Prevent concurrent access to local breaker in rerank #128162
Merged
Conversation
This file contains hidden or 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
|
Pinging @elastic/es-analytical-engine (Team:Analytics) |
elasticsearchmachine
added
the
Team:Analytics
dnhatn
force-pushed
the
fix-rerank-operator
branch
from
cac99e8 to
1840e7f
Compare
dnhatn
force-pushed
the
fix-rerank-operator
branch
from
1840e7f to
d89c1c5
Compare
ivancea
approved these changes
May 21, 2025
| inferenceRunner.doInference( | ||
| buildInferenceRequest(inputPage), | ||
| ActionListener.wrap( | ||
| inferenceResponse -> outputListener.onResponse(buildOutput(inputPage, inferenceResponse)), |
There was a problem hiding this comment.
If it isn't safe to create blocks here (And therefore update the CB), why is it safe to read a block from transport like we do in LookupFromIndexOperator? I was wondering which rule we should follow in the AsyncOperators to avoid having this issue again
There was a problem hiding this comment.
- We use the global breaker not local breaker for reading blocks from an input stream.
- I have added assertions in ce54583, which should consistently detect such misuse.
nik9000
reviewed
May 21, 2025
| BlockFactory blockFactory = blockFactory(); | ||
| final int overReservedBytes = between(0, 1024 * 1024); | ||
| final int maxOverReservedBytes = between(overReservedBytes, 1024 * 1024); | ||
| var localBreaker = new LocalCircuitBreaker(blockFactory.breaker(), overReservedBytes, maxOverReservedBytes); |
There was a problem hiding this comment.
Should we wrap this in something that asserts that we're on the same thread every time?
17 tasks
afoucret
pushed a commit
to afoucret/elasticsearch
that referenced
this pull request
Jun 4, 2025
When an async operator receives a response, we can't create new blocks on the responding thread because multiple threads may adjust the local breaker simultaneously, leading to a data race. To address this, we can either use the global breaker or delay block creation in getOutput. While using the global block factory is simpler, I prefer the second option to use the local breaker when possible. Therefore, I opted to keep the results in the queue and create new blocks in getOutput. Our tests didn't catch this issue because: (1) only one block is created in the test, and (2) there is no delay when simulating the inference service. Closes elastic#127638 Closes elastic#127051
dnhatn
added a commit
to dnhatn/elasticsearch
that referenced
this pull request
Jun 5, 2025
When an async operator receives a response, we can't create new blocks on the responding thread because multiple threads may adjust the local breaker simultaneously, leading to a data race. To address this, we can either use the global breaker or delay block creation in getOutput. While using the global block factory is simpler, I prefer the second option to use the local breaker when possible. Therefore, I opted to keep the results in the queue and create new blocks in getOutput. Our tests didn't catch this issue because: (1) only one block is created in the test, and (2) there is no delay when simulating the inference service. Closes elastic#127638 Closes elastic#127051
dnhatn
added a commit
that referenced
this pull request
Jun 5, 2025
When an async operator receives a response, we can't create new blocks on the responding thread because multiple threads may adjust the local breaker simultaneously, leading to a data race. To address this, we can either use the global breaker or delay block creation in getOutput. While using the global block factory is simpler, I prefer the second option to use the local breaker when possible. Therefore, I opted to keep the results in the queue and create new blocks in getOutput. Our tests didn't catch this issue because: (1) only one block is created in the test, and (2) there is no delay when simulating the inference service. Closes #127638 Closes #127051
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.
When an async operator receives a response, we can't create new blocks on the responding thread because multiple threads may adjust the local breaker simultaneously, leading to a data race. To address this, we can either use the global breaker or delay block creation in getOutput. While using the global block factory is simpler, I prefer the second option to use the local breaker when possible. Therefore, I opted to keep the results in the queue and create new blocks in getOutput. Our tests didn't catch this issue because: (1) only one block is created in the test, and (2) there is no delay when simulating the inference service.
Closes #127638
Closes #127051