shared: improve interface for more efficient reading #1
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
RaduBerinde
force-pushed
the
better-shared-iface
branch
from
d23bc21 to
376886d
Compare
keyOutOfBounds is called deep in the iterator function call stack and should be inlined.
Remove the arm-specific VFS logic, extending the previously non-arm linux file to apply to all linux builds. This separation originated in cockroachdb#171, due to the SyncFileRange syscall being undefined in arm builds, but the unix package now provides it. Fix cockroachdb#2514.
AWS (and probably some other blob storage providers) use a prefix of the object names to partition the buckets (the length of the prefix is chosen dynamically to get a desired number of partitions). We get best performance when IO load is distributed evenly across the partitions. To achieve this we prepend a hash value to the object name, consisting of 4 hex digits. The hash value is derived from the other parts of the name - the creator ID and creator file num. In this change we also improve the functions that generate names to avoid unnecessary allocations.
Remove leading zeros from the creator ID; it makes the object names unwieldy and it's not very useful (unlike filenums, the IDs aren't ordered in any significant way).
Previously, ingestions would trigger obsolete file deletions during the ingest application step. This has the potential to exacerbate the write "hiccup"/stall induced by sstable ingestion. The sequence numbers assigned to the ingestion are not published until after ingestion application. This prevents any concurrent writes that obtained later sequence numbers from publishing their own sequence numbers. When Options.Experimental.MinDeletionRate is configured, this step to delete obsolete files should have minimal impact. However, when Experimental.MinDeletionRate is not configured, the ingesting goroutine would synchronously perform file deletions, extending the length of the commit stall. Now ingestions don't trigger deletion of obsolete files. The only file deletion that an ingestion operation itself could trigger is deletion of a manifest that was made obsolete during a manifest rotation while committing the version edit. This is very rare, and delaying the deletion until a flush or compaction completes is okay.
RaduBerinde
force-pushed
the
better-shared-iface
branch
from
376886d to
6928c72
Compare
Add a new Metric recording the database's uptime since Open. This can be used in combination with other metrics to integrate over time.
Add a new Compact.Duration metric that measures the cumulative time spent in compactions since Open. This may be used in combination with Metrics.Uptime to compute the effective compaction concurrency (cockroachdb#1934). Callers may compute effective compaction concurrency over an interval by measuring both Uptime and Compact.Duration at the beginning and end of the interval and subtracting.
When I extracted out the readahead logic from the sstable Reader, I accidentally omitted some initialization code - setting `readaheadState.size` to `initialReadaheadSize`. Fixing.
Remove the `genericFileReadable` implementation; since we moved `Prealloc` to be a method on `File`, the regular implementation should always work. Using the same implementation for all cases helps testing - e.g. we can test readahead logic with files in MemFS.
This change improves the objstorage test to allow reading arbitrary parts of objects. We add a test that verifies that the read-ahead implementation does the Prefetch and the reopen that is expected. We also improve logging FS to log `ReadAt` and `Prefetch` calls. Informs cockroachdb#2531.
The current `shared.Storage` interface forces the implementation to perform a metadata operation to look up the object on each read. Some experimental backends have a separate internal open step that doesn't need to be performed every time. This commit improves the interface to allow opening the object once and then performing many reads.
RaduBerinde
force-pushed
the
better-shared-iface
branch
from
6928c72 to
52aefa8
Compare
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.
The current
shared.Storageinterface forces the implementation to perform a metadata operation to look up the object on each read.This commit improves the interface to allow opening the object once and then performing many reads.