release-19.2: batcheval: use write-batches to apply small SSTables #41768
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
|
This change is |
dt
force-pushed
the
backport19.2-41705
branch
from
e814ee3
to
7db6142
Compare
Adding very small SSTables is not great: the benefits of adding the raw SST, in terms of avoiding per-key overhead, are small for small SSTs, while the fixed costs of adding a file are the same — triggering flushes, adding to number of files that need to be compacted, etc — so when files become too small, they can actually be more expensive, per key, than just writing their contents via the regular write path. This patch gives the caller control of how SSTs are ingested — either via the usual direct AddFile or by constructing a normal write-batch instead - in the existing AddSSTable API. An alternative approach would be to have the caller choose a different API if they don't want to add an actual SSTable e.g. switch to use Put or WriteBatch. However the AddSSTable method does much more than just "add an SSTable" and much of the ingestion pipeline is built around those specific semantics: ingesting keys with arbitrary timestamps, key-by-key collision detection, cheaper MVCC stats support, etc. Switching between API methods would mean duplicating some of those semantics into other methods, keeping them in sync, and asking the client to keep track of what is supported by which methods. Instead, allowing the the existing method simply change how it writes its result lets small batches continue to use it, just without being forced to then incur the cost of writing a small file. The result is we have just one bulk-ingest KV API, which provides the specific semantics the bulk-ingestion pipeline needs, regardless of how it writes its result to the storage engine. Requests marked for write-based ingestion can additionally skip the back-pressure mechanisms added for SSTable additions — since we’re not creating files directly, the normal write back pressure still applies, and these requests also don’t risk of bloating the file count limits with tiny files. Release note (performance improvement): bulk-ingestion
dt
force-pushed
the
backport19.2-41705
branch
from
7db6142
to
f938107
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.
Backport 1/1 commits from #41705.
/cc @cockroachdb/release
Adding very small SSTables is not great: the benefits of adding the raw SST, in terms of avoiding per-key overhead, are small for small SSTs, while the fixed costs of adding a file are the same — triggering flushes, adding to number of files that need to be compacted, etc — so when files become too small, they can actually be more expensive, per key, than just writing their contents via the regular write path.
This patch gives the caller control of how SSTs are ingested — either via the usual direct AddFile or by constructing a normal write-batch instead.
Requests marked for write-based ingestion can additionally skip the back-pressure mechanisms added for SSTable additions — since we’re not creating files directly, the normal write back pressure still applies, and these requests also don’t risk of bloating the file count limits with tiny files.
Release note: none.