Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Chris,

Thanks for pointing that out, I hadn't realized that the SubmittedRecords
class has almost exactly the same semantics needed for handling offset
commits in the per-sink record ack API case. However, I agree that it isn't
worth the tradeoff and we've already discussed the backward compatibility
concerns imposed on connector developers if we were to consider deprecating
/ removing the preCommit hook in favor of a new ack-based API.

Thanks,
Yash

On Thu, Jun 29, 2023 at 7:31 PM Chris Egerton 
wrote:

> Hi Yash,
>
> Thanks for your continued work on this tricky feature. I have no further
> comments or suggestions on the KIP and am ready to vote in favor of it.
>
> That said, I did want to quickly respond to this comment:
>
> > On a side note, this also means that the per sink record ack API
> that was proposed earlier wouldn't really work for this case since Kafka
> consumers themselves don't support per message acknowledgement semantics
> (and any sort of manual book-keeping based on offset linearity in a topic
> partition would be affected by things like log compaction, control records
> for transactional use cases etc.) right?
>
> I believe we could still use the SubmittedRecords class [1] (with some
> small tweaks) to track ack'd messages and the latest-committable offsets
> per topic partition, without relying on assumptions about offsets for
> consecutive records consumed from Kafka always differing by one. But at
> this point I think that, although this approach does come with the
> advantage of also enabling fine-grained metrics on record delivery to the
> sink system, it's not worth the tradeoff in intuition since it's less clear
> why users should prefer that API instead of using SinkTask::preCommit.
>
> [1] -
>
> https://github.com/apache/kafka/blob/12be344fdd3b20f338ccab87933b89049ce202a4/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/SubmittedRecords.java
>
> Cheers,
>
> Chris
>
> On Wed, Jun 21, 2023 at 9:46 AM Yash Mayya  wrote:
>
> > Hi Chris,
> >
> > Firstly, thanks for sharing your detailed thoughts on this thorny issue!
> > Point taken on Kafka Connect being a brownfield project and I guess we
> > might just need to trade off elegant / "clean" interfaces for fixing this
> > gap in functionality. Also, thanks for calling out all the existing
> > cross-plugin interactions and also the fact that connectors are not and
> > should not be developed in silos ignoring the rest of the ecosystem. That
> > said, here are my thoughts:
> >
> > > we could replace these methods with headers that the
> > > Connect runtime automatically injects into records directly
> > > before dispatching them to SinkTask::put.
> >
> > Hm, that's an interesting idea to get around the need for connectors to
> > handle potential 'NoSuchMethodError's in calls to
> > SinkRecord::originalTopic/originalKafkaPartition/originalKafkaOffset.
> > However, I'm inclined to agree that retrieving these values from the
> record
> > headers seems even less intuitive and I'm okay with adding this to the
> > rejected alternatives list.
> >
> > > we can consider eliminating the overridden
> > > SinkTask::open/close methods
> >
> > I tried to further explore the idea of keeping just the existing
> > SinkTask::open / SinkTask::close methods but only calling them with
> > post-transform topic partitions and ended up coming to the same
> conclusion
> > that you did earlier in this thread :)
> >
> > The overloaded SinkTask::open / SinkTask::close are currently the biggest
> > sticking points with the latest iteration of this KIP and I'd prefer this
> > elimination for now. The primary reasoning is that the information from
> > open / close on pre-transform topic partitions can be combined with the
> per
> > record information of both pre-transform and post-transform topic
> > partitions to handle most practical use cases without significantly
> > muddying the sink connector related public interfaces. The argument that
> > this makes it harder for sink connectors to deal with post-transform
> topic
> > partitions (i.e. in terms of grouping together or batching records for
> > writing to the sink system) can be countered with the fact that it'll be
> > similarly challenging even with the overloaded method approach of calling
> > open / close with both pre-transform and post-transform topic partitions
> > since the batching would be done on post-transform topic partitions
> whereas
> > offset tracking and reporting for commits would be done on pre-transform
> > topic partitions (and the two won't necessarily serially advance in
> > lockstep). On a side note, this also means that the per sink record ack
> API
> > that was proposed earlier wouldn't really work for this case since Kafka
> > consumers themselves don't support per message acknowledgement semantics
> > (and any sort of manual book-keeping based on offset linearity in a topic
> > partition would be affected by things like log compaction, control
> records
> > 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Chris Egerton]

Hi Yash,

Thanks for your continued work on this tricky feature. I have no further
comments or suggestions on the KIP and am ready to vote in favor of it.

That said, I did want to quickly respond to this comment:

> On a side note, this also means that the per sink record ack API
that was proposed earlier wouldn't really work for this case since Kafka
consumers themselves don't support per message acknowledgement semantics
(and any sort of manual book-keeping based on offset linearity in a topic
partition would be affected by things like log compaction, control records
for transactional use cases etc.) right?

I believe we could still use the SubmittedRecords class [1] (with some
small tweaks) to track ack'd messages and the latest-committable offsets
per topic partition, without relying on assumptions about offsets for
consecutive records consumed from Kafka always differing by one. But at
this point I think that, although this approach does come with the
advantage of also enabling fine-grained metrics on record delivery to the
sink system, it's not worth the tradeoff in intuition since it's less clear
why users should prefer that API instead of using SinkTask::preCommit.

[1] -
https://github.com/apache/kafka/blob/12be344fdd3b20f338ccab87933b89049ce202a4/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/SubmittedRecords.java

Cheers,

Chris

On Wed, Jun 21, 2023 at 9:46 AM Yash Mayya  wrote:

> Hi Chris,
>
> Firstly, thanks for sharing your detailed thoughts on this thorny issue!
> Point taken on Kafka Connect being a brownfield project and I guess we
> might just need to trade off elegant / "clean" interfaces for fixing this
> gap in functionality. Also, thanks for calling out all the existing
> cross-plugin interactions and also the fact that connectors are not and
> should not be developed in silos ignoring the rest of the ecosystem. That
> said, here are my thoughts:
>
> > we could replace these methods with headers that the
> > Connect runtime automatically injects into records directly
> > before dispatching them to SinkTask::put.
>
> Hm, that's an interesting idea to get around the need for connectors to
> handle potential 'NoSuchMethodError's in calls to
> SinkRecord::originalTopic/originalKafkaPartition/originalKafkaOffset.
> However, I'm inclined to agree that retrieving these values from the record
> headers seems even less intuitive and I'm okay with adding this to the
> rejected alternatives list.
>
> > we can consider eliminating the overridden
> > SinkTask::open/close methods
>
> I tried to further explore the idea of keeping just the existing
> SinkTask::open / SinkTask::close methods but only calling them with
> post-transform topic partitions and ended up coming to the same conclusion
> that you did earlier in this thread :)
>
> The overloaded SinkTask::open / SinkTask::close are currently the biggest
> sticking points with the latest iteration of this KIP and I'd prefer this
> elimination for now. The primary reasoning is that the information from
> open / close on pre-transform topic partitions can be combined with the per
> record information of both pre-transform and post-transform topic
> partitions to handle most practical use cases without significantly
> muddying the sink connector related public interfaces. The argument that
> this makes it harder for sink connectors to deal with post-transform topic
> partitions (i.e. in terms of grouping together or batching records for
> writing to the sink system) can be countered with the fact that it'll be
> similarly challenging even with the overloaded method approach of calling
> open / close with both pre-transform and post-transform topic partitions
> since the batching would be done on post-transform topic partitions whereas
> offset tracking and reporting for commits would be done on pre-transform
> topic partitions (and the two won't necessarily serially advance in
> lockstep). On a side note, this also means that the per sink record ack API
> that was proposed earlier wouldn't really work for this case since Kafka
> consumers themselves don't support per message acknowledgement semantics
> (and any sort of manual book-keeping based on offset linearity in a topic
> partition would be affected by things like log compaction, control records
> for transactional use cases etc.) right? Overall, I think that the only
> benefit of the overloaded open / close methods approach is that the
> framework can enable the eventual closure of any post-transform topic
> partition based writers created by sink tasks using the heuristics we
> discussed earlier (via a cache with a time-based eviction policy) which
> doesn't seem worth it at this point.
>
> Thanks,
> Yash
>
> On Mon, May 22, 2023 at 7:30 PM Chris Egerton 
> wrote:
>
> > Hi Yash,
> >
> > I've been following the discussion and have some thoughts. Ultimately I'm
> > still in favor of this KIP and would hate to see it go dormant, though we
> > may end up settling for a 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Chris,

Firstly, thanks for sharing your detailed thoughts on this thorny issue!
Point taken on Kafka Connect being a brownfield project and I guess we
might just need to trade off elegant / "clean" interfaces for fixing this
gap in functionality. Also, thanks for calling out all the existing
cross-plugin interactions and also the fact that connectors are not and
should not be developed in silos ignoring the rest of the ecosystem. That
said, here are my thoughts:

> we could replace these methods with headers that the
> Connect runtime automatically injects into records directly
> before dispatching them to SinkTask::put.

Hm, that's an interesting idea to get around the need for connectors to
handle potential 'NoSuchMethodError's in calls to
SinkRecord::originalTopic/originalKafkaPartition/originalKafkaOffset.
However, I'm inclined to agree that retrieving these values from the record
headers seems even less intuitive and I'm okay with adding this to the
rejected alternatives list.

> we can consider eliminating the overridden
> SinkTask::open/close methods

I tried to further explore the idea of keeping just the existing
SinkTask::open / SinkTask::close methods but only calling them with
post-transform topic partitions and ended up coming to the same conclusion
that you did earlier in this thread :)

The overloaded SinkTask::open / SinkTask::close are currently the biggest
sticking points with the latest iteration of this KIP and I'd prefer this
elimination for now. The primary reasoning is that the information from
open / close on pre-transform topic partitions can be combined with the per
record information of both pre-transform and post-transform topic
partitions to handle most practical use cases without significantly
muddying the sink connector related public interfaces. The argument that
this makes it harder for sink connectors to deal with post-transform topic
partitions (i.e. in terms of grouping together or batching records for
writing to the sink system) can be countered with the fact that it'll be
similarly challenging even with the overloaded method approach of calling
open / close with both pre-transform and post-transform topic partitions
since the batching would be done on post-transform topic partitions whereas
offset tracking and reporting for commits would be done on pre-transform
topic partitions (and the two won't necessarily serially advance in
lockstep). On a side note, this also means that the per sink record ack API
that was proposed earlier wouldn't really work for this case since Kafka
consumers themselves don't support per message acknowledgement semantics
(and any sort of manual book-keeping based on offset linearity in a topic
partition would be affected by things like log compaction, control records
for transactional use cases etc.) right? Overall, I think that the only
benefit of the overloaded open / close methods approach is that the
framework can enable the eventual closure of any post-transform topic
partition based writers created by sink tasks using the heuristics we
discussed earlier (via a cache with a time-based eviction policy) which
doesn't seem worth it at this point.

Thanks,
Yash

On Mon, May 22, 2023 at 7:30 PM Chris Egerton 
wrote:

> Hi Yash,
>
> I've been following the discussion and have some thoughts. Ultimately I'm
> still in favor of this KIP and would hate to see it go dormant, though we
> may end up settling for a less-invasive option.
>
>
> On the topic of abstraction and inter-plugin interactions:
>
> First, there already are instances of cross-plugin interactions. Logical
> type handling is probably the biggest example: a source connector embeds
> metadata in the schema for record keys/values it emits that notifies
> downstream converters about how to handle them. We provide support for some
> logical types in Connect out of the box, but there's nothing stopping
> connector and converter developers from implementing their own logical type
> support using the exact same mechanism and different logical type names,
> which is already done by Debezium, to name one example.
>
> Second, although it's been a goal of Connect to abstract away parts of
> building a data pipeline so that, e.g., connector developers don't have to
> be concerned with converters or consumers, in reality, this layer of
> abstraction has already been eroded. The example that most-readily comes to
> mind is how source tasks are notified of the offsets of records that
> they've emitted after they've been published to Kafka via
> SourceTask::commitRecord [1].
>
> But, more importantly, it's unlikely that connectors are being developed in
> complete isolation. Nobody's going to implement the SinkConnector /
> SinkTask interfaces and then throw that code off to someone else to figure
> out all the details of deployment, configuration, testing, etc. Developers
> will probably have to be aware of at least the converter interface, some of
> the available implementations of it, and 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Chris Egerton]

Hi Yash,

I've been following the discussion and have some thoughts. Ultimately I'm
still in favor of this KIP and would hate to see it go dormant, though we
may end up settling for a less-invasive option.


On the topic of abstraction and inter-plugin interactions:

First, there already are instances of cross-plugin interactions. Logical
type handling is probably the biggest example: a source connector embeds
metadata in the schema for record keys/values it emits that notifies
downstream converters about how to handle them. We provide support for some
logical types in Connect out of the box, but there's nothing stopping
connector and converter developers from implementing their own logical type
support using the exact same mechanism and different logical type names,
which is already done by Debezium, to name one example.

Second, although it's been a goal of Connect to abstract away parts of
building a data pipeline so that, e.g., connector developers don't have to
be concerned with converters or consumers, in reality, this layer of
abstraction has already been eroded. The example that most-readily comes to
mind is how source tasks are notified of the offsets of records that
they've emitted after they've been published to Kafka via
SourceTask::commitRecord [1].

But, more importantly, it's unlikely that connectors are being developed in
complete isolation. Nobody's going to implement the SinkConnector /
SinkTask interfaces and then throw that code off to someone else to figure
out all the details of deployment, configuration, testing, etc. Developers
will probably have to be aware of at least the converter interface, some of
the available implementations of it, and some details of Kafka clients
(e.g., consumer groups for sink connectors). And this isn't a bad
thing--it's unlikely that someone will write a Kafka connector without
having or benefitting from some understanding of Kafka and the steps of the
data pipeline that it will be a part of.

Bringing this to the practical topic of discussion--transformations--I
think it's actually in everyone's best interests for connector developers
to be aware of transformations. This isn't just because of the specific
problem that the KIP is trying to address. It's because there's plenty of
logic that can be implemented via SMT that a naive connector developer will
think that they have to implement on their own, which will ultimately lead
to a sub-par experience for people who end up using those connectors due to
inconsistent semantics (especially lack of predicates), inconsistent
configuration syntax, increased chances for bugs, and FUD ("why wasn't this
implemented as an SMT?").

Finally, although preserving clean, composable interfaces that can be
understood in isolation is a great principle to start with, we are now in
what Anna McDonald recently referred to as "brownfield" space for Connect.
We can't go back in time and redesign the SMT interface/contracts to make
things cleaner. And I don't think it's fair to anyone to suddenly drop
support for SMTs that mutate t/p/o information for sink records, especially
since these can be used gainfully with plenty of existing sink connectors.

Ultimately I still think the path forward that's best for the users is to
make the impossible possible by addressing this long-standing API gap in
Connect. Yes, it adds to the cognitive burden for connector developers, but
if they can tolerate it, the end result is better for everyone involved,
and if they can't, it's likely that the end result will be a preservation
of existing behavior, which leaves us no worse than before.


With all that said, I've thought about how to minimize or at least hide the
API changes as much as possible. I've had two thoughts:

1. On the
SinkRecord::originalTopic/originalKafkaPartition/originalKafkaOffset front,
we could replace these methods with headers that the Connect runtime
automatically injects into records directly before dispatching them to
SinkTask::put. The names can be the proposed method names (e.g.,
"originalTopic"). I believe this is inferior to the current proposal and
should be a rejected alternative, but it at least seemed worth floating in
the name of compromise. I dislike this approach for two reasons: first, it
seems even less intuitive, and second, it doesn't come with the benefit of
encouraging connector developers to understand the SMT interface and take
it into account when designing connectors.

2. Although I'd hate to see the same bookkeeping logic implemented in
multiple connectors, we can consider eliminating the overridden
SinkTask::open/close methods. A note should be added to both methods
clarifying that they are only invoked with the original, pre-transform
topic partitions, and developers will be on their own if they want to deal
with post-transform topic partitions instead. I'm on the fence with this
one, but if it's a choice between passing this KIP without modifying
SinkTask::open/close, or letting the KIP go 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Greg,

Thanks for the response and sorry for the late reply.

> Currently the AK tests have a lot of calls to, for example, new
> SinkRecord(String topic, int partition, Schema keySchema,
> Object key, Schema valueSchema, Object value, long kafkaOffset)
> , a constructor without the original T/P/O values. I assumed that for
> backwards compatibility these constructors would still be usable in
> new runtimes. I imagine that there are also tests in downstream projects
> which make use of these constructors, whenever a Transform, Predicate,
> or Task is tested without a corresponding Converter. My question was
> about what values are chosen for the original T/P/O methods when these
> constructors are used after an upgrade to the latest connect-api.

That's a good question - since this should only primarily affect testing I
think it should be acceptable to simply use the topic, partition and
kafkaOffset values as the originalTopic, originalKafkaPartition
and originalKafkaOffset?

> If you inject the original T/P/O only before and after the chain, SMTs
> after an SMT which changes the original T/P/O will see whatever the
earlier
> SMT emitted. Is this intentional, or should this be avoided?

Hmm, this sounds like a misbehaving / badly implemented SMTs since there
doesn't seem to be any reasonable situation where an SMT should modify a
sink record's original topic / partition / offset data so I'm not in favor
of introducing checks and guards in the framework for this.

Another point that I've been pondering about is the one you raised about
the composability of Connect's plugin ecosystem and the special case
handling we're adding to sink connector plugins to work with certain
transformation plugin types. This really doesn't seem like a good precedent
to be setting / starting (since there don't seem to be any other such
"snowflake" inter-plugin interactions) in my opinion. The alternative of
completely managing this in the framework (and only exposing the virtual
coordinates to the sink tasks) doesn't seem too appealing either due to the
backward compatibility concerns while maintaining existing support and
functionality such as the possibility of implementing exactly-once
semantics, ability for tasks to rewind consumer offsets arbitrarily (which
might require the introduction of some form of persistence for the physical
<-> virtual coordinate mapping) etc. Unfortunately, even though this is a
long standing problem that all of us want to fix, I'm considering moving
this KIP into a dormant / inactive state since there doesn't seem to be a
design that satisfies all the general principles that the Kafka Connect
framework has striven to uphold.

Thanks,
Yash

On Tue, Mar 14, 2023 at 3:31 AM Greg Harris 
wrote:

> Yash,
>
> > 'm not sure I follow - are you asking about how the tests will be updated
> post this change or about how upgrades will look like for clusters in
> production?
>
> Currently the AK tests have a lot of calls to, for example, new
> SinkRecord(String topic, int partition, Schema keySchema, Object key,
> Schema valueSchema, Object value, long kafkaOffset), a constructor without
> the original T/P/O values. I assumed that for backwards compatibility these
> constructors would still be usable in new runtimes.
> I imagine that there are also tests in downstream projects which make use
> of these constructors, whenever a Transform, Predicate, or Task is tested
> without a corresponding Converter. My question was about what values are
> chosen for the original T/P/O methods when these constructors are used
> after an upgrade to the latest connect-api.
>
> > There shouldn't be any difference in behavior here - the framework will
> add
> the original T/P/O metadata to the record after the entire transformation
> chain has been applied and just before sending the record to the task for
> processing. The KIP doesn't propose that transformations themselves should
> also be able to retrieve original T/P/O information for a sink record.
>
> The KIP includes this: "Note that while the record's offset can't be
> modified via the standard SinkRecord::newRecord methods that SMTs are
> expected to use, SinkRecord has public constructors that would allow SMTs
> to return records with modified offsets. This is why the proposed changes
> include a new SinkRecord::originalKafkaOffset method as well."
> In order to use the new or old SinkRecord constructors outside of the
> newRecord methods, SMTs will downcast the previous record and may access
> the original T/P/O methods. They may or may not forward this to the next
> SMT, and they may or may not use it in their own computation.
> Since this is acknowledged as a possible implementation, I was just asking
> about when one SMT changes the original T/P/O, what should later SMTs and
> predicates see from the original T/P/O methods?
> If you inject the original T/P/O only before and after the chain, SMTs
> after an SMT which changes the original T/P/O will see whatever 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Greg Harris]

Yash,

> 'm not sure I follow - are you asking about how the tests will be updated
post this change or about how upgrades will look like for clusters in
production?

Currently the AK tests have a lot of calls to, for example, new
SinkRecord(String topic, int partition, Schema keySchema, Object key,
Schema valueSchema, Object value, long kafkaOffset), a constructor without
the original T/P/O values. I assumed that for backwards compatibility these
constructors would still be usable in new runtimes.
I imagine that there are also tests in downstream projects which make use
of these constructors, whenever a Transform, Predicate, or Task is tested
without a corresponding Converter. My question was about what values are
chosen for the original T/P/O methods when these constructors are used
after an upgrade to the latest connect-api.

> There shouldn't be any difference in behavior here - the framework will
add
the original T/P/O metadata to the record after the entire transformation
chain has been applied and just before sending the record to the task for
processing. The KIP doesn't propose that transformations themselves should
also be able to retrieve original T/P/O information for a sink record.

The KIP includes this: "Note that while the record's offset can't be
modified via the standard SinkRecord::newRecord methods that SMTs are
expected to use, SinkRecord has public constructors that would allow SMTs
to return records with modified offsets. This is why the proposed changes
include a new SinkRecord::originalKafkaOffset method as well."
In order to use the new or old SinkRecord constructors outside of the
newRecord methods, SMTs will downcast the previous record and may access
the original T/P/O methods. They may or may not forward this to the next
SMT, and they may or may not use it in their own computation.
Since this is acknowledged as a possible implementation, I was just asking
about when one SMT changes the original T/P/O, what should later SMTs and
predicates see from the original T/P/O methods?
If you inject the original T/P/O only before and after the chain, SMTs
after an SMT which changes the original T/P/O will see whatever the earlier
SMT emitted. Is this intentional, or should this be avoided?
For existing SMTs use the SinkRecord constructor, either directly or via
subclasses of ConnectRecord, they will drop the original T/P/O and fall
back to the logic from question (1).

> The rejected alternative basically says that we can't do a
deterministic mapping from virtual coordinates to physical coordinates
without doing a lot of book-keeping.

I suppose there is a possible implementation of metadata book-keeping which
provides a reasonable system of virtual coordinates, it just ended up
equivalent to hydrating intermediate topics to compute a consistent record
ordering. I wasn't convinced by calling it "book-keeping" since i've seen
that phrase used to disregard much less complicated state management, and
had to see exactly where that solution becomes unreasonable.

Thanks,
Greg

On Sun, Mar 12, 2023 at 6:30 AM Yash Mayya  wrote:

> Hi Greg,
>
> Thanks for the detailed review!
>
> > What is the expected state/behavior for SinkRecords
> > which do not have original T/P/O information after the
> > upgrade? Just browsing, it appears that tests make
> > extensive use of the existing public SinkRecord
> > constructors  for both Transformations and Connectors.
>
> I'm not sure I follow - are you asking about how the tests will be updated
> post this change or about how upgrades will look like for clusters in
> production? For the latter, we won't have to worry about sink records
> without original T/P/O information at all once a cluster is fully rolled
> and we will make it (hopefully) abundantly clear that connectors need to
> account for missing original T/P/O getter methods if they expect to be
> deployed on older Connect runtimes.
>
> > What is the expected behavior for Transformation
> > implementations which do not use the newRecord
> > methods and instead use public SinkRecord constructors?
> > The KIP mentions this as a justification for the
> > originalKafkaOffset method, but if existing implementations
> > are using the existing constructors, those constructors won't
> > forward the original T/P/O information to later transforms or
> > the task.
>
> There shouldn't be any difference in behavior here - the framework will add
> the original T/P/O metadata to the record after the entire transformation
> chain has been applied and just before sending the record to the task for
> processing. The KIP doesn't propose that transformations themselves should
> also be able to retrieve original T/P/O information for a sink record.
>
> > This reasoning and the KIP design seems to imply that the
> > connector is better equipped to solve this problem than the
> > framework, but the stated reasons are not convincing for me.
>
> This was added to the KIP by the original author, but I don't think the
> 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Greg,

Thanks for the detailed review!

> What is the expected state/behavior for SinkRecords
> which do not have original T/P/O information after the
> upgrade? Just browsing, it appears that tests make
> extensive use of the existing public SinkRecord
> constructors  for both Transformations and Connectors.

I'm not sure I follow - are you asking about how the tests will be updated
post this change or about how upgrades will look like for clusters in
production? For the latter, we won't have to worry about sink records
without original T/P/O information at all once a cluster is fully rolled
and we will make it (hopefully) abundantly clear that connectors need to
account for missing original T/P/O getter methods if they expect to be
deployed on older Connect runtimes.

> What is the expected behavior for Transformation
> implementations which do not use the newRecord
> methods and instead use public SinkRecord constructors?
> The KIP mentions this as a justification for the
> originalKafkaOffset method, but if existing implementations
> are using the existing constructors, those constructors won't
> forward the original T/P/O information to later transforms or
> the task.

There shouldn't be any difference in behavior here - the framework will add
the original T/P/O metadata to the record after the entire transformation
chain has been applied and just before sending the record to the task for
processing. The KIP doesn't propose that transformations themselves should
also be able to retrieve original T/P/O information for a sink record.

> This reasoning and the KIP design seems to imply that the
> connector is better equipped to solve this problem than the
> framework, but the stated reasons are not convincing for me.

This was added to the KIP by the original author, but I don't think the
intention was to imply that the connector is better equipped to solve this
problem than the framework. The intention is to provide complete
information to the connector ("physical" and "virtual coordinates" instead
of the currently incomplete "virtual coordinates" as you've termed it) so
that connectors can use the virtual coordinates for writing data to the
sink system and physical coordinates for offset reporting back to the
framework. The rejected alternative basically says that we can't do a
deterministic mapping from virtual coordinates to physical coordinates
without doing a lot of book-keeping.

I agree with the rest of your analysis on the tradeoffs between the
proposed approach versus the seemingly more attractive approach of handling
everything purely in the framework and only exposing "virtual coordinates"
to the connectors. I think the biggest thorn here is maintaining backward
compatibility with the considerable ecosystem of existing connectors which
is something Connect has always been burdened by.

Thanks,
Yash

On Wed, Mar 8, 2023 at 6:54 AM Greg Harris 
wrote:

> Hi Yash,
>
> I always use this issue as an example of a bug being caused by design
> rather than by implementation error, and once it's fixed I'll need to find
> something else to talk about :)
> So glad to see this get fixed!
>
> I'll chime in to support some of the earlier discussions that seem to have
> been resolved:
>
> 1. With respect to SinkRecord methods vs an overloaded put(): I agree with
> the current design but I justify it a little bit differently than has
> already been discussed.
> If we were designing this interface on day 1 without backwards
> compatibility in mind, which design would make more sense? Or for a
> different framing: In the future when old runtimes and connectors are
> retired and the old interfaces are removed, which design is going to look
> more strange and unmotivated?
> Applied to this design decision, I would say that the original T/P/O are
> properties of a single SinkRecord and make sense as getters, and it would
> be strange to store them in an auxiliary map.
>
> 2. Following up this change with a compatibility library to make the
> interface easier to use is the right choice to make here. This change
> should be focused on correctness in allowing developers to fix the
> incompatibility and we can be concerned with coming up with a more
> ergonomic solution in the compatibility library.
> The API should be focused on generality, correctness, and performance
> because those cannot be worked-around after the fact. Connector
> implementations and/or libraries can be concerned with trading off some
> generality and/or performance for ease-of-use.
>
> 3. I think that the difference in behavior of the new open/close methods as
> compared to the old methods is significant, and requires good documentation
> to help connector developers avoid lazy and incorrect migrations. I am
> happy to have that addressed in code review after the KIP is approved.
>
> I had some questions:
>
> 4. What is the expected state/behavior for SinkRecords which do not have
> original T/P/O information after the upgrade? Just 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Greg Harris]

Hi Yash,

I always use this issue as an example of a bug being caused by design
rather than by implementation error, and once it's fixed I'll need to find
something else to talk about :)
So glad to see this get fixed!

I'll chime in to support some of the earlier discussions that seem to have
been resolved:

1. With respect to SinkRecord methods vs an overloaded put(): I agree with
the current design but I justify it a little bit differently than has
already been discussed.
If we were designing this interface on day 1 without backwards
compatibility in mind, which design would make more sense? Or for a
different framing: In the future when old runtimes and connectors are
retired and the old interfaces are removed, which design is going to look
more strange and unmotivated?
Applied to this design decision, I would say that the original T/P/O are
properties of a single SinkRecord and make sense as getters, and it would
be strange to store them in an auxiliary map.

2. Following up this change with a compatibility library to make the
interface easier to use is the right choice to make here. This change
should be focused on correctness in allowing developers to fix the
incompatibility and we can be concerned with coming up with a more
ergonomic solution in the compatibility library.
The API should be focused on generality, correctness, and performance
because those cannot be worked-around after the fact. Connector
implementations and/or libraries can be concerned with trading off some
generality and/or performance for ease-of-use.

3. I think that the difference in behavior of the new open/close methods as
compared to the old methods is significant, and requires good documentation
to help connector developers avoid lazy and incorrect migrations. I am
happy to have that addressed in code review after the KIP is approved.

I had some questions:

4. What is the expected state/behavior for SinkRecords which do not have
original T/P/O information after the upgrade? Just browsing, it appears
that tests make extensive use of the existing public SinkRecord
constructors for both Transformations and Connectors.

5. What is the expected behavior for Transformation implementations which
do not use the newRecord methods and instead use public SinkRecord
constructors? The KIP mentions this as a justification for the
originalKafkaOffset method, but if existing implementations are using the
existing constructors, those constructors won't forward the original T/P/O
information to later transforms or the task.

For the last few points, I want to discuss this rejected alternative:

> Address the offsets problem entirely within the framework, doing some
kind of mapping from the transformed topic back to the original topic.
> * This would only work in the cases where there’s no overlap between the
transformed topic names, but would break for the rest of the
transformations (e.g. static transformation, topic = “a”).
> * Even if we wanted to limit the support to those cases, it would require
considerable bookkeeping to add a validation to verify that the
transformation chain adheres to that expectation (and fail fast if it
doesn’t).

6. This reasoning and the KIP design seems to imply that the connector is
better equipped to solve this problem than the framework, but the stated
reasons are not convincing for me.
* A static transformation still causes an offset collision in the connector
* The connector is not permitted to see the transformation chain to do any
fail-fast assertions

Suppose we were to think of the records at the end of the transformation
chain as being in "virtual partitions" with "virtual offsets".
For example, with identity-routing SMTs, the virtual coordinates are
exactly the same as the underlying physical coordinates. For 1-1 renames,
each virtual topic would be the renamed topic corresponding to the
underlying topic. For fan-out from one topic to multiple virtual topics,
virtual offsets would use the underlying kafka offsets with gaps for
records going to other virtual partitions. Virtual topics with dropped
records have similar gaps in the offsets.
Currently, these virtual coordinates are passed into the connector via
SinkTask::put, but SinkTask::open/close/preCommit and
SinkTaskContext::assignment/offsets/pause/resume all use physical
coordinates.
This proposal patches put,open, and close to have both physical and virtual
coordinates, but leaves the other methods with physical coordinates. After
this proposal, connectors would be intentionally made aware of the
distinction between physical and virtual coordinates, and manage their own
bookkeeping for the two systems.

To avoid that connector logic, we could use virtual coordinates in all
connector calls, never revealing that they are different from the physical
coordinates. There's a whole design shopping list that we'd need:
* Renumbering mechanism for disambiguating and making virtual offsets
monotonic in the case of topic/partition collisions
* Data 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Chris Egerton]

Hi Yash,

We'll probably want to make a few tweaks to the Javadocs for the new
methods (I'm imagining that notes on compatibility with older versions will
be required), but I believe what's proposed in the KIP is good enough to
approve with the understanding that it may not exactly match what gets
implemented/merged.

LGTM, thanks again for the KIP!

Cheers,

Chris

On Tue, Feb 21, 2023 at 12:18 PM Yash Mayya  wrote:

> Hi Chris,
>
> > we might try to introduce a framework-level configuration
> > property to dictate which of the pre-transform and post-transform
> > topic partitions are used for the fallback call to the single-arg
> > variant if a task class has not overridden the multi-arg variant
>
> Thanks for the explanation and I agree that this will be a tad bit too
> convoluted. :)
>
> Please do let me know if you'd like any further amendments to the KIP!
>
> Thanks,
> Yash
>
> On Tue, Feb 21, 2023 at 8:42 PM Chris Egerton 
> wrote:
>
> > Hi Yash,
> >
> > I think the use case for pre-transform TPO coordinates (and topic
> partition
> > writers created/destroyed in close/open) tends to boil down to
> exactly-once
> > semantics, where it's desirable to preserve the guarantees that Kafka
> > provides (every record has a unique TPO trio, and records are ordered by
> > offset within a topic partition).
> >
> > It's my understanding that this approach is utilized in several
> connectors
> > out there today, and it might break these connectors to start using the
> > post-transform topic partitions automatically in their open/close
> methods.
> >
> > If we want to get really fancy with this and try to obviate or at least
> > reduce the need for per-connector code changes, we might try to
> introduce a
> > framework-level configuration property to dictate which of the
> > pre-transform and post-transform topic partitions are used for the
> fallback
> > call to the single-arg variant if a task class has not overridden the
> > multi-arg variant. But I think this is going a bit too far and would
> prefer
> > to keep things simple(r) for now.
> >
> > Cheers,
> >
> > Chris
> >
> >
> > On Sun, Feb 19, 2023 at 2:34 AM Yash Mayya  wrote:
> >
> > > Hi Chris,
> > >
> > > > I was actually envisioning something like `void
> > > > open(Collection originalPartitions,
> > > > Collection transformedPartitions)`
> > >
> > > Ah okay, this does make a lot more sense. Sorry, I think I
> misunderstood
> > > you earlier. I do agree with you that this seems better than splitting
> it
> > > off into two new sets of open / close methods from a complexity
> > standpoint.
> > >
> > > > Plus, if a connector is intentionally designed to use
> > > > pre-transformation topic partitions in its open/close
> > > > methods, wouldn't we just be trading one form of the
> > > >  problem for another by making this switch?
> > >
> > > On thinking about this a bit more, I'm not so convinced that we need to
> > > expose the pre-transform / original topic partitions in the new open /
> > > close methods. The purpose of the open / close methods is to allow sink
> > > tasks to allocate and deallocate resources for each topic partition
> > > assigned to the task and the purpose of topic-mutating SMTs is to
> > > essentially modify the source topic name from the point of view of the
> > sink
> > > connector. Why would a sink connector ever need to or want to allocate
> > > resources for pre-transform topic partitions? Is the argument here that
> > > since we'll be exposing both the pre-transform and post-transform topic
> > > partitions per record, we should also expose the same info via open /
> > close
> > > and allow sink connector implementations to disregard topic-mutating
> SMTs
> > > completely if they wanted to?
> > >
> > > Either way, I've gone ahead and updated the KIP to reflect all of
> > > our previous discussion here since it had become quite outdated. I've
> > also
> > > updated the KIP title from "Sink Connectors: Support topic-mutating
> SMTs
> > > for async connectors (preCommit users)" to "Allow sink connectors to be
> > > used with topic-mutating SMTs" since the improvements to the open /
> close
> > > mechanism doesn't pertain only to asynchronous sink connectors. The new
> > KIP
> > > URL is:
> > >
> > >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/KIP-793%3A+Allow+sink+connectors+to+be+used+with+topic-mutating+SMTs
> > >
> > >
> > > Thanks,
> > > Yash
> > >
> > > On Tue, Feb 14, 2023 at 11:39 PM Chris Egerton  >
> > > wrote:
> > >
> > > > Hi Yash,
> > > >
> > > > I was actually envisioning something like `void
> > > > open(Collection
> > > > originalPartitions, Collection
> transformedPartitions)`,
> > > > since we already convert and transform each batch of records that we
> > poll
> > > > from the sink task's consumer en masse, meaning we could discover
> > several
> > > > new transformed partitions in between consecutive calls to
> > SinkTask::put.
> > > >
> > > > It's also worth noting that we'll probably want to 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Chris,

> we might try to introduce a framework-level configuration
> property to dictate which of the pre-transform and post-transform
> topic partitions are used for the fallback call to the single-arg
> variant if a task class has not overridden the multi-arg variant

Thanks for the explanation and I agree that this will be a tad bit too
convoluted. :)

Please do let me know if you'd like any further amendments to the KIP!

Thanks,
Yash

On Tue, Feb 21, 2023 at 8:42 PM Chris Egerton 
wrote:

> Hi Yash,
>
> I think the use case for pre-transform TPO coordinates (and topic partition
> writers created/destroyed in close/open) tends to boil down to exactly-once
> semantics, where it's desirable to preserve the guarantees that Kafka
> provides (every record has a unique TPO trio, and records are ordered by
> offset within a topic partition).
>
> It's my understanding that this approach is utilized in several connectors
> out there today, and it might break these connectors to start using the
> post-transform topic partitions automatically in their open/close methods.
>
> If we want to get really fancy with this and try to obviate or at least
> reduce the need for per-connector code changes, we might try to introduce a
> framework-level configuration property to dictate which of the
> pre-transform and post-transform topic partitions are used for the fallback
> call to the single-arg variant if a task class has not overridden the
> multi-arg variant. But I think this is going a bit too far and would prefer
> to keep things simple(r) for now.
>
> Cheers,
>
> Chris
>
>
> On Sun, Feb 19, 2023 at 2:34 AM Yash Mayya  wrote:
>
> > Hi Chris,
> >
> > > I was actually envisioning something like `void
> > > open(Collection originalPartitions,
> > > Collection transformedPartitions)`
> >
> > Ah okay, this does make a lot more sense. Sorry, I think I misunderstood
> > you earlier. I do agree with you that this seems better than splitting it
> > off into two new sets of open / close methods from a complexity
> standpoint.
> >
> > > Plus, if a connector is intentionally designed to use
> > > pre-transformation topic partitions in its open/close
> > > methods, wouldn't we just be trading one form of the
> > >  problem for another by making this switch?
> >
> > On thinking about this a bit more, I'm not so convinced that we need to
> > expose the pre-transform / original topic partitions in the new open /
> > close methods. The purpose of the open / close methods is to allow sink
> > tasks to allocate and deallocate resources for each topic partition
> > assigned to the task and the purpose of topic-mutating SMTs is to
> > essentially modify the source topic name from the point of view of the
> sink
> > connector. Why would a sink connector ever need to or want to allocate
> > resources for pre-transform topic partitions? Is the argument here that
> > since we'll be exposing both the pre-transform and post-transform topic
> > partitions per record, we should also expose the same info via open /
> close
> > and allow sink connector implementations to disregard topic-mutating SMTs
> > completely if they wanted to?
> >
> > Either way, I've gone ahead and updated the KIP to reflect all of
> > our previous discussion here since it had become quite outdated. I've
> also
> > updated the KIP title from "Sink Connectors: Support topic-mutating SMTs
> > for async connectors (preCommit users)" to "Allow sink connectors to be
> > used with topic-mutating SMTs" since the improvements to the open / close
> > mechanism doesn't pertain only to asynchronous sink connectors. The new
> KIP
> > URL is:
> >
> >
> https://cwiki.apache.org/confluence/display/KAFKA/KIP-793%3A+Allow+sink+connectors+to+be+used+with+topic-mutating+SMTs
> >
> >
> > Thanks,
> > Yash
> >
> > On Tue, Feb 14, 2023 at 11:39 PM Chris Egerton 
> > wrote:
> >
> > > Hi Yash,
> > >
> > > I was actually envisioning something like `void
> > > open(Collection
> > > originalPartitions, Collection transformedPartitions)`,
> > > since we already convert and transform each batch of records that we
> poll
> > > from the sink task's consumer en masse, meaning we could discover
> several
> > > new transformed partitions in between consecutive calls to
> SinkTask::put.
> > >
> > > It's also worth noting that we'll probably want to deprecate the
> existing
> > > open/close methods, at which point keeping one non-deprecated variant
> of
> > > each seems more appealing and less complex than keeping two.
> > >
> > > Honestly though, I think we're both on the same page enough that I
> > wouldn't
> > > object to either approach. We've probably reached the saturation point
> > for
> > > ROI here and as long as we provide developers a way to get the
> > information
> > > they need from the runtime and take care to add Javadocs and update our
> > > docs page (possibly including the connector development quickstart), it
> > > should be fine.
> > >
> > > At this point, it might be worth updating 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Chris Egerton]

Hi Yash,

I think the use case for pre-transform TPO coordinates (and topic partition
writers created/destroyed in close/open) tends to boil down to exactly-once
semantics, where it's desirable to preserve the guarantees that Kafka
provides (every record has a unique TPO trio, and records are ordered by
offset within a topic partition).

It's my understanding that this approach is utilized in several connectors
out there today, and it might break these connectors to start using the
post-transform topic partitions automatically in their open/close methods.

If we want to get really fancy with this and try to obviate or at least
reduce the need for per-connector code changes, we might try to introduce a
framework-level configuration property to dictate which of the
pre-transform and post-transform topic partitions are used for the fallback
call to the single-arg variant if a task class has not overridden the
multi-arg variant. But I think this is going a bit too far and would prefer
to keep things simple(r) for now.

Cheers,

Chris


On Sun, Feb 19, 2023 at 2:34 AM Yash Mayya  wrote:

> Hi Chris,
>
> > I was actually envisioning something like `void
> > open(Collection originalPartitions,
> > Collection transformedPartitions)`
>
> Ah okay, this does make a lot more sense. Sorry, I think I misunderstood
> you earlier. I do agree with you that this seems better than splitting it
> off into two new sets of open / close methods from a complexity standpoint.
>
> > Plus, if a connector is intentionally designed to use
> > pre-transformation topic partitions in its open/close
> > methods, wouldn't we just be trading one form of the
> >  problem for another by making this switch?
>
> On thinking about this a bit more, I'm not so convinced that we need to
> expose the pre-transform / original topic partitions in the new open /
> close methods. The purpose of the open / close methods is to allow sink
> tasks to allocate and deallocate resources for each topic partition
> assigned to the task and the purpose of topic-mutating SMTs is to
> essentially modify the source topic name from the point of view of the sink
> connector. Why would a sink connector ever need to or want to allocate
> resources for pre-transform topic partitions? Is the argument here that
> since we'll be exposing both the pre-transform and post-transform topic
> partitions per record, we should also expose the same info via open / close
> and allow sink connector implementations to disregard topic-mutating SMTs
> completely if they wanted to?
>
> Either way, I've gone ahead and updated the KIP to reflect all of
> our previous discussion here since it had become quite outdated. I've also
> updated the KIP title from "Sink Connectors: Support topic-mutating SMTs
> for async connectors (preCommit users)" to "Allow sink connectors to be
> used with topic-mutating SMTs" since the improvements to the open / close
> mechanism doesn't pertain only to asynchronous sink connectors. The new KIP
> URL is:
>
> https://cwiki.apache.org/confluence/display/KAFKA/KIP-793%3A+Allow+sink+connectors+to+be+used+with+topic-mutating+SMTs
>
>
> Thanks,
> Yash
>
> On Tue, Feb 14, 2023 at 11:39 PM Chris Egerton 
> wrote:
>
> > Hi Yash,
> >
> > I was actually envisioning something like `void
> > open(Collection
> > originalPartitions, Collection transformedPartitions)`,
> > since we already convert and transform each batch of records that we poll
> > from the sink task's consumer en masse, meaning we could discover several
> > new transformed partitions in between consecutive calls to SinkTask::put.
> >
> > It's also worth noting that we'll probably want to deprecate the existing
> > open/close methods, at which point keeping one non-deprecated variant of
> > each seems more appealing and less complex than keeping two.
> >
> > Honestly though, I think we're both on the same page enough that I
> wouldn't
> > object to either approach. We've probably reached the saturation point
> for
> > ROI here and as long as we provide developers a way to get the
> information
> > they need from the runtime and take care to add Javadocs and update our
> > docs page (possibly including the connector development quickstart), it
> > should be fine.
> >
> > At this point, it might be worth updating the KIP based on recent
> > discussion so that others can see the latest proposal, and we can both
> take
> > a look and make sure everything looks good enough before opening a vote
> > thread.
> >
> > Finally, I think you make a convincing case for a time-based eviction
> > policy. I wasn't thinking about the fairly common SMT pattern of
> deriving a
> > topic name from, e.g., a record field or header.
> >
> > Cheers,
> >
> > Chris
> >
> > On Tue, Feb 14, 2023 at 11:42 AM Yash Mayya 
> wrote:
> >
> > > Hi Chris,
> > >
> > > > Plus, if a connector is intentionally designed to
> > > > use pre-transformation topic partitions in its
> > > > open/close methods, wouldn't we just be trading
> > > > one 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Chris,

> I was actually envisioning something like `void
> open(Collection originalPartitions,
> Collection transformedPartitions)`

Ah okay, this does make a lot more sense. Sorry, I think I misunderstood
you earlier. I do agree with you that this seems better than splitting it
off into two new sets of open / close methods from a complexity standpoint.

> Plus, if a connector is intentionally designed to use
> pre-transformation topic partitions in its open/close
> methods, wouldn't we just be trading one form of the
>  problem for another by making this switch?

On thinking about this a bit more, I'm not so convinced that we need to
expose the pre-transform / original topic partitions in the new open /
close methods. The purpose of the open / close methods is to allow sink
tasks to allocate and deallocate resources for each topic partition
assigned to the task and the purpose of topic-mutating SMTs is to
essentially modify the source topic name from the point of view of the sink
connector. Why would a sink connector ever need to or want to allocate
resources for pre-transform topic partitions? Is the argument here that
since we'll be exposing both the pre-transform and post-transform topic
partitions per record, we should also expose the same info via open / close
and allow sink connector implementations to disregard topic-mutating SMTs
completely if they wanted to?

Either way, I've gone ahead and updated the KIP to reflect all of
our previous discussion here since it had become quite outdated. I've also
updated the KIP title from "Sink Connectors: Support topic-mutating SMTs
for async connectors (preCommit users)" to "Allow sink connectors to be
used with topic-mutating SMTs" since the improvements to the open / close
mechanism doesn't pertain only to asynchronous sink connectors. The new KIP
URL is:
https://cwiki.apache.org/confluence/display/KAFKA/KIP-793%3A+Allow+sink+connectors+to+be+used+with+topic-mutating+SMTs


Thanks,
Yash

On Tue, Feb 14, 2023 at 11:39 PM Chris Egerton 
wrote:

> Hi Yash,
>
> I was actually envisioning something like `void
> open(Collection
> originalPartitions, Collection transformedPartitions)`,
> since we already convert and transform each batch of records that we poll
> from the sink task's consumer en masse, meaning we could discover several
> new transformed partitions in between consecutive calls to SinkTask::put.
>
> It's also worth noting that we'll probably want to deprecate the existing
> open/close methods, at which point keeping one non-deprecated variant of
> each seems more appealing and less complex than keeping two.
>
> Honestly though, I think we're both on the same page enough that I wouldn't
> object to either approach. We've probably reached the saturation point for
> ROI here and as long as we provide developers a way to get the information
> they need from the runtime and take care to add Javadocs and update our
> docs page (possibly including the connector development quickstart), it
> should be fine.
>
> At this point, it might be worth updating the KIP based on recent
> discussion so that others can see the latest proposal, and we can both take
> a look and make sure everything looks good enough before opening a vote
> thread.
>
> Finally, I think you make a convincing case for a time-based eviction
> policy. I wasn't thinking about the fairly common SMT pattern of deriving a
> topic name from, e.g., a record field or header.
>
> Cheers,
>
> Chris
>
> On Tue, Feb 14, 2023 at 11:42 AM Yash Mayya  wrote:
>
> > Hi Chris,
> >
> > > Plus, if a connector is intentionally designed to
> > > use pre-transformation topic partitions in its
> > > open/close methods, wouldn't we just be trading
> > > one form of the problem for another by making this
> > > switch?
> >
> > Thanks, this makes sense, and given that the KIP already proposes a way
> for
> > sink connector implementations to distinguish between pre-transform and
> > post-transform topics per record, I think I'm convinced that going with
> new
> > `open()` / `close()` methods is the right approach. However, I still feel
> > like having overloaded methods will make it a lot less unintuitive given
> > that the two sets of methods would be different in terms of when they're
> > called and what arguments they are passed (also I'm presuming that the
> > overloaded methods you're prescribing will only have a single
> > `TopicPartition` rather than a `Collection` as their
> > parameters). I guess my concern is largely around the fact that it won't
> be
> > possible to distinguish between the overloaded methods' use cases just
> from
> > the method signatures. I agree that naming is going to be difficult here,
> > but I think that having two sets of `SinkTask::openXyz` /
> > `SinkTask::closeXyz` methods will be less complicated to understand from
> a
> > connector developer perspective (as compared to overloaded methods with
> > only differing documentation). Of your suggested options, I think
> > 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Chris Egerton]

Hi Yash,

I was actually envisioning something like `void
open(Collection
originalPartitions, Collection transformedPartitions)`,
since we already convert and transform each batch of records that we poll
from the sink task's consumer en masse, meaning we could discover several
new transformed partitions in between consecutive calls to SinkTask::put.

It's also worth noting that we'll probably want to deprecate the existing
open/close methods, at which point keeping one non-deprecated variant of
each seems more appealing and less complex than keeping two.

Honestly though, I think we're both on the same page enough that I wouldn't
object to either approach. We've probably reached the saturation point for
ROI here and as long as we provide developers a way to get the information
they need from the runtime and take care to add Javadocs and update our
docs page (possibly including the connector development quickstart), it
should be fine.

At this point, it might be worth updating the KIP based on recent
discussion so that others can see the latest proposal, and we can both take
a look and make sure everything looks good enough before opening a vote
thread.

Finally, I think you make a convincing case for a time-based eviction
policy. I wasn't thinking about the fairly common SMT pattern of deriving a
topic name from, e.g., a record field or header.

Cheers,

Chris

On Tue, Feb 14, 2023 at 11:42 AM Yash Mayya  wrote:

> Hi Chris,
>
> > Plus, if a connector is intentionally designed to
> > use pre-transformation topic partitions in its
> > open/close methods, wouldn't we just be trading
> > one form of the problem for another by making this
> > switch?
>
> Thanks, this makes sense, and given that the KIP already proposes a way for
> sink connector implementations to distinguish between pre-transform and
> post-transform topics per record, I think I'm convinced that going with new
> `open()` / `close()` methods is the right approach. However, I still feel
> like having overloaded methods will make it a lot less unintuitive given
> that the two sets of methods would be different in terms of when they're
> called and what arguments they are passed (also I'm presuming that the
> overloaded methods you're prescribing will only have a single
> `TopicPartition` rather than a `Collection` as their
> parameters). I guess my concern is largely around the fact that it won't be
> possible to distinguish between the overloaded methods' use cases just from
> the method signatures. I agree that naming is going to be difficult here,
> but I think that having two sets of `SinkTask::openXyz` /
> `SinkTask::closeXyz` methods will be less complicated to understand from a
> connector developer perspective (as compared to overloaded methods with
> only differing documentation). Of your suggested options, I think
> `openPreTransform` / `openPostTransform` are the most comprehensible ones.
>
> > BTW, I wouldn't say that we can't make assumptions
> > about the relationships between pre- and post-transformation
> >  topic partitions.
>
> I meant that the framework wouldn't be able to deterministically know when
> to close a post-transform topic partition given that SMTs could use
> per-record data / metadata to manipulate the topic names as and how
> required (which supports the suggestion to use an eviction policy based
> mechanism to call SinkTask::close for post-transform topic partitions).
>
> > We might utilize a policy that assumes a deterministic
> > mapping from the former to the latter, for example.
>
> Wouldn't this be making the assumption that SMTs only use the topic name
> itself and no other data / metadata while computing the new topic name? Are
> you suggesting that since this assumption could work for a majority of
> SMTs, it might be more efficient overall in terms of reducing the number of
> "false-positive" calls to `SinkTask::closePostTransform` (and we'll also be
> able to call `SinkTask::closePostTransform` immediately after topic
> partitions are revoked from the consumer)? I was thinking something more
> generic along the lines of a simple time based eviction policy that
> wouldn't be making any assumptions regarding the SMT implementations.
> Either way, I do like your earlier suggestion of keeping this logic
> internal and not painting ourselves into a corner by promising any
> particular behavior in the KIP.
>
> Thanks,
> Yash
>
> On Tue, Feb 14, 2023 at 1:08 AM Chris Egerton 
> wrote:
>
> > Hi Yash,
> >
> > I think the key difference between adding methods/overloads related to
> > SinkTask::open/SinkTask::close and SinkTask::put is that this isn't
> > auxiliary information that may or may not be useful to connector
> > developers. It's actually critical for them to understand the difference
> > between the two concepts here, even if they look very similar. And yes, I
> > do believe that switching from pre-transform to post-transform topic
> > partitions is too big a change in behavior here. Plus, if a 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Chris,

> Plus, if a connector is intentionally designed to
> use pre-transformation topic partitions in its
> open/close methods, wouldn't we just be trading
> one form of the problem for another by making this
> switch?

Thanks, this makes sense, and given that the KIP already proposes a way for
sink connector implementations to distinguish between pre-transform and
post-transform topics per record, I think I'm convinced that going with new
`open()` / `close()` methods is the right approach. However, I still feel
like having overloaded methods will make it a lot less unintuitive given
that the two sets of methods would be different in terms of when they're
called and what arguments they are passed (also I'm presuming that the
overloaded methods you're prescribing will only have a single
`TopicPartition` rather than a `Collection` as their
parameters). I guess my concern is largely around the fact that it won't be
possible to distinguish between the overloaded methods' use cases just from
the method signatures. I agree that naming is going to be difficult here,
but I think that having two sets of `SinkTask::openXyz` /
`SinkTask::closeXyz` methods will be less complicated to understand from a
connector developer perspective (as compared to overloaded methods with
only differing documentation). Of your suggested options, I think
`openPreTransform` / `openPostTransform` are the most comprehensible ones.

> BTW, I wouldn't say that we can't make assumptions
> about the relationships between pre- and post-transformation
>  topic partitions.

I meant that the framework wouldn't be able to deterministically know when
to close a post-transform topic partition given that SMTs could use
per-record data / metadata to manipulate the topic names as and how
required (which supports the suggestion to use an eviction policy based
mechanism to call SinkTask::close for post-transform topic partitions).

> We might utilize a policy that assumes a deterministic
> mapping from the former to the latter, for example.

Wouldn't this be making the assumption that SMTs only use the topic name
itself and no other data / metadata while computing the new topic name? Are
you suggesting that since this assumption could work for a majority of
SMTs, it might be more efficient overall in terms of reducing the number of
"false-positive" calls to `SinkTask::closePostTransform` (and we'll also be
able to call `SinkTask::closePostTransform` immediately after topic
partitions are revoked from the consumer)? I was thinking something more
generic along the lines of a simple time based eviction policy that
wouldn't be making any assumptions regarding the SMT implementations.
Either way, I do like your earlier suggestion of keeping this logic
internal and not painting ourselves into a corner by promising any
particular behavior in the KIP.

Thanks,
Yash

On Tue, Feb 14, 2023 at 1:08 AM Chris Egerton 
wrote:

> Hi Yash,
>
> I think the key difference between adding methods/overloads related to
> SinkTask::open/SinkTask::close and SinkTask::put is that this isn't
> auxiliary information that may or may not be useful to connector
> developers. It's actually critical for them to understand the difference
> between the two concepts here, even if they look very similar. And yes, I
> do believe that switching from pre-transform to post-transform topic
> partitions is too big a change in behavior here. Plus, if a connector is
> intentionally designed to use pre-transformation topic partitions in its
> open/close methods, wouldn't we just be trading one form of the problem for
> another by making this switch?
>
> One possible alternative to overloading the existing methods is to split
> SinkTask::open into openOriginal (or possibly openPhysical or
> openPreTransform) and openTransformed (or openLogical or
> openPostTransform), with a similar change for SinkTask::close. The default
> implementation for SinkTask::openOriginal can be to call SinkTask::open,
> and the same can go for SinkTask::close. However, I prefer overloading the
> existing methods since this alternative increases complexity and none of
> the names are very informative.
>
> BTW, I wouldn't say that we can't make assumptions about the relationships
> between pre- and post-transformation topic partitions. We might utilize a
> policy that assumes a deterministic mapping from the former to the latter,
> for example. The distinction I'd draw is that the assumptions we make can
> and probably should favor some cases in terms of performance (i.e.,
> reducing the number of unnecessary calls to close/open over a given sink
> task's lifetime), but should not lead to guaranteed resource leaks or
> failure to obey API contract in any cases.
>
> Cheers,
>
> Chris
>
> On Mon, Feb 13, 2023 at 10:54 AM Yash Mayya  wrote:
>
> > Hi Chris,
> >
> > > especially if connectors are intentionally designed around
> > > original topic partitions instead of transformed ones.
> >
> > Ha, that's a good point and 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Chris Egerton]

Hi Yash,

I think the key difference between adding methods/overloads related to
SinkTask::open/SinkTask::close and SinkTask::put is that this isn't
auxiliary information that may or may not be useful to connector
developers. It's actually critical for them to understand the difference
between the two concepts here, even if they look very similar. And yes, I
do believe that switching from pre-transform to post-transform topic
partitions is too big a change in behavior here. Plus, if a connector is
intentionally designed to use pre-transformation topic partitions in its
open/close methods, wouldn't we just be trading one form of the problem for
another by making this switch?

One possible alternative to overloading the existing methods is to split
SinkTask::open into openOriginal (or possibly openPhysical or
openPreTransform) and openTransformed (or openLogical or
openPostTransform), with a similar change for SinkTask::close. The default
implementation for SinkTask::openOriginal can be to call SinkTask::open,
and the same can go for SinkTask::close. However, I prefer overloading the
existing methods since this alternative increases complexity and none of
the names are very informative.

BTW, I wouldn't say that we can't make assumptions about the relationships
between pre- and post-transformation topic partitions. We might utilize a
policy that assumes a deterministic mapping from the former to the latter,
for example. The distinction I'd draw is that the assumptions we make can
and probably should favor some cases in terms of performance (i.e.,
reducing the number of unnecessary calls to close/open over a given sink
task's lifetime), but should not lead to guaranteed resource leaks or
failure to obey API contract in any cases.

Cheers,

Chris

On Mon, Feb 13, 2023 at 10:54 AM Yash Mayya  wrote:

> Hi Chris,
>
> > especially if connectors are intentionally designed around
> > original topic partitions instead of transformed ones.
>
> Ha, that's a good point and reminds me of Hyrum's Law [1] :)
>
> > I think we have to provide connector developers with some
> > way to differentiate between the two, but maybe there's a way
> >  to do this that I haven't thought of yet
>
> I can't think of a better way to do this either; would invoking the
> existing `SinkTask::open` and `SinkTask::close` methods with post-transform
> topic partitions instead of pre-transform topic partitions not be
> acceptable even in a minor / major AK release? I feel like the proposed
> approach of adding overloaded `SinkTask::open` / `SinkTask::close` methods
> to differentiate between pre-transform and post-transform topic partitions
> has similar pitfalls to the idea of the overloaded `SinkTask::put` method
> we discarded earlier.
>
> > Either way, I'm glad that the general idea of a cache and
> > eviction policy for SinkTask::close seem reasonable; if
> > we decide to go this route, it might make sense for the KIP
> > to include an outline of one or more high-level strategies
> > we might take, but without promising any particular behavior
> > beyond occasionally calling SinkTask::close for post-transform
> > topic partitions. I'm hoping that this logic can stay internal,
> > and by notpainting ourselves into a corner with the KIP, we
> > give ourselves leeway to tweak it in the future if necessary
> > without filing another KIP or introducing a pluggable interface.
>
> Thanks, that's a good idea. Given the flexibility of SMTs, the framework
> can't really make any assumptions around topic partitions post
> transformation nor does it have any way to definitively get any such
> information from transformations which is why the idea of a cache with an
> eviction policy makes perfect sense!
>
> [1] - https://www.hyrumslaw.com/
>
>
> Thanks,
> Yash
>
> On Thu, Feb 9, 2023 at 9:38 PM Chris Egerton 
> wrote:
>
> > Hi Yash,
> >
> > > So it looks like with the current state of affairs, sink tasks that
> only
> > instantiate writers in the SinkTask::open method (and don't do the lazy
> > instantiation in SinkTask::put that you mentioned) might fail when used
> > with topic/partition mutating SMTs even if they don't do any asynchronous
> > processing?
> >
> > Yep, exactly 
> >
> > > What do you think about retaining just the existing methods
> > but changing when they're called in the Connect runtime? For instance,
> > instead of calling SinkTask::open after partition assignment post a
> > consumer group rebalance, we could cache the currently "seen" topic
> > partitions (post transformation) and before each call to SinkTask::put
> > check whether there's any new "unseen" topic partitions, and if so call
> > SinkTask::open (and also update the cache of course).
> >
> > IMO the issue here is that it's a drastic change in behavior to start
> > invoking SinkTask::open and SinkTask::close with post-transform topic
> > partitions instead of pre-transform, especially if connectors are
> > intentionally designed around original topic partitions 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Chris,

> especially if connectors are intentionally designed around
> original topic partitions instead of transformed ones.

Ha, that's a good point and reminds me of Hyrum's Law [1] :)

> I think we have to provide connector developers with some
> way to differentiate between the two, but maybe there's a way
>  to do this that I haven't thought of yet

I can't think of a better way to do this either; would invoking the
existing `SinkTask::open` and `SinkTask::close` methods with post-transform
topic partitions instead of pre-transform topic partitions not be
acceptable even in a minor / major AK release? I feel like the proposed
approach of adding overloaded `SinkTask::open` / `SinkTask::close` methods
to differentiate between pre-transform and post-transform topic partitions
has similar pitfalls to the idea of the overloaded `SinkTask::put` method
we discarded earlier.

> Either way, I'm glad that the general idea of a cache and
> eviction policy for SinkTask::close seem reasonable; if
> we decide to go this route, it might make sense for the KIP
> to include an outline of one or more high-level strategies
> we might take, but without promising any particular behavior
> beyond occasionally calling SinkTask::close for post-transform
> topic partitions. I'm hoping that this logic can stay internal,
> and by notpainting ourselves into a corner with the KIP, we
> give ourselves leeway to tweak it in the future if necessary
> without filing another KIP or introducing a pluggable interface.

Thanks, that's a good idea. Given the flexibility of SMTs, the framework
can't really make any assumptions around topic partitions post
transformation nor does it have any way to definitively get any such
information from transformations which is why the idea of a cache with an
eviction policy makes perfect sense!

[1] - https://www.hyrumslaw.com/


Thanks,
Yash

On Thu, Feb 9, 2023 at 9:38 PM Chris Egerton 
wrote:

> Hi Yash,
>
> > So it looks like with the current state of affairs, sink tasks that only
> instantiate writers in the SinkTask::open method (and don't do the lazy
> instantiation in SinkTask::put that you mentioned) might fail when used
> with topic/partition mutating SMTs even if they don't do any asynchronous
> processing?
>
> Yep, exactly 
>
> > What do you think about retaining just the existing methods
> but changing when they're called in the Connect runtime? For instance,
> instead of calling SinkTask::open after partition assignment post a
> consumer group rebalance, we could cache the currently "seen" topic
> partitions (post transformation) and before each call to SinkTask::put
> check whether there's any new "unseen" topic partitions, and if so call
> SinkTask::open (and also update the cache of course).
>
> IMO the issue here is that it's a drastic change in behavior to start
> invoking SinkTask::open and SinkTask::close with post-transform topic
> partitions instead of pre-transform, especially if connectors are
> intentionally designed around original topic partitions instead of
> transformed ones. I think we have to provide connector developers with some
> way to differentiate between the two, but maybe there's a way to do this
> that I haven't thought of yet. Interested to hear your thoughts.
>
> Either way, I'm glad that the general idea of a cache and eviction policy
> for SinkTask::close seem reasonable; if we decide to go this route, it
> might make sense for the KIP to include an outline of one or more
> high-level strategies we might take, but without promising any particular
> behavior beyond occasionally calling SinkTask::close for post-transform
> topic partitions. I'm hoping that this logic can stay internal, and by not
> painting ourselves into a corner with the KIP, we give ourselves leeway to
> tweak it in the future if necessary without filing another KIP or
> introducing a pluggable interface.
>
> Cheers,
>
> Chris
>
> On Thu, Feb 9, 2023 at 7:39 AM Yash Mayya  wrote:
>
> > Hi Chris,
> >
> > Thanks for the feedback.
> >
> > 1) That's a fair point; while I did scan everything publicly available on
> > GitHub, you're right in that it won't cover all possible SMTs that are
> out
> > there. Thanks for the example use-case as well, I've updated the KIP to
> add
> > the two new proposed methods.
> >
> > 2) So it looks like with the current state of affairs, sink tasks that
> only
> > instantiate writers in the SinkTask::open method (and don't do the lazy
> > instantiation in SinkTask::put that you mentioned) might fail when used
> > with topic/partition mutating SMTs even if they don't do any asynchronous
> > processing? Since they could encounter records in SinkTask::put with
> > topics/partitions that they might not have created writers for. Thanks
> for
> > pointing this out, it's definitely another incompatibility that needs to
> be
> > called out and fixed. The overloaded method approach is interesting, but
> > comes with the caveat of yet more new methods that will need to 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Chris Egerton]

Hi Yash,

> So it looks like with the current state of affairs, sink tasks that only
instantiate writers in the SinkTask::open method (and don't do the lazy
instantiation in SinkTask::put that you mentioned) might fail when used
with topic/partition mutating SMTs even if they don't do any asynchronous
processing?

Yep, exactly 

> What do you think about retaining just the existing methods
but changing when they're called in the Connect runtime? For instance,
instead of calling SinkTask::open after partition assignment post a
consumer group rebalance, we could cache the currently "seen" topic
partitions (post transformation) and before each call to SinkTask::put
check whether there's any new "unseen" topic partitions, and if so call
SinkTask::open (and also update the cache of course).

IMO the issue here is that it's a drastic change in behavior to start
invoking SinkTask::open and SinkTask::close with post-transform topic
partitions instead of pre-transform, especially if connectors are
intentionally designed around original topic partitions instead of
transformed ones. I think we have to provide connector developers with some
way to differentiate between the two, but maybe there's a way to do this
that I haven't thought of yet. Interested to hear your thoughts.

Either way, I'm glad that the general idea of a cache and eviction policy
for SinkTask::close seem reasonable; if we decide to go this route, it
might make sense for the KIP to include an outline of one or more
high-level strategies we might take, but without promising any particular
behavior beyond occasionally calling SinkTask::close for post-transform
topic partitions. I'm hoping that this logic can stay internal, and by not
painting ourselves into a corner with the KIP, we give ourselves leeway to
tweak it in the future if necessary without filing another KIP or
introducing a pluggable interface.

Cheers,

Chris

On Thu, Feb 9, 2023 at 7:39 AM Yash Mayya  wrote:

> Hi Chris,
>
> Thanks for the feedback.
>
> 1) That's a fair point; while I did scan everything publicly available on
> GitHub, you're right in that it won't cover all possible SMTs that are out
> there. Thanks for the example use-case as well, I've updated the KIP to add
> the two new proposed methods.
>
> 2) So it looks like with the current state of affairs, sink tasks that only
> instantiate writers in the SinkTask::open method (and don't do the lazy
> instantiation in SinkTask::put that you mentioned) might fail when used
> with topic/partition mutating SMTs even if they don't do any asynchronous
> processing? Since they could encounter records in SinkTask::put with
> topics/partitions that they might not have created writers for. Thanks for
> pointing this out, it's definitely another incompatibility that needs to be
> called out and fixed. The overloaded method approach is interesting, but
> comes with the caveat of yet more new methods that will need to be
> implemented by existing connectors if they want to make use of this new
> functionality. What do you think about retaining just the existing methods
> but changing when they're called in the Connect runtime? For instance,
> instead of calling SinkTask::open after partition assignment post a
> consumer group rebalance, we could cache the currently "seen" topic
> partitions (post transformation) and before each call to SinkTask::put
> check whether there's any new "unseen" topic partitions, and if so call
> SinkTask::open (and also update the cache of course). I don't think this
> would break the existing contract with sink tasks where SinkTask::open is
> expected to be called for a topic partition before any records from the
> topic partition are sent via SinkTask::put? The SinkTask::close case is a
> lot trickier however, and would require some sort of cache eviction policy
> that would be deemed appropriate as you pointed out too.
>
> Thanks,
> Yash
>
> On Mon, Feb 6, 2023 at 11:27 PM Chris Egerton 
> wrote:
>
> > Hi Yash,
> >
> > I've had some time to think on this KIP and I think I'm in agreement
> about
> > not blocking it on an official compatibility library or adding the "ack"
> > API for sink records.
> >
> > I only have two more thoughts:
> >
> > 1. Because it is possible to manipulate sink record partitions and
> offsets
> > with the current API we provide for transformations, I still believe
> > methods should be added to the SinkRecord class to expose the original
> > partition and offset, not just the original topic. The additional
> cognitive
> > burden from these two methods is going to be minimal anyways; once users
> > understand the difference between the transformed topic name and the
> > original one, it's going to be trivial for them to understand how that
> same
> > difference applies for partitions and offsets. It's not enough to scan
> the
> > set of SMTs provided out of the box with Connect, ones developed by
> > Confluent, or even everything available on GitHub, since there may be
> > 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Chris,

Thanks for the feedback.

1) That's a fair point; while I did scan everything publicly available on
GitHub, you're right in that it won't cover all possible SMTs that are out
there. Thanks for the example use-case as well, I've updated the KIP to add
the two new proposed methods.

2) So it looks like with the current state of affairs, sink tasks that only
instantiate writers in the SinkTask::open method (and don't do the lazy
instantiation in SinkTask::put that you mentioned) might fail when used
with topic/partition mutating SMTs even if they don't do any asynchronous
processing? Since they could encounter records in SinkTask::put with
topics/partitions that they might not have created writers for. Thanks for
pointing this out, it's definitely another incompatibility that needs to be
called out and fixed. The overloaded method approach is interesting, but
comes with the caveat of yet more new methods that will need to be
implemented by existing connectors if they want to make use of this new
functionality. What do you think about retaining just the existing methods
but changing when they're called in the Connect runtime? For instance,
instead of calling SinkTask::open after partition assignment post a
consumer group rebalance, we could cache the currently "seen" topic
partitions (post transformation) and before each call to SinkTask::put
check whether there's any new "unseen" topic partitions, and if so call
SinkTask::open (and also update the cache of course). I don't think this
would break the existing contract with sink tasks where SinkTask::open is
expected to be called for a topic partition before any records from the
topic partition are sent via SinkTask::put? The SinkTask::close case is a
lot trickier however, and would require some sort of cache eviction policy
that would be deemed appropriate as you pointed out too.

Thanks,
Yash

On Mon, Feb 6, 2023 at 11:27 PM Chris Egerton 
wrote:

> Hi Yash,
>
> I've had some time to think on this KIP and I think I'm in agreement about
> not blocking it on an official compatibility library or adding the "ack"
> API for sink records.
>
> I only have two more thoughts:
>
> 1. Because it is possible to manipulate sink record partitions and offsets
> with the current API we provide for transformations, I still believe
> methods should be added to the SinkRecord class to expose the original
> partition and offset, not just the original topic. The additional cognitive
> burden from these two methods is going to be minimal anyways; once users
> understand the difference between the transformed topic name and the
> original one, it's going to be trivial for them to understand how that same
> difference applies for partitions and offsets. It's not enough to scan the
> set of SMTs provided out of the box with Connect, ones developed by
> Confluent, or even everything available on GitHub, since there may be
> closed-source projects out there that rely on this ability. One potential
> use case could be re-routing partitions between Kafka and some other
> sharded system.
>
> 2. We still have to address the SinkTask::open [1] and SinkTask::close [2]
> methods. If a connector writes to the external system using the transformed
> topic partitions it reads from Kafka, then it's possible for the connector
> to lazily instantiate writers for topic partitions as it encounters them
> from records provided in SinkTask::put. However, connectors also need a way
> to de-allocate those writers (and the resources used by them) over time,
> which they can't do as easily. One possible approach here is to overload
> SinkTask::open and SinkTask::close with variants that distinguish between
> transformed and original topic partitions, and default to invoking the
> existing methods with just the original topic partitions. We would then
> have several options for how the Connect runtime can invoke these methods,
> but in general, an approach that guarantees that tasks are notified of
> transformed topic partitions in SinkTask::open before any records for that
> partition are given to it in SinkTask::put, and makes a best-effort attempt
> to close transformed topic partitions that appear to no longer be in use
> based on some eviction policy, would probably be sufficient.
>
> [1] -
>
> https://kafka.apache.org/33/javadoc/org/apache/kafka/connect/sink/SinkTask.html#open(java.util.Collection)
> [2] -
>
> https://kafka.apache.org/33/javadoc/org/apache/kafka/connect/sink/SinkTask.html#close(java.util.Collection)
>
> Cheers,
>
> Chris
>
> On Sat, Nov 5, 2022 at 5:46 AM Yash Mayya  wrote:
>
> > Hi Chris,
> >
> > Thanks a lot for your inputs!
> >
> > > would provide a simple, clean interface for developers to determine
> > > which features are supported by the version of the Connect runtime
> > > that their plugin has been deployed onto
> >
> > I do like the idea of having such a public compatibility library - I
> think
> > it would remove a lot of restrictions from framework development 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Chris Egerton]

Hi Yash,

I've had some time to think on this KIP and I think I'm in agreement about
not blocking it on an official compatibility library or adding the "ack"
API for sink records.

I only have two more thoughts:

1. Because it is possible to manipulate sink record partitions and offsets
with the current API we provide for transformations, I still believe
methods should be added to the SinkRecord class to expose the original
partition and offset, not just the original topic. The additional cognitive
burden from these two methods is going to be minimal anyways; once users
understand the difference between the transformed topic name and the
original one, it's going to be trivial for them to understand how that same
difference applies for partitions and offsets. It's not enough to scan the
set of SMTs provided out of the box with Connect, ones developed by
Confluent, or even everything available on GitHub, since there may be
closed-source projects out there that rely on this ability. One potential
use case could be re-routing partitions between Kafka and some other
sharded system.

2. We still have to address the SinkTask::open [1] and SinkTask::close [2]
methods. If a connector writes to the external system using the transformed
topic partitions it reads from Kafka, then it's possible for the connector
to lazily instantiate writers for topic partitions as it encounters them
from records provided in SinkTask::put. However, connectors also need a way
to de-allocate those writers (and the resources used by them) over time,
which they can't do as easily. One possible approach here is to overload
SinkTask::open and SinkTask::close with variants that distinguish between
transformed and original topic partitions, and default to invoking the
existing methods with just the original topic partitions. We would then
have several options for how the Connect runtime can invoke these methods,
but in general, an approach that guarantees that tasks are notified of
transformed topic partitions in SinkTask::open before any records for that
partition are given to it in SinkTask::put, and makes a best-effort attempt
to close transformed topic partitions that appear to no longer be in use
based on some eviction policy, would probably be sufficient.

[1] -
https://kafka.apache.org/33/javadoc/org/apache/kafka/connect/sink/SinkTask.html#open(java.util.Collection)
[2] -
https://kafka.apache.org/33/javadoc/org/apache/kafka/connect/sink/SinkTask.html#close(java.util.Collection)

Cheers,

Chris

On Sat, Nov 5, 2022 at 5:46 AM Yash Mayya  wrote:

> Hi Chris,
>
> Thanks a lot for your inputs!
>
> > would provide a simple, clean interface for developers to determine
> > which features are supported by the version of the Connect runtime
> > that their plugin has been deployed onto
>
> I do like the idea of having such a public compatibility library - I think
> it would remove a lot of restrictions from framework development if it were
> to be widely adopted.
>
> > we might consider adding an API to "ack" sink records
>
> I agree that this does seem like a more intuitive and clean API, but I'm
> concerned about the backward compatibility headache we'd be imposing on all
> existing sink connectors. Connector developers will have to maintain two
> separate ways of doing offset management if they want to use the new API
> but continue supporting older versions of Kafka Connect.
>
> For now, I've reverted the KIP to the previous iteration which proposed the
> addition of a new `SinkRecord` method to obtain the original Kafka topic
> pre-transformation. One thing to note is that I've removed the method for
> obtaining the original Kafka partition after a cursory search showed that
> use cases for partition modifying SMTs are primarily on the source
> connector side.
>
> Thanks,
> Yash
>
> On Tue, Nov 1, 2022 at 9:22 PM Chris Egerton 
> wrote:
>
> > Hi all,
> >
> > I have more comments I'd like to make on this KIP when I have time (sorry
> > for the delay, Yash, and thanks for your patience!), but I did want to
> > chime in and say that I'm also not sure about overloading SinkTask::put.
> I
> > share the concerns about creating an intuitive, simple API that Yash has
> > raised. In addition, this approach doesn't seem very sustainable--what do
> > we do if we encounter another case in the future that would warrant a
> > similar solution? We probably don't want to create three, four, etc.
> > overloaded variants of the method, each of which would have to be
> > implemented by connector developers who want to both leverage the latest
> > and greatest connector APIs and maintain compatibility with connect
> > Clusters running older versions.
> >
> > I haven't been able to flesh this out into a design worth publishing in
> its
> > own KIP yet, but one alternative I've pitched to a few people with
> > generally positive interest has been to develop an official compatibility
> > library for Connect developers. This library would be released as its own
> 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Chris,

Thanks a lot for your inputs!

> would provide a simple, clean interface for developers to determine
> which features are supported by the version of the Connect runtime
> that their plugin has been deployed onto

I do like the idea of having such a public compatibility library - I think
it would remove a lot of restrictions from framework development if it were
to be widely adopted.

> we might consider adding an API to "ack" sink records

I agree that this does seem like a more intuitive and clean API, but I'm
concerned about the backward compatibility headache we'd be imposing on all
existing sink connectors. Connector developers will have to maintain two
separate ways of doing offset management if they want to use the new API
but continue supporting older versions of Kafka Connect.

For now, I've reverted the KIP to the previous iteration which proposed the
addition of a new `SinkRecord` method to obtain the original Kafka topic
pre-transformation. One thing to note is that I've removed the method for
obtaining the original Kafka partition after a cursory search showed that
use cases for partition modifying SMTs are primarily on the source
connector side.

Thanks,
Yash

On Tue, Nov 1, 2022 at 9:22 PM Chris Egerton 
wrote:

> Hi all,
>
> I have more comments I'd like to make on this KIP when I have time (sorry
> for the delay, Yash, and thanks for your patience!), but I did want to
> chime in and say that I'm also not sure about overloading SinkTask::put. I
> share the concerns about creating an intuitive, simple API that Yash has
> raised. In addition, this approach doesn't seem very sustainable--what do
> we do if we encounter another case in the future that would warrant a
> similar solution? We probably don't want to create three, four, etc.
> overloaded variants of the method, each of which would have to be
> implemented by connector developers who want to both leverage the latest
> and greatest connector APIs and maintain compatibility with connect
> Clusters running older versions.
>
> I haven't been able to flesh this out into a design worth publishing in its
> own KIP yet, but one alternative I've pitched to a few people with
> generally positive interest has been to develop an official compatibility
> library for Connect developers. This library would be released as its own
> Maven artifact (separate from connect-api, connect-runtime, etc.) and would
> provide a simple, clean interface for developers to determine which
> features are supported by the version of the Connect runtime that their
> plugin has been deployed onto. Under the hood, this library might use
> reflection to determine whether classes, methods, etc. are available, but
> the developer wouldn't have to do anything more than check (for example)
> `Features.SINK_TASK_ERRANT_RECORD_REPORTER.enabled()` to know at any point
> in the lifetime of their connector/task whether that feature is provided by
> the runtime.
>
> One other high-level comment: this doesn't address every case, but we might
> consider adding an API to "ack" sink records. This could use the
> SubmittedRecords class [1] (with some slight tweaks) under the hood to
> track the latest-acked offset for each topic partition. This way, connector
> developers won't be responsible for tracking offsets at all in their sink
> tasks (eliminating issues with the accuracy of post-transformation T/P/O
> sink record information), and they'll only have to notify the Connect
> framework when a record has been successfully dispatched to the external
> system. This provides a cleaner, friendlier API, and also enables more
> fine-grained metrics like the ones proposed in KIP-767 [2].
>
> [1] -
>
> https://github.com/apache/kafka/blob/9ab140d5419d735baae45aff56ffce7f5622744f/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/SubmittedRecords.java
> [2] -
>
> https://cwiki.apache.org/confluence/display/KAFKA/KIP-767%3A+Connect+Latency+Metrics
>
> Cheers,
>
> Chris
>
> On Tue, Nov 1, 2022 at 11:21 AM Yash Mayya  wrote:
>
> > Hi Randall,
> >
> > It's been a while for this one but the more I think about it, the more I
> > feel like the current approach with a new overloaded `SinkTask::put`
> method
> > might not be optimal. We're trying to fix a pretty corner case bug here
> > (usage of topic mutating SMTs with sink connectors that do their own
> offset
> > tracking) and I'm not sure that warrants a change to such a central
> > interface method. The new `SinkTask::put` method just seems somewhat odd
> > and it may not be very understandable for a new reader - I don't think
> this
> > should be the case for a public interface method. Furthermore, even with
> > elaborate documentation in place, I'm not sure if it'll be very obvious
> to
> > most people what the purpose of having these two `put` methods is and how
> > they should be used by sink task implementations. What do you think?
> >
> > Thanks,
> > Yash
> >
> > On Mon, Oct 10, 2022 at 9:33 PM Yash Mayya  wrote:

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Chris Egerton]

Hi all,

I have more comments I'd like to make on this KIP when I have time (sorry
for the delay, Yash, and thanks for your patience!), but I did want to
chime in and say that I'm also not sure about overloading SinkTask::put. I
share the concerns about creating an intuitive, simple API that Yash has
raised. In addition, this approach doesn't seem very sustainable--what do
we do if we encounter another case in the future that would warrant a
similar solution? We probably don't want to create three, four, etc.
overloaded variants of the method, each of which would have to be
implemented by connector developers who want to both leverage the latest
and greatest connector APIs and maintain compatibility with connect
Clusters running older versions.

I haven't been able to flesh this out into a design worth publishing in its
own KIP yet, but one alternative I've pitched to a few people with
generally positive interest has been to develop an official compatibility
library for Connect developers. This library would be released as its own
Maven artifact (separate from connect-api, connect-runtime, etc.) and would
provide a simple, clean interface for developers to determine which
features are supported by the version of the Connect runtime that their
plugin has been deployed onto. Under the hood, this library might use
reflection to determine whether classes, methods, etc. are available, but
the developer wouldn't have to do anything more than check (for example)
`Features.SINK_TASK_ERRANT_RECORD_REPORTER.enabled()` to know at any point
in the lifetime of their connector/task whether that feature is provided by
the runtime.

One other high-level comment: this doesn't address every case, but we might
consider adding an API to "ack" sink records. This could use the
SubmittedRecords class [1] (with some slight tweaks) under the hood to
track the latest-acked offset for each topic partition. This way, connector
developers won't be responsible for tracking offsets at all in their sink
tasks (eliminating issues with the accuracy of post-transformation T/P/O
sink record information), and they'll only have to notify the Connect
framework when a record has been successfully dispatched to the external
system. This provides a cleaner, friendlier API, and also enables more
fine-grained metrics like the ones proposed in KIP-767 [2].

[1] -
https://github.com/apache/kafka/blob/9ab140d5419d735baae45aff56ffce7f5622744f/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/SubmittedRecords.java
[2] -
https://cwiki.apache.org/confluence/display/KAFKA/KIP-767%3A+Connect+Latency+Metrics

Cheers,

Chris

On Tue, Nov 1, 2022 at 11:21 AM Yash Mayya  wrote:

> Hi Randall,
>
> It's been a while for this one but the more I think about it, the more I
> feel like the current approach with a new overloaded `SinkTask::put` method
> might not be optimal. We're trying to fix a pretty corner case bug here
> (usage of topic mutating SMTs with sink connectors that do their own offset
> tracking) and I'm not sure that warrants a change to such a central
> interface method. The new `SinkTask::put` method just seems somewhat odd
> and it may not be very understandable for a new reader - I don't think this
> should be the case for a public interface method. Furthermore, even with
> elaborate documentation in place, I'm not sure if it'll be very obvious to
> most people what the purpose of having these two `put` methods is and how
> they should be used by sink task implementations. What do you think?
>
> Thanks,
> Yash
>
> On Mon, Oct 10, 2022 at 9:33 PM Yash Mayya  wrote:
>
> > Hi Randall,
> >
> > Thanks a lot for your valuable feedback so far! I've updated the KIP
> based
> > on our discussion above. Could you please take another look?
> >
> > Thanks,
> > Yash
> >
> > On Tue, Oct 4, 2022 at 12:40 AM Randall Hauch  wrote:
> >
> >> On Mon, Oct 3, 2022 at 11:45 AM Yash Mayya 
> wrote:
> >>
> >> > Hi Randall,
> >> >
> >> > Thanks for elaborating. I think these are all very good points and I
> see
> >> > why the overloaded `SinkTask::put` method is a cleaner solution
> overall.
> >> >
> >> > > public void put(Collection records, Map >> > TopicPartition> updatedTopicPartitions)
> >> >
> >> > I think this should be
> >> >
> >> > `public void put(Collection records, Map >> > TopicPartition> originalTopicPartitions)`
> >> >
> >> > instead because the sink records themselves have the updated topic
> >> > partitions (i.e. after all transformations have been applied) and the
> >> KIP
> >> > is proposing a way for the tasks to be able to access the original
> topic
> >> > partition (i.e. before transformations have been applied).
> >> >
> >>
> >> Sounds good.
> >>
> >>
> >> >
> >> > > Of course, if the developer does not need separate methods, they can
> >> > easily have the older `put` method simply delegate to the newer
> method.
> >> >
> >> > If the developer does not need separate methods (i.e. they don't need
> to
> >> > use this new addition), they 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Randall,

It's been a while for this one but the more I think about it, the more I
feel like the current approach with a new overloaded `SinkTask::put` method
might not be optimal. We're trying to fix a pretty corner case bug here
(usage of topic mutating SMTs with sink connectors that do their own offset
tracking) and I'm not sure that warrants a change to such a central
interface method. The new `SinkTask::put` method just seems somewhat odd
and it may not be very understandable for a new reader - I don't think this
should be the case for a public interface method. Furthermore, even with
elaborate documentation in place, I'm not sure if it'll be very obvious to
most people what the purpose of having these two `put` methods is and how
they should be used by sink task implementations. What do you think?

Thanks,
Yash

On Mon, Oct 10, 2022 at 9:33 PM Yash Mayya  wrote:

> Hi Randall,
>
> Thanks a lot for your valuable feedback so far! I've updated the KIP based
> on our discussion above. Could you please take another look?
>
> Thanks,
> Yash
>
> On Tue, Oct 4, 2022 at 12:40 AM Randall Hauch  wrote:
>
>> On Mon, Oct 3, 2022 at 11:45 AM Yash Mayya  wrote:
>>
>> > Hi Randall,
>> >
>> > Thanks for elaborating. I think these are all very good points and I see
>> > why the overloaded `SinkTask::put` method is a cleaner solution overall.
>> >
>> > > public void put(Collection records, Map> > TopicPartition> updatedTopicPartitions)
>> >
>> > I think this should be
>> >
>> > `public void put(Collection records, Map> > TopicPartition> originalTopicPartitions)`
>> >
>> > instead because the sink records themselves have the updated topic
>> > partitions (i.e. after all transformations have been applied) and the
>> KIP
>> > is proposing a way for the tasks to be able to access the original topic
>> > partition (i.e. before transformations have been applied).
>> >
>>
>> Sounds good.
>>
>>
>> >
>> > > Of course, if the developer does not need separate methods, they can
>> > easily have the older `put` method simply delegate to the newer method.
>> >
>> > If the developer does not need separate methods (i.e. they don't need to
>> > use this new addition), they can simply continue implementing just the
>> > older `put` method right?
>> >
>>
>> Correct. We should update the JavaDoc of both methods to make this clear,
>> and in general how the two methods should are used and should be
>> implemented. That can be part of the PR, and the KIP doesn't need this
>> wording.
>>
>> >
>> > > Finally, this gives us a roadmap for *eventually* deprecating the
>> older
>> > method, once the Connect runtime versions without this change are old
>> > enough.
>> >
>> > I'm not sure we'd ever want to deprecate the older method. Most common
>> sink
>> > connector implementations do not do their own offset tracking with
>> > asynchronous processing and will probably never have a need for the
>> > additional parameter `Map
>> > originalTopicPartitions` in the proposed new `put` method. These
>> connectors
>> > can continue implementing only the existing `SinkTask::put` method which
>> > will be called by the default implementation of the newer overloaded
>> `put`
>> > method.
>> >
>>
>> +1
>>
>>
>> >
>> > > the pre-commit methods use the same `Map> > OffsetAndMetadata> currentOffsets` data structure I'm suggesting be
>> used.
>> >
>> > The data structure you're suggesting be used is a `Map> > TopicPartition>` which will map `SinkRecord` objects to the original
>> topic
>> > partition of the corresponding `ConsumerRecord` right? To clarify, this
>> is
>> > a new data structure that will need to be managed in the
>> `WorkerSinkTask`.
>> >
>>
>> Ah, you're right. Thanks for the correction.
>>
>> Best regards,
>> Randall
>>
>>
>> > Thanks,
>> > Yash
>>
>>
>> > On Mon, Oct 3, 2022 at 1:20 AM Randall Hauch  wrote:
>> >
>> > > Hi, Yash.
>> > >
>> > > I'm not sure I quite understand why it would be "easier" for connector
>> > > > developers to account for implementing two different overloaded
>> `put`
>> > > > methods (assuming that they want to use this new feature) versus
>> using
>> > a
>> > > > try-catch block around `SinkRecord` access methods?
>> > >
>> > >
>> > > Using a try-catch to try around an API method that *might* be there
>> is a
>> > > very unusual thing for most developers. Unfortunately, we've had to
>> > resort
>> > > to this atypical approach with Connect in places when there was no
>> good
>> > > alternative. We seem to relying upon pattern because it's easier for
>> us,
>> > > not because it offers a better experience for Connector developers.
>> IMO,
>> > if
>> > > there's a practical alternative that uses normal development practices
>> > and
>> > > techniques, then we should use that alternative. IIUC, there is at
>> least
>> > > one practical alternative for this KIP that would not require
>> developers
>> > to
>> > > use the unusual try-catch to handle the case where methods are not
>> found.
>> > >
>> > > I also think 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Randall,

Thanks a lot for your valuable feedback so far! I've updated the KIP based
on our discussion above. Could you please take another look?

Thanks,
Yash

On Tue, Oct 4, 2022 at 12:40 AM Randall Hauch  wrote:

> On Mon, Oct 3, 2022 at 11:45 AM Yash Mayya  wrote:
>
> > Hi Randall,
> >
> > Thanks for elaborating. I think these are all very good points and I see
> > why the overloaded `SinkTask::put` method is a cleaner solution overall.
> >
> > > public void put(Collection records, Map > TopicPartition> updatedTopicPartitions)
> >
> > I think this should be
> >
> > `public void put(Collection records, Map > TopicPartition> originalTopicPartitions)`
> >
> > instead because the sink records themselves have the updated topic
> > partitions (i.e. after all transformations have been applied) and the KIP
> > is proposing a way for the tasks to be able to access the original topic
> > partition (i.e. before transformations have been applied).
> >
>
> Sounds good.
>
>
> >
> > > Of course, if the developer does not need separate methods, they can
> > easily have the older `put` method simply delegate to the newer method.
> >
> > If the developer does not need separate methods (i.e. they don't need to
> > use this new addition), they can simply continue implementing just the
> > older `put` method right?
> >
>
> Correct. We should update the JavaDoc of both methods to make this clear,
> and in general how the two methods should are used and should be
> implemented. That can be part of the PR, and the KIP doesn't need this
> wording.
>
> >
> > > Finally, this gives us a roadmap for *eventually* deprecating the older
> > method, once the Connect runtime versions without this change are old
> > enough.
> >
> > I'm not sure we'd ever want to deprecate the older method. Most common
> sink
> > connector implementations do not do their own offset tracking with
> > asynchronous processing and will probably never have a need for the
> > additional parameter `Map
> > originalTopicPartitions` in the proposed new `put` method. These
> connectors
> > can continue implementing only the existing `SinkTask::put` method which
> > will be called by the default implementation of the newer overloaded
> `put`
> > method.
> >
>
> +1
>
>
> >
> > > the pre-commit methods use the same `Map > OffsetAndMetadata> currentOffsets` data structure I'm suggesting be used.
> >
> > The data structure you're suggesting be used is a `Map > TopicPartition>` which will map `SinkRecord` objects to the original
> topic
> > partition of the corresponding `ConsumerRecord` right? To clarify, this
> is
> > a new data structure that will need to be managed in the
> `WorkerSinkTask`.
> >
>
> Ah, you're right. Thanks for the correction.
>
> Best regards,
> Randall
>
>
> > Thanks,
> > Yash
>
>
> > On Mon, Oct 3, 2022 at 1:20 AM Randall Hauch  wrote:
> >
> > > Hi, Yash.
> > >
> > > I'm not sure I quite understand why it would be "easier" for connector
> > > > developers to account for implementing two different overloaded `put`
> > > > methods (assuming that they want to use this new feature) versus
> using
> > a
> > > > try-catch block around `SinkRecord` access methods?
> > >
> > >
> > > Using a try-catch to try around an API method that *might* be there is
> a
> > > very unusual thing for most developers. Unfortunately, we've had to
> > resort
> > > to this atypical approach with Connect in places when there was no good
> > > alternative. We seem to relying upon pattern because it's easier for
> us,
> > > not because it offers a better experience for Connector developers.
> IMO,
> > if
> > > there's a practical alternative that uses normal development practices
> > and
> > > techniques, then we should use that alternative. IIUC, there is at
> least
> > > one practical alternative for this KIP that would not require
> developers
> > to
> > > use the unusual try-catch to handle the case where methods are not
> found.
> > >
> > > I also think having two `put` methods is easier when the Connector has
> to
> > > do different things for different Connect runtimes, too. One of those
> > > methods is called by newer Connect runtimes with the new behavior, and
> > the
> > > other method is called by an older Connect runtime. Of course, if the
> > > developer does not need separate methods, they can easily have the
> older
> > > `put` method simply delegate to the newer method.
> > >
> > > Finally, this gives us a roadmap for *eventually* deprecating the older
> > > method, once the Connect runtime versions without this change are old
> > > enough.
> > >
> > > I think the advantage of going with the
> > > > proposed approach in the KIP is that it wouldn't require extra
> > > book-keeping
> > > > (the Map > > > TopicPartition> in `WorkerSinkTask` in your proposed approach)
> > > >
> > >
> > > The connector does have to do some of this bookkeeping in how they
> track
> > > the topic partition offsets used in the `preCommit`, and the pre-commit
> > > methods use the 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Randall Hauch]

On Mon, Oct 3, 2022 at 11:45 AM Yash Mayya  wrote:

> Hi Randall,
>
> Thanks for elaborating. I think these are all very good points and I see
> why the overloaded `SinkTask::put` method is a cleaner solution overall.
>
> > public void put(Collection records, Map TopicPartition> updatedTopicPartitions)
>
> I think this should be
>
> `public void put(Collection records, Map TopicPartition> originalTopicPartitions)`
>
> instead because the sink records themselves have the updated topic
> partitions (i.e. after all transformations have been applied) and the KIP
> is proposing a way for the tasks to be able to access the original topic
> partition (i.e. before transformations have been applied).
>

Sounds good.


>
> > Of course, if the developer does not need separate methods, they can
> easily have the older `put` method simply delegate to the newer method.
>
> If the developer does not need separate methods (i.e. they don't need to
> use this new addition), they can simply continue implementing just the
> older `put` method right?
>

Correct. We should update the JavaDoc of both methods to make this clear,
and in general how the two methods should are used and should be
implemented. That can be part of the PR, and the KIP doesn't need this
wording.

>
> > Finally, this gives us a roadmap for *eventually* deprecating the older
> method, once the Connect runtime versions without this change are old
> enough.
>
> I'm not sure we'd ever want to deprecate the older method. Most common sink
> connector implementations do not do their own offset tracking with
> asynchronous processing and will probably never have a need for the
> additional parameter `Map
> originalTopicPartitions` in the proposed new `put` method. These connectors
> can continue implementing only the existing `SinkTask::put` method which
> will be called by the default implementation of the newer overloaded `put`
> method.
>

+1


>
> > the pre-commit methods use the same `Map OffsetAndMetadata> currentOffsets` data structure I'm suggesting be used.
>
> The data structure you're suggesting be used is a `Map TopicPartition>` which will map `SinkRecord` objects to the original topic
> partition of the corresponding `ConsumerRecord` right? To clarify, this is
> a new data structure that will need to be managed in the `WorkerSinkTask`.
>

Ah, you're right. Thanks for the correction.

Best regards,
Randall


> Thanks,
> Yash


> On Mon, Oct 3, 2022 at 1:20 AM Randall Hauch  wrote:
>
> > Hi, Yash.
> >
> > I'm not sure I quite understand why it would be "easier" for connector
> > > developers to account for implementing two different overloaded `put`
> > > methods (assuming that they want to use this new feature) versus using
> a
> > > try-catch block around `SinkRecord` access methods?
> >
> >
> > Using a try-catch to try around an API method that *might* be there is a
> > very unusual thing for most developers. Unfortunately, we've had to
> resort
> > to this atypical approach with Connect in places when there was no good
> > alternative. We seem to relying upon pattern because it's easier for us,
> > not because it offers a better experience for Connector developers. IMO,
> if
> > there's a practical alternative that uses normal development practices
> and
> > techniques, then we should use that alternative. IIUC, there is at least
> > one practical alternative for this KIP that would not require developers
> to
> > use the unusual try-catch to handle the case where methods are not found.
> >
> > I also think having two `put` methods is easier when the Connector has to
> > do different things for different Connect runtimes, too. One of those
> > methods is called by newer Connect runtimes with the new behavior, and
> the
> > other method is called by an older Connect runtime. Of course, if the
> > developer does not need separate methods, they can easily have the older
> > `put` method simply delegate to the newer method.
> >
> > Finally, this gives us a roadmap for *eventually* deprecating the older
> > method, once the Connect runtime versions without this change are old
> > enough.
> >
> > I think the advantage of going with the
> > > proposed approach in the KIP is that it wouldn't require extra
> > book-keeping
> > > (the Map > > TopicPartition> in `WorkerSinkTask` in your proposed approach)
> > >
> >
> > The connector does have to do some of this bookkeeping in how they track
> > the topic partition offsets used in the `preCommit`, and the pre-commit
> > methods use the same `Map
> > currentOffsets`
> > data structure I'm suggesting be used.
> >
> > I hope that helps.
> >
> > Best regards,
> >
> > Randall
> >
> > On Mon, Sep 26, 2022 at 9:38 AM Yash Mayya  wrote:
> >
> > > Hi Randall,
> > >
> > > Thanks for reviewing the KIP!
> > >
> > > > That latter logic can get quite ugly.
> > >
> > > I'm not sure I quite understand why it would be "easier" for connector
> > > developers to account for implementing two different overloaded `put`
> > > 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Randall,

Thanks for elaborating. I think these are all very good points and I see
why the overloaded `SinkTask::put` method is a cleaner solution overall.

> public void put(Collection records, Map updatedTopicPartitions)

I think this should be

`public void put(Collection records, Map originalTopicPartitions)`

instead because the sink records themselves have the updated topic
partitions (i.e. after all transformations have been applied) and the KIP
is proposing a way for the tasks to be able to access the original topic
partition (i.e. before transformations have been applied).

> Of course, if the developer does not need separate methods, they can
easily have the older `put` method simply delegate to the newer method.

If the developer does not need separate methods (i.e. they don't need to
use this new addition), they can simply continue implementing just the
older `put` method right?

> Finally, this gives us a roadmap for *eventually* deprecating the older
method, once the Connect runtime versions without this change are old
enough.

I'm not sure we'd ever want to deprecate the older method. Most common sink
connector implementations do not do their own offset tracking with
asynchronous processing and will probably never have a need for the
additional parameter `Map
originalTopicPartitions` in the proposed new `put` method. These connectors
can continue implementing only the existing `SinkTask::put` method which
will be called by the default implementation of the newer overloaded `put`
method.

> the pre-commit methods use the same `Map currentOffsets` data structure I'm suggesting be used.

The data structure you're suggesting be used is a `Map` which will map `SinkRecord` objects to the original topic
partition of the corresponding `ConsumerRecord` right? To clarify, this is
a new data structure that will need to be managed in the `WorkerSinkTask`.


Thanks,
Yash

On Mon, Oct 3, 2022 at 1:20 AM Randall Hauch  wrote:

> Hi, Yash.
>
> I'm not sure I quite understand why it would be "easier" for connector
> > developers to account for implementing two different overloaded `put`
> > methods (assuming that they want to use this new feature) versus using a
> > try-catch block around `SinkRecord` access methods?
>
>
> Using a try-catch to try around an API method that *might* be there is a
> very unusual thing for most developers. Unfortunately, we've had to resort
> to this atypical approach with Connect in places when there was no good
> alternative. We seem to relying upon pattern because it's easier for us,
> not because it offers a better experience for Connector developers. IMO, if
> there's a practical alternative that uses normal development practices and
> techniques, then we should use that alternative. IIUC, there is at least
> one practical alternative for this KIP that would not require developers to
> use the unusual try-catch to handle the case where methods are not found.
>
> I also think having two `put` methods is easier when the Connector has to
> do different things for different Connect runtimes, too. One of those
> methods is called by newer Connect runtimes with the new behavior, and the
> other method is called by an older Connect runtime. Of course, if the
> developer does not need separate methods, they can easily have the older
> `put` method simply delegate to the newer method.
>
> Finally, this gives us a roadmap for *eventually* deprecating the older
> method, once the Connect runtime versions without this change are old
> enough.
>
> I think the advantage of going with the
> > proposed approach in the KIP is that it wouldn't require extra
> book-keeping
> > (the Map > TopicPartition> in `WorkerSinkTask` in your proposed approach)
> >
>
> The connector does have to do some of this bookkeeping in how they track
> the topic partition offsets used in the `preCommit`, and the pre-commit
> methods use the same `Map
> currentOffsets`
> data structure I'm suggesting be used.
>
> I hope that helps.
>
> Best regards,
>
> Randall
>
> On Mon, Sep 26, 2022 at 9:38 AM Yash Mayya  wrote:
>
> > Hi Randall,
> >
> > Thanks for reviewing the KIP!
> >
> > > That latter logic can get quite ugly.
> >
> > I'm not sure I quite understand why it would be "easier" for connector
> > developers to account for implementing two different overloaded `put`
> > methods (assuming that they want to use this new feature) versus using a
> > try-catch block around `SinkRecord` access methods? In both cases, a
> > connector developer would need to write additional code in order to
> ensure
> > that their connector continues working with older Connect runtimes.
> > Furthermore, we would probably need to carefully document how the
> > implementation for the older `put` method should look like for connectors
> > that want to use this new feature. I think the advantage of going with
> the
> > proposed approach in the KIP is that it wouldn't require extra
> book-keeping
> > (the Map > TopicPartition> in 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Randall Hauch]

Hi, Yash.

I'm not sure I quite understand why it would be "easier" for connector
> developers to account for implementing two different overloaded `put`
> methods (assuming that they want to use this new feature) versus using a
> try-catch block around `SinkRecord` access methods?


Using a try-catch to try around an API method that *might* be there is a
very unusual thing for most developers. Unfortunately, we've had to resort
to this atypical approach with Connect in places when there was no good
alternative. We seem to relying upon pattern because it's easier for us,
not because it offers a better experience for Connector developers. IMO, if
there's a practical alternative that uses normal development practices and
techniques, then we should use that alternative. IIUC, there is at least
one practical alternative for this KIP that would not require developers to
use the unusual try-catch to handle the case where methods are not found.

I also think having two `put` methods is easier when the Connector has to
do different things for different Connect runtimes, too. One of those
methods is called by newer Connect runtimes with the new behavior, and the
other method is called by an older Connect runtime. Of course, if the
developer does not need separate methods, they can easily have the older
`put` method simply delegate to the newer method.

Finally, this gives us a roadmap for *eventually* deprecating the older
method, once the Connect runtime versions without this change are old
enough.

I think the advantage of going with the
> proposed approach in the KIP is that it wouldn't require extra book-keeping
> (the Map TopicPartition> in `WorkerSinkTask` in your proposed approach)
>

The connector does have to do some of this bookkeeping in how they track
the topic partition offsets used in the `preCommit`, and the pre-commit
methods use the same `Map currentOffsets`
data structure I'm suggesting be used.

I hope that helps.

Best regards,

Randall

On Mon, Sep 26, 2022 at 9:38 AM Yash Mayya  wrote:

> Hi Randall,
>
> Thanks for reviewing the KIP!
>
> > That latter logic can get quite ugly.
>
> I'm not sure I quite understand why it would be "easier" for connector
> developers to account for implementing two different overloaded `put`
> methods (assuming that they want to use this new feature) versus using a
> try-catch block around `SinkRecord` access methods? In both cases, a
> connector developer would need to write additional code in order to ensure
> that their connector continues working with older Connect runtimes.
> Furthermore, we would probably need to carefully document how the
> implementation for the older `put` method should look like for connectors
> that want to use this new feature. I think the advantage of going with the
> proposed approach in the KIP is that it wouldn't require extra book-keeping
> (the Map TopicPartition> in `WorkerSinkTask` in your proposed approach) and also the
> fact that the try-catch based logic is an already established pattern
> through
>
> https://cwiki.apache.org/confluence/display/KAFKA/KIP-610%3A+Error+Reporting+in+Sink+Connectors
> and other KIPs which added methods to source/sink connector/task contexts.
>
> Let me know if you still feel that having a new overloaded put method is a
> cleaner solution and I'd be happy to reconsider!
>
> Thanks,
> Yash
>
> On Thu, Sep 22, 2022 at 11:18 PM Randall Hauch  wrote:
>
> > Hi, Yash. Thanks for picking up this KIP and discussion.
> >
> > The KIP includes this rejected alternative:
> >
> > > 4. Update SinkTask.put in any way to pass the new information outside
> > > SinkRecord (e.g. a Map or a derived class)
> > >
> > >-
> > >
> > >Much more disruptive change without considerable pros
> > >
> > >
> > One advantage about doing this is that sink connector implementations can
> > more easily implement two different "put(...)" methods to handle running
> in
> > a variety of runtimes, without having to use try-catch logic around the
> > newer SinkRecord access methods. That latter logic can get quite ugly.
> >
> > For example, the existing `put` method has this signature:
> >
> > public abstract void put(Collection records);
> >
> > If we added an overloaded method that passed in a map of the old
> > topic+partition for each record (and defined the absence of an entry as
> > having an unchanged topic and partition):
> >
> > public void put(Collection records, Map > TopicPartition> updatedTopicPartitions) {
> > put(records);
> > }
> >
> > then a `SinkTask` implementation that wants to use this new feature could
> > simply implement both methods:
> >
> > public void put(Collection records) {
> > // Running in an older runtime, so no tracking of SMT-modified topic
> names
> > or partitions
> > put(records, Map.of());
> > }
> >
> > public void put(Collection records, Map > TopicPartition> updatedTopicPartitions) {
> > // real logic here
> > }
> >
> > This seems a lot easier than having to use try-catch logic, yet 

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Randall,

Thanks for reviewing the KIP!

> That latter logic can get quite ugly.

I'm not sure I quite understand why it would be "easier" for connector
developers to account for implementing two different overloaded `put`
methods (assuming that they want to use this new feature) versus using a
try-catch block around `SinkRecord` access methods? In both cases, a
connector developer would need to write additional code in order to ensure
that their connector continues working with older Connect runtimes.
Furthermore, we would probably need to carefully document how the
implementation for the older `put` method should look like for connectors
that want to use this new feature. I think the advantage of going with the
proposed approach in the KIP is that it wouldn't require extra book-keeping
(the Map in `WorkerSinkTask` in your proposed approach) and also the
fact that the try-catch based logic is an already established pattern
through
https://cwiki.apache.org/confluence/display/KAFKA/KIP-610%3A+Error+Reporting+in+Sink+Connectors
and other KIPs which added methods to source/sink connector/task contexts.

Let me know if you still feel that having a new overloaded put method is a
cleaner solution and I'd be happy to reconsider!

Thanks,
Yash

On Thu, Sep 22, 2022 at 11:18 PM Randall Hauch  wrote:

> Hi, Yash. Thanks for picking up this KIP and discussion.
>
> The KIP includes this rejected alternative:
>
> > 4. Update SinkTask.put in any way to pass the new information outside
> > SinkRecord (e.g. a Map or a derived class)
> >
> >-
> >
> >Much more disruptive change without considerable pros
> >
> >
> One advantage about doing this is that sink connector implementations can
> more easily implement two different "put(...)" methods to handle running in
> a variety of runtimes, without having to use try-catch logic around the
> newer SinkRecord access methods. That latter logic can get quite ugly.
>
> For example, the existing `put` method has this signature:
>
> public abstract void put(Collection records);
>
> If we added an overloaded method that passed in a map of the old
> topic+partition for each record (and defined the absence of an entry as
> having an unchanged topic and partition):
>
> public void put(Collection records, Map TopicPartition> updatedTopicPartitions) {
> put(records);
> }
>
> then a `SinkTask` implementation that wants to use this new feature could
> simply implement both methods:
>
> public void put(Collection records) {
> // Running in an older runtime, so no tracking of SMT-modified topic names
> or partitions
> put(records, Map.of());
> }
>
> public void put(Collection records, Map TopicPartition> updatedTopicPartitions) {
> // real logic here
> }
>
> This seems a lot easier than having to use try-catch logic, yet still
> allows sink connectors to utilize the new functionality and still work with
> older Connect runtimes.
>
> WDYT?
>
> Randall
>
>
> On Thu, Sep 8, 2022 at 7:03 AM Yash Mayya  wrote:
>
> > Hi all,
> >
> > I would like to (re)start a new discussion thread on KIP-793 (Kafka
> > Connect) which proposes some additions to the public SinkRecord interface
> > in order to support topic mutating SMTs for sink connectors that do their
> > own offset tracking.
> >
> > Links:
> >
> > KIP:
> >
> https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=191336830
> >
> > Older discussion thread:
> > https://lists.apache.org/thread/00kcth6057jdcsyzgy1x8nb2s1cymy8h,
> > https://lists.apache.org/thread/rzqkm0q5y5v3vdjhg8wqppxbkw7nyopj
> >
> > Jira: https://issues.apache.org/jira/browse/KAFKA-13431
> >
> >
> > Thanks,
> > Yash
> >
>

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Randall Hauch]

Hi, Yash. Thanks for picking up this KIP and discussion.

The KIP includes this rejected alternative:

> 4. Update SinkTask.put in any way to pass the new information outside
> SinkRecord (e.g. a Map or a derived class)
>
>-
>
>Much more disruptive change without considerable pros
>
>
One advantage about doing this is that sink connector implementations can
more easily implement two different "put(...)" methods to handle running in
a variety of runtimes, without having to use try-catch logic around the
newer SinkRecord access methods. That latter logic can get quite ugly.

For example, the existing `put` method has this signature:

public abstract void put(Collection records);

If we added an overloaded method that passed in a map of the old
topic+partition for each record (and defined the absence of an entry as
having an unchanged topic and partition):

public void put(Collection records, Map updatedTopicPartitions) {
put(records);
}

then a `SinkTask` implementation that wants to use this new feature could
simply implement both methods:

public void put(Collection records) {
// Running in an older runtime, so no tracking of SMT-modified topic names
or partitions
put(records, Map.of());
}

public void put(Collection records, Map updatedTopicPartitions) {
// real logic here
}

This seems a lot easier than having to use try-catch logic, yet still
allows sink connectors to utilize the new functionality and still work with
older Connect runtimes.

WDYT?

Randall


On Thu, Sep 8, 2022 at 7:03 AM Yash Mayya  wrote:

> Hi all,
>
> I would like to (re)start a new discussion thread on KIP-793 (Kafka
> Connect) which proposes some additions to the public SinkRecord interface
> in order to support topic mutating SMTs for sink connectors that do their
> own offset tracking.
>
> Links:
>
> KIP:
> https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=191336830
>
> Older discussion thread:
> https://lists.apache.org/thread/00kcth6057jdcsyzgy1x8nb2s1cymy8h,
> https://lists.apache.org/thread/rzqkm0q5y5v3vdjhg8wqppxbkw7nyopj
>
> Jira: https://issues.apache.org/jira/browse/KAFKA-13431
>
>
> Thanks,
> Yash
>

[DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi all,

I would like to (re)start a new discussion thread on KIP-793 (Kafka
Connect) which proposes some additions to the public SinkRecord interface
in order to support topic mutating SMTs for sink connectors that do their
own offset tracking.

Links:

KIP:
https://cwiki.apache.org/confluence/pages/viewpage.action?pageId=191336830

Older discussion thread:
https://lists.apache.org/thread/00kcth6057jdcsyzgy1x8nb2s1cymy8h,
https://lists.apache.org/thread/rzqkm0q5y5v3vdjhg8wqppxbkw7nyopj

Jira: https://issues.apache.org/jira/browse/KAFKA-13431


Thanks,
Yash

RE: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Apologies for the poor formatting on the quoted bits in the previous email.

On 2021/11/03 22:17:06 Diego Erdody wrote:
> Hello,
>
> I'd like to propose a small KIP to add a new field to SinkRecord in order
> to add support for topic-mutating SMTs (e.g. RegexRouter) to asynchronous
> Sink Connectors (the ones that override preCommit for internal offset
> tracking, like S3
> <
https://github.com/confluentinc/kafka-connect-storage-cloud/blob/master/kafka-connect-s3/src/main/java/io/confluent/connect/s3/S3SinkTask.java#L274
>
> ).
>
> Links:
>
> - KIP-793: Sink Connectors: Support topic-mutating SMTs for async
> connectors (preCommit users)
> 
> - PR #11464 
>
> Thanks,
>
> Diego
>

RE: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Yash Mayya]

Hi Diego,

Thanks for writing this KIP. Are you still planning to work on this? If
not, I'd be happy to try and take this to completion!

Hi Chris,

Thanks for your valuable inputs as always!

> It looks like this KIP does not take the SinkTask::open and
SinkTask::close methods into account

> I think we may want to add this type of support to the KIP so that we can
solve the mutating SMT/asynchronous sink connector problem once and for all

Could you please clarify what you mean here? Do you mean to say that the
partitions passed to the SinkTask::open / SinkTask::close methods should
not be the original topic partitions? If so, how would we be able to change
that while maintaining backward compatibility? Would we want to add new
methods to SinkTask's public API which connectors could choose to implement
instead? Also, currently SinkTask::open is called when partitions are
assigned to the task's consumer in a consumer rebalance; at this point
there is no way to know of any "transformed" topic partitions - that can
only be determined after the consumer polls and records are converted +
transformed. Similarly, SinkTask::close is called when partitions are
revoked from the task's consumer in a consumer rebalance. Are you
suggesting we change when SinkTask::open and SinkTask::close are called
(maybe by book-keeping all currently known post-transformation topic
partitions)?

>  I agree with the rationale for not exposing more of the original
consumer record for the most part, but what about the record's offset?

Should we pollute the public API with a method that in all likelihood will
never be used? Maybe I'm lacking imagination here, but why would an SMT
ever want to modify the record's offset? Is there any such SMT currently?
It seems to me that adding such a method may just be unnecessary complexity
and a potential source of confusion to Connect developers, WDYT?

> do you think it'd make sense to separate out the newly-proposed
SinkTask::originalTopicPartition method into separate
SinkTask::originalTopic and SinkTask::originalKafkaPartition methods, to
stay in line with the convention that's been loosely set by the existing,
separate SinkTask::topic and SinkTask::kafkaPartition methods?

I would tend to agree with this rationale.

Thanks,
Yash

On 2021/11/03 22:17:06 Diego Erdody wrote:
> Hello,
>
> I'd like to propose a small KIP to add a new field to SinkRecord in order
> to add support for topic-mutating SMTs (e.g. RegexRouter) to asynchronous
> Sink Connectors (the ones that override preCommit for internal offset
> tracking, like S3
> <
https://github.com/confluentinc/kafka-connect-storage-cloud/blob/master/kafka-connect-s3/src/main/java/io/confluent/connect/s3/S3SinkTask.java#L274
>
> ).
>
> Links:
>
> - KIP-793: Sink Connectors: Support topic-mutating SMTs for async
> connectors (preCommit users)
> 
> - PR #11464 
>
> Thanks,
>
> Diego
>

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Randall Hauch]

Thanks for the proposed KIP, Diego. I plan to review this in more
detail in the coming weeks.

Best regards,

Randall

On Wed, Nov 3, 2021 at 5:42 PM Chris Egerton
 wrote:
>
> Hi Diego,
>
> This is a long time coming and I'm glad to see someone's finally gotten
> around to filling in this feature gap for Connect.
>
> It looks like this KIP does not take the SinkTask::open and SinkTask::close
> methods into account (
> https://kafka.apache.org/30/javadoc/org/apache/kafka/connect/sink/SinkTask.html#open(java.util.Collection
> /
> https://kafka.apache.org/30/javadoc/org/apache/kafka/connect/sink/SinkTask.html#close(java.util.Collection)).
> Is this intentional? If so, it'd be nice to see a rationale for leaving
> this out in the rejected alternatives so; if not, I think we may want to
> add this type of support to the KIP so that we can solve the mutating
> SMT/asynchronous sink connector problem once and for all, instead of
> narrowing but not closing the existing feature gap. We may want to take the
> current effort to add support for cooperative consumer groups (
> https://issues.apache.org/jira/browse/KAFKA-12487 /
> https://github.com/apache/kafka/pull/10563) into account if we opt to add
> support for open/close, since the current behavior of Connect (which
> involves invoking SinkTask::close for every topic partition every time a
> consumer rebalance occurs, then invoking SinkTask::open for all
> still-assigned partitions) may be easier to reason about, but is likely
> going to change soon (although it is an option to hold off on that work if
> this KIP is given priority, which is definitely a valid option).
>
> It also looks like we're only exposing the original topic partition to
> connector developers. I agree with the rationale for not exposing more of
> the original consumer record for the most part, but what about the record's
> offset? Although it's not possible to override the Kafka offset for a sink
> record via the standard SinkRecord::newRecord methods (
> https://kafka.apache.org/30/javadoc/org/apache/kafka/connect/sink/SinkRecord.html#newRecord(java.lang.String,java.lang.Integer,org.apache.kafka.connect.data.Schema,java.lang.Object,org.apache.kafka.connect.data.Schema,java.lang.Object,java.lang.Long)
> /
> https://kafka.apache.org/30/javadoc/org/apache/kafka/connect/sink/SinkRecord.html#newRecord(java.lang.String,java.lang.Integer,org.apache.kafka.connect.data.Schema,java.lang.Object,org.apache.kafka.connect.data.Schema,java.lang.Object,java.lang.Long,java.lang.Iterable)),
> there are still public constructors available for the SinkRecord class that
> can be leveraged by SMTs to return new SinkRecord instances that don't have
> the same Kafka offset as the one that they've mutated. Do you think it may
> be worth the additional maintenance burden and API complexity to
> accommodate this case, with something like a SinkTask::originalKafkaOffset
> method?
>
> I'm also wondering about how exactly this method will be implemented. Will
> we automatically create a new SinkRecord instance at the end of the
> transformation chain in order to provide the correct topic partition (and
> possibly offset)? If so, this should be called out since it means that
> transformations that return custom subclasses of SinkRecord will no longer
> be able to do so (or rather, they will still be able to, but these custom
> subclasses will never be visible to sink tasks).
>
> Finally, a small nit: do you think it'd make sense to separate out the
> newly-proposed SinkTask::originalTopicPartition method into separate
> SinkTask::originalTopic and SinkTask::originalKafkaPartition methods, to
> stay in line with the convention that's been loosely set by the existing,
> separate SinkTask::topic and SinkTask::kafkaPartition methods?
>
> I'm personally looking forward to leveraging this improvement in the
> BigQuery sink connector I help maintain because we recently added a new
> write mode that uses asynchronous writes and SinkTask::preCommit, but
> encourage users to use SMTs to redirect records to different
> datasets/tables in BigQuery, which is currently impossible in that write
> mode. Thanks for taking this on!
>
> Cheers,
>
> Chris
>
> On Wed, Nov 3, 2021 at 6:17 PM Diego Erdody  wrote:
>
> > Hello,
> >
> > I'd like to propose a small KIP to add a new field to SinkRecord in order
> > to add support for topic-mutating SMTs (e.g. RegexRouter) to asynchronous
> > Sink Connectors (the ones that override preCommit for internal offset
> > tracking, like S3
> > <
> > https://github.com/confluentinc/kafka-connect-storage-cloud/blob/master/kafka-connect-s3/src/main/java/io/confluent/connect/s3/S3SinkTask.java#L274
> > >
> > ).
> >
> > Links:
> >
> > - KIP-793: Sink Connectors: Support topic-mutating SMTs for async
> > connectors (preCommit users)
> > 
> > - PR #11464 
> >
> > Thanks,
> >
> > Diego
> >

Re: [DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Chris Egerton]

Hi Diego,

This is a long time coming and I'm glad to see someone's finally gotten
around to filling in this feature gap for Connect.

It looks like this KIP does not take the SinkTask::open and SinkTask::close
methods into account (
https://kafka.apache.org/30/javadoc/org/apache/kafka/connect/sink/SinkTask.html#open(java.util.Collection
/
https://kafka.apache.org/30/javadoc/org/apache/kafka/connect/sink/SinkTask.html#close(java.util.Collection)).
Is this intentional? If so, it'd be nice to see a rationale for leaving
this out in the rejected alternatives so; if not, I think we may want to
add this type of support to the KIP so that we can solve the mutating
SMT/asynchronous sink connector problem once and for all, instead of
narrowing but not closing the existing feature gap. We may want to take the
current effort to add support for cooperative consumer groups (
https://issues.apache.org/jira/browse/KAFKA-12487 /
https://github.com/apache/kafka/pull/10563) into account if we opt to add
support for open/close, since the current behavior of Connect (which
involves invoking SinkTask::close for every topic partition every time a
consumer rebalance occurs, then invoking SinkTask::open for all
still-assigned partitions) may be easier to reason about, but is likely
going to change soon (although it is an option to hold off on that work if
this KIP is given priority, which is definitely a valid option).

It also looks like we're only exposing the original topic partition to
connector developers. I agree with the rationale for not exposing more of
the original consumer record for the most part, but what about the record's
offset? Although it's not possible to override the Kafka offset for a sink
record via the standard SinkRecord::newRecord methods (
https://kafka.apache.org/30/javadoc/org/apache/kafka/connect/sink/SinkRecord.html#newRecord(java.lang.String,java.lang.Integer,org.apache.kafka.connect.data.Schema,java.lang.Object,org.apache.kafka.connect.data.Schema,java.lang.Object,java.lang.Long)
/
https://kafka.apache.org/30/javadoc/org/apache/kafka/connect/sink/SinkRecord.html#newRecord(java.lang.String,java.lang.Integer,org.apache.kafka.connect.data.Schema,java.lang.Object,org.apache.kafka.connect.data.Schema,java.lang.Object,java.lang.Long,java.lang.Iterable)),
there are still public constructors available for the SinkRecord class that
can be leveraged by SMTs to return new SinkRecord instances that don't have
the same Kafka offset as the one that they've mutated. Do you think it may
be worth the additional maintenance burden and API complexity to
accommodate this case, with something like a SinkTask::originalKafkaOffset
method?

I'm also wondering about how exactly this method will be implemented. Will
we automatically create a new SinkRecord instance at the end of the
transformation chain in order to provide the correct topic partition (and
possibly offset)? If so, this should be called out since it means that
transformations that return custom subclasses of SinkRecord will no longer
be able to do so (or rather, they will still be able to, but these custom
subclasses will never be visible to sink tasks).

Finally, a small nit: do you think it'd make sense to separate out the
newly-proposed SinkTask::originalTopicPartition method into separate
SinkTask::originalTopic and SinkTask::originalKafkaPartition methods, to
stay in line with the convention that's been loosely set by the existing,
separate SinkTask::topic and SinkTask::kafkaPartition methods?

I'm personally looking forward to leveraging this improvement in the
BigQuery sink connector I help maintain because we recently added a new
write mode that uses asynchronous writes and SinkTask::preCommit, but
encourage users to use SMTs to redirect records to different
datasets/tables in BigQuery, which is currently impossible in that write
mode. Thanks for taking this on!

Cheers,

Chris

On Wed, Nov 3, 2021 at 6:17 PM Diego Erdody  wrote:

> Hello,
>
> I'd like to propose a small KIP to add a new field to SinkRecord in order
> to add support for topic-mutating SMTs (e.g. RegexRouter) to asynchronous
> Sink Connectors (the ones that override preCommit for internal offset
> tracking, like S3
> <
> https://github.com/confluentinc/kafka-connect-storage-cloud/blob/master/kafka-connect-s3/src/main/java/io/confluent/connect/s3/S3SinkTask.java#L274
> >
> ).
>
> Links:
>
> - KIP-793: Sink Connectors: Support topic-mutating SMTs for async
> connectors (preCommit users)
> 
> - PR #11464 
>
> Thanks,
>
> Diego
>

[DISCUSS] KIP-793: Sink Connectors: Support topic-mutating SMTs for async connectors (preCommit users)

[Diego Erdody]

Hello,

I'd like to propose a small KIP to add a new field to SinkRecord in order
to add support for topic-mutating SMTs (e.g. RegexRouter) to asynchronous
Sink Connectors (the ones that override preCommit for internal offset
tracking, like S3

).

Links:

- KIP-793: Sink Connectors: Support topic-mutating SMTs for async
connectors (preCommit users)

- PR #11464 

Thanks,

Diego