call Vec::shrink_to_vec in a few strategic spots
#120
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
I think I started with a contiguous representation but switched to nested allocs. Which is kind of unfortunate, and I'd like to fix that, but for now we should document reality.
Previously the heap usage required for storing the lengths of each pattern was not included in the memory usage reported.
This commit reduces the *actual* heap memory used by noncontiguous NFAs, contiguous NFAs and DFAs. Previously, all three would report memory usage that only included what was actually being used within each Vec. But each Vec might actually have a lot more heap memory inside of it in the form of unused capacity. Potentially ~twice as much. This is because a Vec amortizes its growth. So when Vecs are used to build NFAs and DFAs, some of them wind up with potentially a lot of additional capacity that goes unused and unreported. This commit partially fixes that by calling shrink_to_vec in a few strategic places. However, there are still some nested Vecs, notably in the noncontiguous NFA, that remain unshrinked. I tried doing it, but the perf impact it had on construction was non-trivial. This is a good example of one of those things where you of course know that Vecs amortize their growth by reserving potentially extra capacity that yuo won't end up using, but still remain ignorant of the broader effects of this strategy in aggregate. Especially at the extremes when you start trying to build Aho-Corasick automatons with millions of patterns. We should think more carefully about using linked lists in the non-contiguous NFA to reduce memory usage, or perhaps thinking of some other way to avoid nested Vecs. It turns out that they end up penalizing us quite a bit. This commit was inspired by the investigation I did as part of this issue: G-Research/ahocorasick_rs#62
BurntSushi
force-pushed
the
ag/shrink-vecs
branch
from
fba9661 to
284737f
Compare
BurntSushi
added a commit
to BurntSushi/ripgrep
that referenced
this pull request
Aug 15, 2023
This brings in [1,2], which improves memory usage substantially when Aho-Corasick is used. [1]: BurntSushi/aho-corasick#120 [2]: BurntSushi/aho-corasick#121
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.
See the commit messages for more details.
This work was overall inspired by the investigation I did in this issue: G-Research/ahocorasick_rs#62
TL;DR - Creating lots of
Vecs can lead to a lot of spare unused capacity hanging around. This invariably leads to potentially large discrepancies between a process's peak memory usage and the memory usage as reported by something likeNFA::memory_usage.