[IMPROVED] NumPending calculations and subject index memory in filestore #4960
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
Stree is an adaptive radix tree implementation used for storing and retrieving literal subjects. It also allows quick matching to wildcard subjects, which is it's major design goal along with using less memory in high subject cardinality situations. This will be used in the filestore implementation to replace the PSIM hash map which was fast at insert and lookup but suffered when trying to filter based on wildcard subjects. This is used specifically in calculations on NumPending with a wildcard, and given we push folks to use larger muxed streams with down filtered consumers and/or mirrors this was becoming a performance issue. Signed-off-by: Derek Collison <derek@nats.io>
Signed-off-by: Derek Collison <derek@nats.io>
Signed-off-by: Derek Collison <derek@nats.io>
Jarema
approved these changes
Jan 18, 2024
Signed-off-by: Derek Collison <derek@nats.io>
Signed-off-by: Neil Twigg <neil@nats.io>
The `_pre` and `cparts` copies are more often than not unnecessary and result in potentially gigabytes of allocations which can slow down functions like `NumPending`. Passing `pre` and `nparts` down recursively without copies is usually safe, even where there are appends in those, because there is no concurrency and those appends will not modify the slice length back at the callsite. Instead we'll only reallocate either when `append` does so naturally (we've ran out of capacity) or when we know specifically that we're modifying something in-place (like in `matchParts`). This improves performance of things like `NumPending` with high subject cardinality. Signed-off-by: Neil Twigg <neil@nats.io>
When using `make(x, y, z)`, there is a heap escape due to the non-constant size/capacity. Try to stay on the stack instead, reducing GC pressure. Signed-off-by: Neil Twigg <neil@nats.io>
In #4974 I removed preallocated buffers for `pre` as the copies were unnecessary on each single recursion, however it turns out that having a preallocation up front removes quite a few unnecessary allocations from subject construction, relieving further pressure on the GC. Signed-off-by: Neil Twigg <neil@nats.io>
In #4974 I removed preallocated buffers for `pre` as the copies were unnecessary on each single recursion, however it turns out that having a preallocation up front removes quite a few unnecessary allocations from subject construction as the underlying memory gets reused throughout the iteration or match process, relieving further pressure on the GC. Signed-off-by: Neil Twigg <neil@nats.io>
Signed-off-by: Neil Twigg <neil@nats.io>
…ing memory for the GC. The solution was to have a node return its children as a []node. Since node256 is sparse the upper layers need to check for nil, but this improved the performance. Signed-off-by: Derek Collison <derek@nats.io>
derekcollison
added a commit
that referenced
this pull request
Jan 22, 2024
This updates the memory store to use the new subject tree from #4960 for per-subject tracking. Signed-off-by: Neil Twigg <neil@nats.io>
neilalexander
added a commit
that referenced
this pull request
Jan 22, 2024
…ore (#4960) Swapped out psim as a go hashmap for our stree implementation. Stree is an adaptive radix tree implementation used for storing and retrieving literal subjects. It also allows quick matching to wildcard subjects, which is it's major design goal along with using less memory in high subject cardinality situations. This will be used in the filestore implementation to replace the PSIM hash map which was fast at insert and lookup but suffered when trying to filter based on wildcard subjects. This is used specifically in calculations on NumPending with a wildcard, and given we push folks to use larger muxed streams with down filtered consumers and/or mirrors this was becoming a performance issue. Signed-off-by: Derek Collison <derek@nats.io> --------- Signed-off-by: Derek Collison <derek@nats.io> Signed-off-by: Neil Twigg <neil@nats.io> Co-authored-by: Neil Twigg <neil@nats.io>
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.
Swapped out psim as a go hashmap for our stree implementation.
Stree is an adaptive radix tree implementation used for storing and retrieving literal subjects. It also allows quick matching to wildcard subjects, which is it's major design goal along with using less memory in high subject cardinality situations.
This will be used in the filestore implementation to replace the PSIM hash map which was fast at insert and lookup but suffered when trying to filter based on wildcard subjects.
This is used specifically in calculations on NumPending with a wildcard, and given we push folks to use larger muxed streams with down filtered consumers and/or mirrors this was becoming a performance issue.
Signed-off-by: Derek Collison derek@nats.io