Improving file eviction performance #696
Draft
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
…erloaded We have been using a file removal semaphore with weight 5,000 (half of Go's default 10,000 maximum OS threads, beyond which Go will crash), in an attempt to avoid crashing when the filesystem/storage layer can't keep up with our requirements. This change renames that semaphore to `diskWaitSem` and also uses it for disk-write operations. When the semaphore cannot be acquired for disk-writes, we return HTTP 503 (service unavailable) or gRPC RESOURCE_EXHAUSTED error codes to the client. Relates to buchgr#638
This commit: - Performs evictions from a single background goroutine that receives files to be removed via a channel. - Throttles number of concurrent Put requests with semaphore (but not rejecting them). In order to: - Avoid crashing on high load. - Achieve up to 3 times faster cache eviction. - Achieve up to 70% higher write throughput in scenario with many cache evictions. The cache can grow above max_size when asynchronous files removals do not catch up with new file writes. This is addressed in the following part 2 commit. This issue was masqueraded in previous bazel-remote versions by instead running out of operating system threads and crash. Change-Id: Ifa2ed6c5a093adbb407750a0d38a4181a07f227f
Introduce a disk_size_limit for the total disk space of: - Files currently in the cache. - Reserved space for files currently being uploaded. - Evicted files not yet removed. Setting this limit is optional (at least for now). Reservations for Put requests are rejected when disk_size_limit is exceeded. The prometheus gauge bazel_remote_disk_cache_size_bytes is updated to be a max value for the previous 30 seconds, in order to be aware of short spikes when tuning the disk_size_limit configuration. There is also a new prometheus gauge bazel_remote_disk_cache_size_bytes_limit showing current configured limits in order to help visualize if current size is getting close to the limit and help tuning the disk_size_limit. Change-Id: Iaec29af9a2e02796c29f294b993989783d575c4b
Use access logger instead of error logger when requests are rejected due to overload, in order to avoid too verbose error log when many requests are rejected. Number of rejects can also be monitored via codes in the prometheus metrics http_request_duration_seconds_count and grpc_server_handled_total Change-Id: I5d5999360b3e49b153fd6f122e2244d4789cf2ff
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.
No description provided.