Use proto encoding for scheduler workflow next time cache #5277
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
ast2023
approved these changes
Jan 10, 2024
service/worker/scheduler/workflow.go
Outdated
| cache := s.nextTimeCacheV2 | ||
| start := cache.StartTime.AsTime() | ||
| afterOffset := int64(after.Sub(start)) | ||
| for i, nextOffset := range cache.NextTimes { |
There was a problem hiding this comment.
Could this block (3 assignments and the for loop) be moved into a separate function?
dnr
added a commit
to dnr/temporal
that referenced
this pull request
Apr 10, 2024
dnr
added a commit
that referenced
this pull request
Apr 10, 2024
…5698) ## What changed? Activate schedule workflow logic changes. Some of this code is not in 1.23.0, but we can patch those PRs into 1.23.1 to support downgrades to the 1.23 series. ## Why? Fix bugs, support new features, make more efficient. ## How did you test it? existing tests (on those PRs) ## Potential risks schedule workflow determinism errors
dnr
added a commit
that referenced
this pull request
Apr 10, 2024
…5698) ## What changed? Activate schedule workflow logic changes. Some of this code is not in 1.23.0, but we can patch those PRs into 1.23.1 to support downgrades to the 1.23 series. ## Why? Fix bugs, support new features, make more efficient. ## How did you test it? existing tests (on those PRs) ## Potential risks schedule workflow determinism errors
yycptt
pushed a commit
that referenced
this pull request
Apr 27, 2024
The scheduler workflow keeps a cache of a sequence of calls to getNextTime to reduce overhead. The cache was serialized as json, this changes it to serialize as proto. This is made forwards and backwards compatible by trying to deserialize into both types. The json decoding of this cache showed up as a significant cost in a cpu profile of a worker that was doing a lot of schedule workflow replays. In a simple benchmark of decoding only on my laptop, decoding the proto was about 20 times faster with no jitter (14 with jitter), and the encoded proto was 8 times smaller (5 with jitter). Encoding was about 4 times faster (but that's less important). ``` no jitter: BenchmarkDecodeJson-12 178678 33114 ns/op 1848 B/op 13 allocs/op BenchmarkDecodeProto-12 2249361 1571 ns/op 288 B/op 3 allocs/op with jitter: BenchmarkDecodeJson-12 110619 32657 ns/op 1848 B/op 13 allocs/op BenchmarkDecodeProto-12 1565641 2336 ns/op 400 B/op 4 allocs/op ``` existing tests (especially replay test for backwards + forwards compatibility) The code is a little more complicated and there could be bugs in the conversion. Note that both json and proto support nanosecond resolution for timestamps.
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 changed?
The scheduler workflow keeps a cache of a sequence of calls to getNextTime to reduce overhead. The cache was serialized as json, this changes it to serialize as proto. This is made forwards and backwards compatible by trying to deserialize into both types.
Why?
The json decoding of this cache showed up as a significant cost in a cpu profile of a worker that was doing a lot of schedule workflow replays. In a simple benchmark of decoding only on my laptop, decoding the proto was about 20 times faster with no jitter (14 with jitter), and the encoded proto was 8 times smaller (5 with jitter). Encoding was about 4 times faster (but that's less important).
How did you test it?
existing tests (especially replay test for backwards + forwards compatibility)
Potential risks
The code is a little more complicated and there could be bugs in the conversion. Note that both json and proto support nanosecond resolution for timestamps.