Uh oh!
There was an error while loading. Please reload this page.
HBASE-27795: Define RPC API for cache cleaning - #5492
Closed
Kota-SH wants to merge 1 commit into
Closed
Conversation
Apache-HBase
commented
Nov 2, 2023
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 2, 2023
💔 -1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 2, 2023
🎊 +1 overall
This message was automatically generated. |
wchevreuil
requested changes
Nov 2, 2023
wchevreuil
left a comment
Contributor
There was a problem hiding this comment.
Thanks for this initial work, @Kota-SH ! I have a few remarks, though. Please let me know if those make sense to you.
Comment on lines
+2637
to
+2661
| /** | ||
| * Clean BucketCache | ||
| */ |
Contributor
There was a problem hiding this comment.
Nit: Let's be more specific. What does it mean to clean BucketCache? We are evicting everything? And why this is only for BucketCache, not for all BlockCache implementations?
| } | ||
| message CleanBucketCacheResponse { | ||
| repeated string uncached_files = 1; |
Contributor
There was a problem hiding this comment.
Could we return a Map<uncached_files, num_blocks_evicted> ?
| return Optional.empty(); | ||
| } | ||
| default List<HStoreFile> getCachedButClosedFiles() { |
Contributor
There was a problem hiding this comment.
I like the idea of exposing this info.
And let's use Optional to avoid returning null anti-pattern, just like the getFullyCachedFiles method above.
Also the naming is not very intuitive. We could call it getFilesWithStaleBlocks and let's put an explanatory javadoc.
Additionally, BlockCache interface should defined the clean method as well. It's up for the BlockCache implementation to deal with the eviction logic.
| } | ||
| @Override | ||
| public CleanBucketCacheResponse cleanBucketCache(RpcController controller, |
Contributor
There was a problem hiding this comment.
I'm confused why we are implementing the eviction logic here. The RSRpcServices should be just a facade, finding "orphans" blocks and evict it should be a responsibility of the cache implementation class.
Apache-HBase
commented
Nov 14, 2023
💔 -1 overall
This message was automatically generated. |
1 similar comment
Apache-HBase
commented
Nov 14, 2023
💔 -1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 14, 2023
💔 -1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 14, 2023
💔 -1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 15, 2023
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 15, 2023
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 15, 2023
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 15, 2023
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 15, 2023
💔 -1 overall
This message was automatically generated. |
Uh oh!
There was an error while loading. Please reload this page.
| @Override | ||
| public Optional<Map<String, Integer>> uncacheStaleBlocks() { | ||
| return l2Cache.uncacheStaleBlocks(); |
Contributor
There was a problem hiding this comment.
We should do this to L1 as well.
| public void setFilesWithStaleBlocks(HRegion hRegion) { | ||
| for (HStore hs : hRegion.getStores()) { | ||
| filesWithStaleBlocks.addAll(hs.getStorefiles()); |
Contributor
There was a problem hiding this comment.
Shouldn't we check first which files from the region's store have blocks in the cache? If the total cache size is smaller than the store file size, we might not be able to cache all files.
Also, I wonder what's the heap usage impact of keeping such extra list. Could you check that with a heap dump?
Apache-HBase
commented
Nov 22, 2023
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 22, 2023
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 22, 2023
💔 -1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 29, 2023
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 29, 2023
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Nov 29, 2023
🎊 +1 overall
This message was automatically generated. |
Kota-SH
commented
Dec 1, 2023
ContributorAuthor
@wchevreuil - Could you please review this updated patch and provide feedback? Thanks, |
Apache-HBase
commented
Dec 1, 2023
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Dec 1, 2023
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Dec 1, 2023
🎊 +1 overall
This message was automatically generated. |
| * @return A map of filename and number of blocks evicted. | ||
| */ | ||
| default Optional<Map<String, Integer>> | ||
| uncacheStaleBlocks(RegionAvailabilityChecker regionAvailabilityChecker) { |
Contributor
There was a problem hiding this comment.
I think @Apache9 meant to use the RegionServerServices interface in the method signature, not creating a new interface.
| Map<String, Integer> uncachedStaleBlocksMap = | ||
| l1Cache.uncacheStaleBlocks(regionAvailabilityChecker).orElseGet(HashMap::new); | ||
| l2Cache.uncacheStaleBlocks(regionAvailabilityChecker).ifPresent( | ||
| map2 -> map2.forEach((key, value) -> uncachedStaleBlocksMap.merge(key, value, Integer::sum))); |
Contributor
There was a problem hiding this comment.
Why merge? We should just do putAll, no?
| public Optional<Map<String, Integer>> | ||
| uncacheStaleBlocks(RegionAvailabilityChecker regionAvailabilityChecker) { | ||
| Map<String, Integer> evictedFilesWithStaleBlocks = new HashMap<>(); |
Contributor
There was a problem hiding this comment.
Copying my previous concerns shared on the other PR:
Hum, I think we should make this method async. And we shouldn't block the RPC call until its finished. We should dedicate a thread pool with a max size of one.
I'm worried now this might not be a lightweight operation, if the related RPC is synchronous (as implemented here), we might eventually timeout and client might submit another call, eventually exhausting RPC handlers.
Simply making the RPC async isn't enough, we should also make sure we don't have more than one background thread running this.`
Apache-HBase
commented
Feb 6, 2024
💔 -1 overall
This message was automatically generated. |
Apache-HBase
commented
Feb 6, 2024
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Feb 6, 2024
💔 -1 overall
This message was automatically generated. |
| this.getBlockCache().flatMap(BlockCache::getFullyCachedFiles).orElse(Collections.emptyMap()); | ||
| Map<String, Integer> evictedFilesWithStaleBlocks = new ConcurrentHashMap<>(); | ||
| ExecutorService executor = Executors.newFixedThreadPool(6); |
Contributor
There was a problem hiding this comment.
The threadpool should be global for HRegionServer. We don't want a new threadpool created on every client call to this method, that would potentially kill the RS if an operator calls it multiple times on a short period.
Comment on lines
+1675
to
+1687
| Callable<Void> task = () -> { | ||
| HRegion regionOnServer = getRegion(value.getFirst()); | ||
| int blocksEvicted = (regionOnServer == null || !regionOnServer.isAvailable()) | ||
| ? this.getBlockCache().get().evictBlocksByHfileName(fileName) | ||
| : 0; | ||
| evictedFilesWithStaleBlocks.put(fileName, blocksEvicted); | ||
| LOG.info( | ||
| "Uncached {} blocks belonging to the file {} as the region {} " | ||
| + "is not served by the region server {} anymore.", | ||
| blocksEvicted, fileName, value.getFirst(), this.getServerName()); | ||
| return null; | ||
| }; | ||
| tasks.add(task); |
Contributor
There was a problem hiding this comment.
We should do a single task per call to the method, not a task per file, that would create yet another collection in memory with as many objects as the total files cached. On large caches, that would be impacting.
| } finally { | ||
| executor.shutdown(); | ||
| } | ||
| return evictedFilesWithStaleBlocks; |
Contributor
There was a problem hiding this comment.
We should think about an alternative result here, as this map is effectively being updated by the background task, which can be confusing for the caller. I would rather make this as void, or a boolean, returning true indicating that the task is submitted and is running in the background.
Apache-HBase
commented
Feb 6, 2024
🎊 +1 overall
This message was automatically generated. |
Apache-HBase
commented
Feb 7, 2024
💔 -1 overall
This message was automatically generated. |
Apache-HBase
commented
Feb 7, 2024
💔 -1 overall
This message was automatically generated. |
wchevreuil
commented
Dec 17, 2024
Contributor
Closing this as it has been deprecated by HBASE-28923. |
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
Sign up for freeto join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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.
Jira: https://issues.apache.org/jira/browse/HBASE-27795