[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17850702#comment-17850702
]
Bruno Cadonna commented on KAFKA-13295:
---
I re-opened the issue because we did not release the state updater in 3.6 and
also the fix version was set to 3.4 for reasons I do not know.
The state updater will be most likely released in 3.8. After the release I will
resolve the ticket.
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Priority: Critical
> Labels: eos, new-streams-runtime-should-fix
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17704320#comment-17704320
]
Sagar Rao commented on KAFKA-13295:
---
Thanks [~guozhang] , glad to know that!
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos, new-streams-runtime-should-fix
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17703404#comment-17703404
]
Guozhang Wang commented on KAFKA-13295:
---
[~sagarrao] I had a chat with others driving the 4.0 release and it seems like
it will still take some time, given that I will resume helping you to finish
fixing this bug and not wait for that release.
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos, new-streams-runtime-should-fix
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17698327#comment-17698327
]
Sagar Rao commented on KAFKA-13295:
---
Had a discussion with [~guozhang] and the suggestion is to wait for 4.0 release
when EOS-v1 would be deprecated. It would simpler to implement the changes in
EOS-v2.
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos, new-streams-runtime-should-fix
> Fix For: 4.0.0
>
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17654211#comment-17654211
]
Matthias J. Sax commented on KAFKA-13295:
-
Thanks. SG.
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos, new-streams-runtime-should-fix
> Fix For: 3.5.0
>
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17653025#comment-17653025
]
A. Sophie Blee-Goldman commented on KAFKA-13295:
[~mjsax] unfortunately no, it's a separate issue as we would not even attempt
the commit to begin with in this case.
I've been chatting with Sagar about it but didn't want to push something into
3.4 at the last minute, however I do think we should try to fix in 3.5
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos, new-streams-runtime-should-fix
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17652620#comment-17652620
]
Matthias J. Sax commented on KAFKA-13295:
-
[~ableegoldman] – I am wondering if this issue is resolved via
https://issues.apache.org/jira/browse/KAFKA-14294 implicitly?
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos, new-streams-runtime-should-fix
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17516945#comment-17516945
]
Bruno Cadonna commented on KAFKA-13295:
---
Removing from 3.2. release since code freeze passed.
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos, new-streams-runtime-should-fix
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17443694#comment-17443694
]
David Jacot commented on KAFKA-13295:
-
Moving it to the next release as we are part the code freeze for 3.1.0.
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos
> Fix For: 3.2.0
>
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17429067#comment-17429067
]
A. Sophie Blee-Goldman commented on KAFKA-13295:
1) The rebalance callbacks can take a bit to wrap your head around, hence the
comment trying to explain more or less the flow of task management...but
basically, at the end of each rebalance the following are invoked in this order:
# StreamsRebalanceListener#onPartitionsRevoked (only if there are revoked
partitions)
# StreamsPartitionAssignor#onAssignment (always)
# StreamsRebalanceListener#onPartitionsAssigned (always)
Callback #1 invokes TaskManager#handleRevocation, which cleans up and commits
any active tasks that were revoked. Then callback #2 invokes
TaskManager#handleAssignment, which is what that comment is referring to. This
is where you'll want to add the necessary logic
2) It's probably most instructive to look at the #handleRevocation
implementation, which also covers the case of committing tasks that are not
revoked (because if we have to commit one active task that's revoked, we have
to commit all of them) and how to handle exceptions there. In this case, since
it's EOS by definition, the only possible exceptions are TaskCorruptedException
and a general RuntimeException. For the former case, you can just
closeDirtyAndRevive, while for the latter you should keep track of the
exception and then rethrow it at the end of #handleAssignment
3) Yes, but only if the commit succeeded (ie didn't throw either of the above)
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos
> Fix For: 3.1.0
>
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17426321#comment-17426321
]
Sagar Rao commented on KAFKA-13295:
---
hey [~ableegoldman], whenever you get the chance, plz help me with the couple
of questions from the comment above.
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos
> Fix For: 3.1.0
>
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17422290#comment-17422290
]
Sagar Rao commented on KAFKA-13295:
---
[~ableegoldman], yes it does. I wanted to ask the same question, do we need to
commit offsets for all the tasks or reduce it somehow and you have answered it.
I have a few of more questions ->
1) [~guozhang] has mentioned that we can invoke commitOffsetOrTransaction in
the onPartitionAssigned method. However, in StreamPartitionRebalanceListener,
there's a comment stating that:
{code:java}
//all task management is already handled by:
//
org.apache.kafka.streams.processor.internals.StreamsPartitionAssignor.onAssignment{code}
So, do we need to add the logic in onAssignment? I haven't quite looked at the
flow entirely and hence asking this question
2) How do we handle cases if a TaskCorruptedException or TimeoutException is
thrown during commiting. I see these have been handled in handleRevocation
whereby these are considered dirtyTasks and then closed. So, how do we handle
failure scenarios for offset committing ?
3) Do we also need to write checkpoints post committing?
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos
> Fix For: 3.1.0
>
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17421822#comment-17421822
]
A. Sophie Blee-Goldman commented on KAFKA-13295:
{quote}we always blindly do a commitOffsetOrTransaction as well so that if
there's any ongoing transactions, it would be committed before we potentially
go on to restore the newly assigned tasks.
{quote}
I still think we want to keep the scope of committing offsets as narrow as
possible, since committing adds a nontrivial overhead to the rebalance. The
ideal fix would be to just start checking on the transaction time during
restoration, and if we find ourselves getting too close to the timeout then we
commit. But that's going to introduce some unnecessary complexity, so a good
middle ground might be to just check if `Task#needsCommit` is true for any of
the active tasks. And maybe also to check if we even have any new tasks to
restore. I think those two checks would be simple enough while still keeping
the committing sufficiently narrow in scope.
[~sagarrao] does that make sense?
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos
> Fix For: 3.1.0
>
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17415486#comment-17415486
]
Sagar Rao commented on KAFKA-13295:
---
[~guozhang]/ [~ableegoldman] I assigned this to myself.. Would go through the
code and see if I can find something :D
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Assignee: Sagar Rao
>Priority: Critical
> Labels: eos
> Fix For: 3.1.0
>
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17415195#comment-17415195
]
Guozhang Wang commented on KAFKA-13295:
---
As we talked about the potential fix in the near term, one possibility is that
`onPartitionAssigned` when new tasks are created, we always blindly do a
commitOffsetOrTransaction as well so that if there's any ongoing transactions,
it would be committed before we potentially go on to restore the newly assigned
tasks.
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Priority: Critical
> Labels: eos
> Fix For: 3.1.0
>
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
[jira] [Commented] (KAFKA-13295) Long restoration times for new tasks can lead to transaction timeouts
[
https://issues.apache.org/jira/browse/KAFKA-13295?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17414652#comment-17414652
]
A. Sophie Blee-Goldman commented on KAFKA-13295:
Note that https://issues.apache.org/jira/browse/KAFKA-13249, while not directly
related, makes it more likely for an application to hit this bug since it can
result in Streams restoring additional records and thus longer restoration times
> Long restoration times for new tasks can lead to transaction timeouts
> -
>
> Key: KAFKA-13295
> URL: https://issues.apache.org/jira/browse/KAFKA-13295
> Project: Kafka
> Issue Type: Bug
> Components: streams
>Reporter: A. Sophie Blee-Goldman
>Priority: Critical
> Labels: eos
> Fix For: 3.1.0
>
>
> In some EOS applications with relatively long restoration times we've noticed
> a series of ProducerFencedExceptions occurring during/immediately after
> restoration. The broker logs were able to confirm these were due to
> transactions timing out.
> In Streams, it turns out we automatically begin a new txn when calling
> {{send}} (if there isn’t already one in flight). A {{send}} occurs often
> outside a commit during active processing (eg writing to the changelog),
> leaving the txn open until the next commit. And if a StreamThread has been
> actively processing when a rebalance results in a new stateful task without
> revoking any existing tasks, the thread won’t actually commit this open txn
> before it goes back into the restoration phase while it builds up state for
> the new task. So the in-flight transaction is left open during restoration,
> during which the StreamThread only consumes from the changelog without
> committing, leaving it vulnerable to timing out when restoration times exceed
> the configured transaction.timeout.ms for the producer client.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
