pq: reduce read contention when caught up #17765
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
By keeping track of the lock-holding reader's demand (a quantity of events and a deadline), we can avoid waking up the reader for each event that is written to the queue, significantly reducing the time that the lock-holding _writer_ is blocked by synchronous read operations. The lock-holding reader is awoken when its demand is met, has expired, or when the writer has rotated pages. | batch size | unpatched | this patch | | ---------- | --------- | ---------- | | 125 | 260k | 498k | | 2000 | 219k | 516k |
🤖 GitHub commentsExpand to view the GitHub comments
Just comment with:
|
|
This pull request does not have a backport label. Could you fix it @yaauie? 🙏
|
kaisecheng
reviewed
Jul 2, 2025
Comment on lines
+935
to
+938
| this.readDemand = new ReadDemand(deadlineMillis, elementsNeeded); | ||
|
|
||
| boolean unElapsed = this.notEmpty.awaitUntil(new Date(deadlineMillis)); | ||
| this.readDemand = null; |
There was a problem hiding this comment.
I don't have big concern but wanna clarify the behaviour.
Reader releases the lock when notEmpty.awaitUntil() is called and do the waiting. Potentially, other readers could acquire the lock and update this.readDemand to a further delay deadlineMillis. Updating this.readDemand is not thread-safe and I believe it is not required to be. The worst case is the deadline extends to (timeout 50ms x number of threads).
Is my understanding correct? If so, can we add a comment to remind that the action is not atomic?
There was a problem hiding this comment.
Other readers cannot acquire the lock because Queue#readSerializedBatch and Queue#nonBlockReadBatch (the only places where a reader can acquire the queue's lock) are synchronized.
[EDIT: I believe that Queue#nonBlockReadBatch is only referenced from tests; I cannot find other references to this, or to non_block_read_batch that could be called from the ruby-side.]
|
💚 Build Succeeded
History
|
yaauie
added
backport-9.1
mergify bot
pushed a commit
that referenced
this pull request
Jul 10, 2025
* pq: reduce read contention when caught up By keeping track of the lock-holding reader's demand (a quantity of events and a deadline), we can avoid waking up the reader for each event that is written to the queue, significantly reducing the time that the lock-holding _writer_ is blocked by synchronous read operations. The lock-holding reader is awoken when its demand is met, has expired, or when the writer has rotated pages. | batch size | unpatched | this patch | | ---------- | --------- | ---------- | | 125 | 260k | 498k | | 2000 | 219k | 516k | * remove superfluous volatile flag (cherry picked from commit ddd519c)
2 tasks
mergify bot
pushed a commit
that referenced
this pull request
Jul 10, 2025
* pq: reduce read contention when caught up By keeping track of the lock-holding reader's demand (a quantity of events and a deadline), we can avoid waking up the reader for each event that is written to the queue, significantly reducing the time that the lock-holding _writer_ is blocked by synchronous read operations. The lock-holding reader is awoken when its demand is met, has expired, or when the writer has rotated pages. | batch size | unpatched | this patch | | ---------- | --------- | ---------- | | 125 | 260k | 498k | | 2000 | 219k | 516k | * remove superfluous volatile flag (cherry picked from commit ddd519c)
mergify bot
pushed a commit
that referenced
this pull request
Jul 10, 2025
* pq: reduce read contention when caught up By keeping track of the lock-holding reader's demand (a quantity of events and a deadline), we can avoid waking up the reader for each event that is written to the queue, significantly reducing the time that the lock-holding _writer_ is blocked by synchronous read operations. The lock-holding reader is awoken when its demand is met, has expired, or when the writer has rotated pages. | batch size | unpatched | this patch | | ---------- | --------- | ---------- | | 125 | 260k | 498k | | 2000 | 219k | 516k | * remove superfluous volatile flag (cherry picked from commit ddd519c)
This was referenced Jul 10, 2025
mergify bot
pushed a commit
that referenced
this pull request
Jul 10, 2025
* pq: reduce read contention when caught up By keeping track of the lock-holding reader's demand (a quantity of events and a deadline), we can avoid waking up the reader for each event that is written to the queue, significantly reducing the time that the lock-holding _writer_ is blocked by synchronous read operations. The lock-holding reader is awoken when its demand is met, has expired, or when the writer has rotated pages. | batch size | unpatched | this patch | | ---------- | --------- | ---------- | | 125 | 260k | 498k | | 2000 | 219k | 516k | * remove superfluous volatile flag (cherry picked from commit ddd519c)
2 tasks
mergify bot
pushed a commit
that referenced
this pull request
Jul 10, 2025
* pq: reduce read contention when caught up By keeping track of the lock-holding reader's demand (a quantity of events and a deadline), we can avoid waking up the reader for each event that is written to the queue, significantly reducing the time that the lock-holding _writer_ is blocked by synchronous read operations. The lock-holding reader is awoken when its demand is met, has expired, or when the writer has rotated pages. | batch size | unpatched | this patch | | ---------- | --------- | ---------- | | 125 | 260k | 498k | | 2000 | 219k | 516k | * remove superfluous volatile flag (cherry picked from commit ddd519c)
2 tasks
yaauie
added a commit
that referenced
this pull request
Jul 10, 2025
* pq: reduce read contention when caught up By keeping track of the lock-holding reader's demand (a quantity of events and a deadline), we can avoid waking up the reader for each event that is written to the queue, significantly reducing the time that the lock-holding _writer_ is blocked by synchronous read operations. The lock-holding reader is awoken when its demand is met, has expired, or when the writer has rotated pages. | batch size | unpatched | this patch | | ---------- | --------- | ---------- | | 125 | 260k | 498k | | 2000 | 219k | 516k | * remove superfluous volatile flag (cherry picked from commit ddd519c) Co-authored-by: Rye Biesemeyer <[email protected]>
yaauie
added a commit
that referenced
this pull request
Jul 10, 2025
* pq: reduce read contention when caught up By keeping track of the lock-holding reader's demand (a quantity of events and a deadline), we can avoid waking up the reader for each event that is written to the queue, significantly reducing the time that the lock-holding _writer_ is blocked by synchronous read operations. The lock-holding reader is awoken when its demand is met, has expired, or when the writer has rotated pages. | batch size | unpatched | this patch | | ---------- | --------- | ---------- | | 125 | 260k | 498k | | 2000 | 219k | 516k | * remove superfluous volatile flag (cherry picked from commit ddd519c) Co-authored-by: Rye Biesemeyer <[email protected]>
yaauie
added a commit
that referenced
this pull request
Jul 10, 2025
* pq: reduce read contention when caught up By keeping track of the lock-holding reader's demand (a quantity of events and a deadline), we can avoid waking up the reader for each event that is written to the queue, significantly reducing the time that the lock-holding _writer_ is blocked by synchronous read operations. The lock-holding reader is awoken when its demand is met, has expired, or when the writer has rotated pages. | batch size | unpatched | this patch | | ---------- | --------- | ---------- | | 125 | 260k | 498k | | 2000 | 219k | 516k | * remove superfluous volatile flag (cherry picked from commit ddd519c) Co-authored-by: Rye Biesemeyer <[email protected]>
yaauie
added a commit
that referenced
this pull request
Jul 10, 2025
* pq: reduce read contention when caught up By keeping track of the lock-holding reader's demand (a quantity of events and a deadline), we can avoid waking up the reader for each event that is written to the queue, significantly reducing the time that the lock-holding _writer_ is blocked by synchronous read operations. The lock-holding reader is awoken when its demand is met, has expired, or when the writer has rotated pages. | batch size | unpatched | this patch | | ---------- | --------- | ---------- | | 125 | 260k | 498k | | 2000 | 219k | 516k | * remove superfluous volatile flag (cherry picked from commit ddd519c) Co-authored-by: Rye Biesemeyer <[email protected]>
yaauie
added a commit
that referenced
this pull request
Jul 10, 2025
* pq: reduce read contention when caught up By keeping track of the lock-holding reader's demand (a quantity of events and a deadline), we can avoid waking up the reader for each event that is written to the queue, significantly reducing the time that the lock-holding _writer_ is blocked by synchronous read operations. The lock-holding reader is awoken when its demand is met, has expired, or when the writer has rotated pages. | batch size | unpatched | this patch | | ---------- | --------- | ---------- | | 125 | 260k | 498k | | 2000 | 219k | 516k | * remove superfluous volatile flag (cherry picked from commit ddd519c) Co-authored-by: Rye Biesemeyer <[email protected]>
1 task
3 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Release notes
What does this PR do?
Eliminates per-event overhead from workers consuming from a persisted queue.
By keeping track of the lock-holding reader's demand (a quantity of events and a deadline), we can avoid waking up the reader for each event that is written to the queue, significantly reducing the time that the lock-holding writer is blocked by synchronous read operations. The lock-holding reader is awoken when its demand is met, has expired, or when the writer has rotated pages.
Why is it important/What is the impact to the user?
Significantly improves PQ throughput.
From my local tests (ARM / M3 Pro 12 CPU / SSD) with
input { java_generator {} } output { sink {} }:Checklist
[ ] I have made corresponding changes to the documentation[ ] I have made corresponding change to the default configuration files (and/or docker env variables)[ ] I have added tests that prove my fix is effective or that my feature worksAuthor's Checklist
How to test this PR locally
Related issues