libroach: disable rocksdb subcompactions #35210
Merged
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
The feature is not widely used so is not the most stable, especially when combined with other uncommonly used features like range tombstones. In my `bin/workload kv --read-percent 0` experiments it is not necessary since the frequent foreground `fdatasync`s prevent compaction from falling behind, with or without subcompactions enabled. Release note: None
|
This change is |
petermattis
approved these changes
Feb 26, 2019
complete! 1 of 0 LGTMs obtained
|
bors r+ |
craig bot
pushed a commit
that referenced
this pull request
Feb 26, 2019
35157: sql: Add "or-if-not-exists" semantics to `CPut` and use it in index entry writes r=bdarnell a=dt This switches index updates to use a new `CPut` behavior, to fix the handling of valid update operations that are processed during the backfill for indexes that have values -- such as UNIQUE indexes as well as non-unique indexes that include STORING columns. We need to use `CPut` for maintaining index entries (at least for UNIQUE indexes) to avoid clobbering entries for other rows. During the backfill we cannot assume an entry with the old value actually exists however. Thus a normal `CPut` on them from the prior value to the new value also asserts they exist, which isn't always true. This issue is a little hard to hit: we only emit a `CPut` from the old value to the new value when a) the key is the same and b) the values are different. If any indexed values changed, we'd be deleting the old entry and `CPut`ting a new one which we (correctly) expect not to exist at all. Only when updating the same key to a new value -- i.e. when handing an update to only a STORING column or some UNIQUE cases -- do we do this in-place CPut from old to new and then only when we happen to be operating on keys not yet reached by the backfiller do we incorrectly assume it exists with value old. Since we put many CPuts in the batch at once, we can't easily retry in SQL with nil vs old values so we need new CPut semantics to fix this. This adds a flag to `CPut` to optionally allow a put which expects a given existing value to succeed if instead that key does not exist. This is useful in sitations where the client is trying to avoid overwriting an existing entry other than the one it thinks should exist, but also wants to allow for it not to exist. Specifically, when writing secondary index entries, particularly for unique indexes, we use `CPut` to ensure that we're not overwriting the index entry for a different row -- if we are, we expect the `CPut` to fail. Fixes #33260. Release note (bug fix): Fix bug when that would return errors when handling valid UPDATEs during periods of an index creation schema change. 35210: libroach: disable rocksdb subcompactions r=ajkr a=ajkr The feature is not widely used so is not the most stable, especially when combined with other uncommonly used features like range tombstones. In my `bin/workload kv --read-percent 0` experiments it is not necessary since the frequent foreground `fdatasync`s prevent compaction from falling behind, with or without subcompactions enabled. Release note: None 35213: Makefile: don't build krb5 on musl r=mjibson a=mjibson It produces a linking error. It may be fixable, but for now it's not important, so disable it. In the future if requested by users we can figure out if/how to fix the musl build. Release note: None Co-authored-by: David Taylor <tinystatemachine@gmail.com> Co-authored-by: Andrew Kryczka <andrew.kryczka2@gmail.com> Co-authored-by: Matt Jibson <matt.jibson@gmail.com>
Build succeeded |
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 feature is not widely used so is not the most stable, especially
when combined with other uncommonly used features like range tombstones.
In my
bin/workload kv --read-percent 0experiments it is not necessarysince the frequent foreground
fdatasyncs prevent compaction fromfalling behind, with or without subcompactions enabled.
Release note: None