perf: allow concurrent decompress away from network loop #162
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
We want to ensure the transcode passes everything that is expected to be run asynchronously over to the decode loop. In general, memcached calls us back with gotData, then receivedStatus, then complete. We use gotData to launch the work onto the transcode threadpool and return control to memcached, which would immediately then call receivedStatus. Previously, receivedStatus and complete were set up to interact and set a value on the underlying future but only by synchronously blocking for the transcode future. Given this callback is happening nearly immediately after the gotData callback, we were firing the transcode and nearly always performing a blocking get, which triggers a synchronous decompression on the network IO loop. This is of course very detrimental to evcache driver performance, since the driver cannot even accept new requests to issue to them to memcached backends, and must wait until decompression completes. In this fix, we rearrange things a little to ensure that if the async decode is requested, that we push the completion status updates to happen only after the async decode completes. This is a little ugly because of the current arrangement of the memcached decoder. A future change might be to overhaul this integration and pull it out of the memcached transcode framework and use something a bit more friendly.
|
This change is tricky, actually, since it slightly changes timing it makes the system more scalable without tuning but a little less efficient. It needs some consideration. Maybe an alternative is to always do it in-thread and encourage users to tune poolSize much higher. |
pkarumanchi9
reviewed
Feb 5, 2025
|
|
||
| if (isWrongKeyReturned(key, k)) return; | ||
| boolean shouldLog = log.isDebugEnabled() && client.getPool().getEVCacheClientPoolManager().shouldLog(appName); | ||
| boolean alwaysSync = false; |
There was a problem hiding this comment.
should be exposed via a fp to go via async path?
pkarumanchi9
reviewed
Feb 5, 2025
| dataSizeDS.record(data.length); | ||
| Transcoder<T> transcoder = (tc == null) ? (Transcoder<T>) getTranscoder() : tc; | ||
| if (tcService == null) { | ||
| log.error("tcService is null, will not be able to decode"); |
There was a problem hiding this comment.
should we do latch.countDown() here?
There was a problem hiding this comment.
fixed in 2b60f17, I restructured it too. Hopefully reduce chance of bugs like this.
…ecode Previously had a concurrency issue and may have dropped some latch countDown. A refactor to track state differently and tidy up a little helps. Add a property to control sync decode threading behavior
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
We want to ensure the transcode passes everything that is expected to be run asynchronously over to the decode loop. In general, memcached calls us back with gotData, then receivedStatus, then complete. We use gotData to launch the work onto the transcode threadpool and return control to memcached, which would immediately then call receivedStatus.
Previously, receivedStatus and complete were set up to interact and set a value on the underlying future but only by synchronously blocking for the transcode future. Given this callback is happening nearly immediately after the gotData callback, we were firing the transcode and nearly always performing a blocking get, which triggers a synchronous decompression on the network IO loop. This is of course very detrimental to evcache driver performance, since the driver cannot even accept new requests to issue to them to memcached backends, and must wait until decompression completes.
In this fix, we rearrange things a little to ensure that if the async decode is requested, that we push the completion status updates to happen only after the async decode completes. This is a little ugly because of the current arrangement of the memcached decoder. A future change might be to overhaul this integration and pull it out of the memcached transcode framework and use something a bit more friendly.