PS-9378: Backport PS-9302 (WRITESET perf improvements) to 8.0 (8.0) #5409
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Outdated suggestions cannot be applied.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.
https://perconadev.atlassian.net/browse/PS-9378
This patch was originally created for the 8.4 branch where the only
available binlog transaction dependency tracking mode is 'WRITESET'.
However, it also makes sense to backport this to 8.0 as the same
code paths are used when user explicitly set the
'binlog_transaction_dependency_tracking' system variable to 'WRITESET'
(by default in 8.0 series, this variable is set to 'COMMIT_ORDER').
Cherry-picked PS-9302
"Improve performance for binlog_transaction_dependency_tracking=WRITESET"
(https://perconadev.atlassian.net/browse/PS-9302) from 8.4.
Problem:
Comparing to 8.0 where the default value of
binlog_transaction_dependency_tracking was COMMIT_ORDER, 8.4 introduces a visible write performance drop.
Cause:
8.4 uses WRITESET dependency tracking. For this tracking we maintain a map of binlog_transaction_dependency_history_size for row_id to newest transaction sequence_number which modified a given row. Every new transaction needs to check its dependency by examining the map (find), which is done with logarithmic complexity.
Solution:
Change std::map to std::unordered_map. This allows us to find a row dependency in constant on average complexity.