[WIP] cache: remove redundant find() call #33376
Conversation
|
The following sections might be updated with supplementary metadata relevant to reviewers and maintainers. Code Coverage & BenchmarksFor details see: https://corecheck.dev/bitcoin/bitcoin/pulls/33376. ReviewsSee the guideline for information on the review process. ConflictsReviewers, this pull request conflicts with the following ones:
If you consider this pull request important, please also help to review the conflicting pull requests. Ideally, start with the one that should be merged first. |
Raimo33
force-pushed
the
batch-write-optimize-find
branch
from
b51190c to
9df8c9d
Compare
This commit reduces the number of implicit calls to ::find() by using an optimistic approach: insert preemptively, remove O(1) if already spent. Benchmarks can't currently measure this change in a realistic manner, because they don't account for cache common related scenarios. This change cannot possibly be a regression since ::try_emplace() is a call to find() under the hood. And ::erase(it) is constant time in the worst case. Currently, the ::find() calls account for ~2.9% of total IBD time as shown in issue bitcoin#32832
Raimo33
force-pushed
the
batch-write-optimize-find
branch
from
9df8c9d to
40e75ea
Compare
|
Multiple people have pushed this already, e.g. |
|
Following up in l0rinc@b23c6a1 as reviewer/coauthor |
4 participants
Summary
This PR halves the number of implicit calls to
::find()by using an optimistic approach: insert preemptively, remove in O(1) if already spent.Motivation
Currently, the
::find()calls that originate fromCCoinsViewCache::BatchWrite()account for ~2.9% of total IBD time as shown by this flamegraph from issue #32832Benchmarks
Benchmarks can't currently measure this change in a realistic manner, because they don't account for cache common related scenarios (see #33375).
This change cannot possibly be a regression since
::try_emplace()is a call to::find()under the hood. And::erase(it)is constant time in the worst case.I'm marking this PR as draft because I ultimately want to measure the impact on a real IBD run.