On 1/19/24 22:49, Robert Bradshaw via dev wrote:
I think this standard design could still be made to work.
Specifically, the graph would contain a DoFn that has the
RequiresTimeSortedInput bit set, and as a single "subtransform" that
has a different DoFn in its spec that does not require this bit to be
set and whose implementation enforces this ordering (say, via state)
before invoking the user's DoFn. This would work fine in Streaming for
any runner, and would work OK for batch as long as the value set for
any key fit into memory (or the batch state implementation spilled to
disk, though that could get really slow). Runners that wanted to do
better (e.g. provide timestamp sorting as part of their batch
grouping, or even internally sort timestamps more efficiently than
could be done via the SDK over the state API) could do so.
Yes, this should work fine for portable runners outside Java.
For Java, such a wrapper might be a bit messy, but could probably be
hard coded above the ByteBuddy wrappers layer.
+1
Maybe we can delegate this to an (internal) annotation that would enable
DoFns to define a subclass of DoFnInvokerFactory.
E.g.
class MyDoFn extends DoFn<> {
@DoFnInvokerFactory
MyDoFnInvokerFactory createDoFnInvokerFactory() {
...
}
}
TBD how much of our infrastructure assumes ParDo transforms do not
contain subtransforms. (We could also provide a different URN for
RequresTimeSortedInput DoFns whose payload would be the DoFn payload,
rather than setting a bit on the payload itself.) Rather than
introducing nesting, we could implement the AnyOf PTransform that
would present the two implementations as siblings (which could be
useful elsewhere). This can be made backward compatible by providing
one of the alternatives as the composite structure. The primary
hesitation I have here is that it prevents much
introspection/manipulation of the pipeline before the runner
capabilities are know.
What we really want is a way to annotate a DoFn as
RequestsTimeSortedInput, together with a way for the runner to
communicate to the SDK whether or not it was able to honor this
request. That may be a more invasive change to the protocol (e.g.
annotating PCollections with ordering properties, which is where it
belongs[1]). I suppose we could let a runner that supports this
capability strip the RequestsTimeSortedInput bit (or set a new bit),
and SDKs that get unmutated transforms would know they have to do the
sorting themselves.
That sounds more like runners that are able to provide the sorting
themselves would have a manual override for the sorted ParDo (be it via
a bit or specific URN), no?
Jan
- Robert
[1] Ordering is an under-defined concept in Beam, but if we're going
to add it my take would be that to do it properly one would want
(1) Annotations on PCollections indicating whether they're unordered
or ordered (by a certain ordering criteria, in this case
timestamp-within-key), which could be largely inferred by
(2) Annotations on PTransforms indicating whether they're
order-creating, order-preserving, or order-requiring (with the default
being unspeciified=order-destroying), again parameterized by an
ordering criteria of some kind, which criteria could for a hierarchy.
On Fri, Jan 19, 2024 at 10:40 AM Kenneth Knowles<k...@apache.org> wrote:
In this design space, what we have done in the past is:
1) ensure that runners all reject pipelines they cannot run correctly
2) if there is a default/workaround/slower implementation, provide it as an
override
This is largely ignoring portability but I think/hope it will still work. At one time I
put some effort into ensuring Java Pipeline objects and proto representations could
roundtrip with all the necessary information for pre-portability runners to still work,
which is the same prereqs as pre-portable "Override" implementations to still
work.
TBH I'm 50/50 on the idea. If something is going to be implemented more slowly
or less scalably as a fallback, I think it may be best to simply be upfront
about being unable to really run it. It would depend on the situation. For
requiring time sorted input, the manual implementation is probably similar to
what a streaming runner might do, so it might make sense.
Kenn
On Fri, Jan 19, 2024 at 11:05 AM Robert Burke<rob...@frantil.com> wrote:
I certainly don't have the deeper java insight here. So one more portable based
reply and then I'll step back on the Java specifics.
Portable runners only really have the "unknown Composite" fallback option,
where if the Composite's URN isn't known to the runner, it should use the subgraph that
is being wrapped.
I suppose the protocol could be expanded : If a composite transform with a
ParDo payload, and urn has features the runner can't handle, then it could use
the fallback graph as well.
The SDK would have then still needed to have construct the fallback graph into
the Pipeline proto. This doesn't sound incompatible with what you've suggested
the Java SDK could do, but it avoids the runner needing to be aware of a
specific implementation requirement around a feature it doesn't support. If it
has to do something specific to support an SDK specific mechanism, that's still
supporting the feature, but I fear it's not a great road to tread on for
runners to add SDK specific implementation details.
If a (portable) runner is going to spend work on doing something to handle
RequiresTimeSortedInput, it's probably easier to handle it generally than to
try to enable a Java specific work around. I'm not even sure how that could
work since the SDK would then need a special interpretation of what a runner
sent back for it to do any SDK side special backup handling, vs the simple
execution of the given transform.
It's entirely possible I've over simplified the "fallback" protocol described
above, so this thread is still useful for my Prism work, especially if I see any similar
situations once I start on the Java Validates Runner suite.
Robert Burke
Beam Go Busybody
On Fri, Jan 19, 2024, 6:41 AM Jan Lukavský<je...@seznam.cz> wrote:
I was primarily focused on Java SDK (and core-contruction-java), but generally
speaking, any SDK can provide default expansion that runners can use so that it
is not (should not be) required to implement this manually.
Currently, in Java SDK, the annotation is wired up into StatefulDoFnRunner,
which (as name suggests) can be used for running stateful DoFns. The problem is
that not every runner is using this facility. Java SDK generally supports
providing default expansions of transforms, but _only for transforms that do
not have to work with dynamic state_. This is not the case for this annotation
- a default implementation for @RequiresTimeSortedInput has to take another
DoFn as input, and wire its lifecycle in a way that elements are buffered in
(dynamically created) buffer and fed into the downstream DoFn only when timer
fires.
If I narrow down my line of thinking, it would be possible to:
a) create something like "dynamic pipeline expansion", which would make it
possible work with PTransforms in this way (probably would require some ByteBuddy magic)
b) wire this up to DoFnInvoker, which takes DoFn and creates class that is
used by runners for feeding data
Option b) would ensure that actually all runners support such expansion, but
seems to be somewhat hacky and too specific to this case. Moreover, it would
require knowledge if the expansion is actually required by the runner (e.g. if
the annotation is supported explicitly - most likely for batch execution).
Therefore I'd be in favor of option a), this might be reusable by a broader
range of default expansions.
In other SDKs than Java this might have different implications, the reason why
it is somewhat more complicated to do dynamic (or generic?) expansions of
PTransforms in Java is mostly due to how DoFns are implemented in terms of
annotations and the DoFnInvokers involved for efficiency.
Jan
On 1/18/24 18:35, Robert Burke wrote:
I agree that variable support across Runners does limit the adoption of a
feature. But it's also then limited if the SDKs and their local / direct
runners don't yet support the feature. The Go SDK doesn't currently have a way
of specifying that annotation, preventing use. (The lack of mention of the
Python direct runner your list implies it's not yet supported by the Python
SDK, and a quick search shows that's likely [0])
While not yet widely available to the other SDKs, Prism, the new Go SDK Local
Runner, maintains data in event time sorted heaps [1]. The intent was to
implement the annotation (among other features) once I start running the Java
and Python Validates Runner suites against it.
I think stateful transforms are getting the event ordering on values for "free"
as a result [2], but there's no special/behavior at present if the DoFn is consuming the
result of a Group By Key.
Part of the issue is that by definition, a GBK "loses" the timestamps of the
values, and doesn't emit them, outside of using them to determine the resulting timestamp
of the Key... [3]. To make use of the timestamp in the aggregation stage a runner would
need to do something different in the GBK, namely sorting by the timestamp as the data is
ingested, and keeping that timestamp around to continue the sort. This prevents a more
efficient implementation of directly arranging the received element bytes into the
Iterator format, requiring a post process filtering. Not hard, but a little dissatisfying.
Skimming through the discussion, I agree with the general utility goal of the
annotation, but as with many Beam features, there may be a discoverability
problem. The feature isn't mentioned in the Programming Guide (AFAICT), and
trying to find anything on the beam site, the top result is the Javadoc for the
annotation (which is good, but you still need to know to look for it), and then
the next time related bit is OrderedListState which doesn't yet have a
meaningful portable representation last I checked [4], once again limiting
adoption.
Probably the most critical bit is, while we have broad "handling" of the annotation, I'm hard
pressed to say we even use the annotation outside of tests. A search [5] doesn't show any
"Transforms" or "IOs" making use of it with the only markdown/documentation about it
being the Beam 2.20.0 release notes saying it's now supported in Flink and Spark [6].
I will say, this isn't grounds for removing the feature, as I can only check what's in
the repo, and not what end users have, but it does indicate we didn't drive the feature
to completion and enable user adoption beyond "This Exists, and we can tell you
about it if you ask.".
AFAICT this is just one of those features we built, but then proceeded not to
use within Beam, and evangelize. This is a point we could certainly do better
on in Beam as a whole.
Robert Burke
Beam Go Busybody
[0]https://github.com/search?q=repo%3Aapache%2Fbeam+TIME_SORTED_INPUT+language%3APython&type=code
[1]https://github.com/apache/beam/blob/master/sdks/go/pkg/beam/runners/prism/internal/engine/elementmanager.go#L93
[2]https://github.com/apache/beam/blob/master/sdks/go/pkg/beam/runners/prism/internal/engine/elementmanager.go#L1094
[3]https://github.com/apache/beam/blob/master/model/pipeline/src/main/proto/org/apache/beam/model/pipeline/v1/beam_runner_api.proto#L1132
[4]https://github.com/apache/beam/issues?q=is%3Aissue+is%3Aopen+OrderedListState
[5]https://github.com/search?q=repo%3Aapache%2Fbeam+RequiresTimeSortedInput&type=code&p=2
[6]https://github.com/apache/beam/blob/b4c23b32f2b80ce052c8a235e5064c69f37df992/website/www/site/content/en/blog/beam-2.20.0.md?plain=1#L46
On 2024/01/18 16:14:56 Jan Lukavský wrote:
Hi,
recently I came across the fact that most runners do not support
@RequiresTimeSortedInput annotation for sorting per-key data by event
timestamp [1]. Actually, runners supporting it seem to be Direct java,
Flink and Dataflow batch (as it is a noop there). The annotation has
use-cases in time-series data processing, in transaction processing and
more. Though it is absolutely possible to implement the time-sorting
manually (e.g. [2]), this is actually efficient only in streaming mode,
in batch mode the runner typically wants to leverage the internal
sort-grouping it already does.
The original idea was to implement this annotation inside
StatefulDoFnRunner, which would be used by majority of runners. It turns
out that this is not the case. The question now is, should we use an
alternative place to implement the annotation (e.g. Pipeline expansion,
or DoFnInvoker) so that more runners can benefit from it automatically
(at least for streaming case, batch case needs to be implemented
manually)? Do the community find the annotation useful? I'm linking a
rather old (and long :)) thread that preceded introduction of the
annotation [3] for more context.
I sense the current adoption of the annotation by runners makes it
somewhat use-less.
Looking forward to any comments on this.
Best,
Jan
[1]
https://beam.apache.org/releases/javadoc/2.53.0/org/apache/beam/sdk/transforms/DoFn.RequiresTimeSortedInput.html
[2]
https://cloud.google.com/spanner/docs/change-streams/use-dataflow#order-by-key
[3]https://lists.apache.org/thread/bkl9kk8l44xw2sw08s7m54k1wsc3n4tn
