Move caching of the store to IndexShard. #30817
Closed
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
In spite of the existing caching, I have seen a number of nodes hot threads where one thread had been spending all its cpu on computing the size of a directory. I am proposing to move the caching of the store size to `IndexShard` so that it has access to the existing logic regarding whether a shard is active or not in order to be able to cache the store size more agressively. The tricky bit is that an inactive shard might still be merged, which may have a significant impact on the store size. This should be especially useful for time-based data since most indices are typically inactive.
jpountz
added
>enhancement
WIP
:Distributed Indexing/Store
Collaborator
|
Pinging @elastic/es-distributed |
Contributor
Author
|
WIP because of the lack of test. |
s1monw
reviewed
May 24, 2018
Contributor
s1monw
left a comment
s1monw
left a comment
There was a problem hiding this comment.
I like that this is much simpler yet, I think it's incorrect or has too many corner cases. like if a reader gets closed we don't refresh the stats since we don't see the deletes. It's not good enough. sorry for pushing down that route.
| private final CounterMetric totalMergesNumDocs = new CounterMetric(); | ||
| private final CounterMetric totalMergesSizeInBytes = new CounterMetric(); | ||
| private final CounterMetric currentMerges = new CounterMetric(); | ||
| private final AtomicLong currentMerges = new AtomicLong(); |
| private final Set<OnGoingMerge> onGoingMerges = ConcurrentCollections.newConcurrentSet(); | ||
| private final Set<OnGoingMerge> readOnlyOnGoingMerges = Collections.unmodifiableSet(onGoingMerges); | ||
| private final MergeSchedulerConfig config; | ||
| private volatile long lastMergeMillis; |
Contributor
There was a problem hiding this comment.
I don't think we need these changes. I guess it would be enough to check Engine#getMergeStats() and then do:
mergeStats.current > 0 || mergeStats.total != previousStats.total?
| } | ||
|
|
||
| @Override | ||
| protected boolean needsRefresh() { |
Contributor
There was a problem hiding this comment.
I am not sure why this is so complex, wouldn't it be enough to override needsRefresh()?
private MergeStats previousStats = new MergeStats();
if (super.needsRefresh()) {
boolean refresh = false;
if (isActive()) {
MergeStats mergeStats = getEngine().getMergeStats();
refresh = mergeStats.current > 0 || mergeStats.total != previousStats.total;
previousStats = mergeStats;
}
return refresh;
}| if (active.get() == false) { | ||
| // We refresh when transitioning to an inactive state to make | ||
| // it easier to cache the store size. | ||
| refresh("transition to inactive"); |
Contributor
There was a problem hiding this comment.
-1 we should not refresh anything except of the internal reader. Visibility guarantees are important.
Contributor
Author
|
Closed in favor of the original PR. |
jpountz
removed
:Distributed Indexing/Store
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.
In spite of the existing caching, I have seen a number of nodes hot threads
where one thread had been spending all its cpu on computing the size of a
directory. I am proposing to move the caching of the store size to
IndexShardso that it has access to the existing logic regarding whether a shard is active
or not in order to be able to cache the store size more agressively.
The tricky bit is that an inactive shard might still be merged, which may have
a significant impact on the store size.
This should be especially useful for time-based data since most indices are
typically inactive.