Memberlist: Optimise receive path by not cloning current ring state. #76
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 memberlist is used as the backing store for the ring, processing updates can be costly if the ring state is very large. This is partly due to the ring state being cloned upon the arrival of every message. It is believed that this can be avoided. Additionally, the reduced memory allocations will reduce the amount of work required by the GC. The included benchmark shows a good improvement: Command: ``` dskit/bench$ go test -v -bench BenchmarkMemberlistReceiveWithRingDesc -benchmem -cpu 6 ``` ``` BenchmarkMemberlistReceiveWithRingDesc-6 1138 975938 ns/op 551572 B/op 1264 allocs/op ``` ``` BenchmarkMemberlistReceiveWithRingDesc-6 3580 319163 ns/op 239799 B/op 37 allocs/op ```
stevesg
commented
Nov 23, 2021
|
I'm getting race conditions when running memberlist tests with |
|
Me too, looking into it. |
This is necessary because the Mergeable implementors are allowed to reference the source state in the returned changes. For example, in `ring.Desc`, the `Tokens` slice can be shared and is not deep copied. For safety, we can just clone the whole thing, given the changes are typically small. ``` BenchmarkMemberlistReceiveWithRingDesc-6 3229 324260 ns/op 239825 B/op 37 allocs/op ```
|
I've added a fix to The performance appears to be unchanged. (Edit: Removed outdated benchmark numbers) |
1 task
bboreham
reviewed
Nov 24, 2021
stevesg
commented
Nov 24, 2021
pracucci
approved these changes
Nov 24, 2021
stevesg
force-pushed
the
memberlist-clone
branch
from
f4c8d07
to
b827143
Compare
replay
reviewed
Nov 24, 2021
replay
reviewed
Nov 24, 2021
replay
reviewed
Nov 24, 2021
replay
approved these changes
Nov 24, 2021
pracucci
approved these changes
Nov 25, 2021
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.
What this PR does:
When memberlist is used as the backing store for the ring, processing updates
can be costly if the ring state is very large. This is partly due to the ring
state being cloned upon the arrival of every message. It is believed that this
can be avoided. Additionally, the reduced memory allocations will reduce the
amount of work required by the GC.
The included benchmark shows a good improvement:
Command:
(Edit: Updated results to take into account adding tokens to benchmark)
Which issue(s) this PR fixes:
Fixes N/A
Checklist
CHANGELOG.mdupdated - the order of entries should be[CHANGE],[FEATURE],[ENHANCEMENT],[BUGFIX]