Poll for backlog in background thread instead of inline #31697
Conversation
Naireen
force-pushed
the
backlog_async
branch
6 times, most recently
from
e7d7074
to
c6221ff
Compare
Naireen
force-pushed
the
backlog_async
branch
6 times, most recently
from
0413d14
to
1127404
Compare
|
R: @scwhittle |
|
Assigning reviewers. If you would like to opt out of this review, comment R: @kennknowles for label java. Available commands:
The PR bot will only process comments in the main thread (not review comments). |
|
Stopping reviewer notifications for this pull request: review requested by someone other than the bot, ceding control. If you'd like to restart, comment |
| return offsetConsumer.position(topicPartition); | ||
| } | ||
| }, | ||
| 1, |
There was a problem hiding this comment.
make this a constant for now so it's more readable/discoverable?
we eventually might want to make it an option or increase it by default to reduce polling.
| @@ -148,6 +150,12 @@ | |||
| abstract class ReadFromKafkaDoFn<K, V> | |||
| extends DoFn<KafkaSourceDescriptor, KV<KafkaSourceDescriptor, KafkaRecord<K, V>>> { | |||
|
|
|||
| private static final int OFFSET_UPDATE_INTERVAL_SECONDS = 1; | |||
|
|
|||
| private transient ScheduledExecutorService backlogFetcherThread = | |||
There was a problem hiding this comment.
this is per-dofn, it seems like it should be some static cache across all of the dofns.
You could see what I was doing for the caching the background reading threads in https://github.com/apache/beam/pull/31786/files#diff-0f4b915b82782e618addba2e443f07b3d4b86ad8b77c2bf7adeb0ad1d6864adb
There was a problem hiding this comment.
Note that this is a little different than the UnboundedReader where there is a single reader instance for each partition+topic and the caching across dofns is managed by the runner with the ReaderCache. Since these are just normal dofns, there may be many different DoFn instances (up to 1-per processing thread) and the same partition+topic may bounce between using different dofn instances.
| @@ -277,7 +292,9 @@ protected void finalize() { | |||
|
|
|||
| @Override | |||
| public long estimate() { | |||
| return memoizedBacklog.get(); | |||
| memoizedBacklog = Preconditions.checkStateNotNull(backlogMap.get(topicPartition)); | |||
| Long backlogValue = memoizedBacklog.get(); | |||
There was a problem hiding this comment.
AFAICT this estimate is also called from the main processing thread and thus may still block issuing the backlog request.
Instead we may want to have estimate be non blocking and just return whatever the latest observed offset was, and separately change the background fetching thread to force refreshes of all the backlogs.
Naireen
force-pushed
the
backlog_async
branch
2 times, most recently
from
c63defc
to
51e4854
Compare
move momization to background thread
Naireen
force-pushed
the
backlog_async
branch
from
99a453e
to
6451e75
Compare
Poll for backlog in background thread instead of inline
Thank you for your contribution! Follow this checklist to help us incorporate your contribution quickly and easily:
addresses #123), if applicable. This will automatically add a link to the pull request in the issue. If you would like the issue to automatically close on merging the pull request, commentfixes #<ISSUE NUMBER>instead.CHANGES.mdwith noteworthy changes.See the Contributor Guide for more tips on how to make review process smoother.
To check the build health, please visit https://github.com/apache/beam/blob/master/.test-infra/BUILD_STATUS.md
GitHub Actions Tests Status (on master branch)
See CI.md for more information about GitHub Actions CI or the workflows README to see a list of phrases to trigger workflows.