kpumuk opened a new pull request, #3741:
URL: https://github.com/apache/thrift/pull/3741
<!-- Explain the changes in the pull request below: -->
The native Ruby Compact Protocol previously called the transport's write
method once per encoded byte for multibyte varints. This created up to five
one-byte Ruby strings for an i32 and ten for an i64, along with the
corresponding dynamic method dispatches.
This change encodes each varint into a bounded C stack buffer and passes it
through the existing transport write boundary once. It preserves the Compact
wire format and generic transport behavior without specializing the
optimization for `MemoryBufferTransport`.
## Benchmarks
Measured on Ruby 4.0.6, aarch64-linux, using:
```text
ruby test/rb/benchmarks/protocol_benchmark.rb --json --scenarios
rb-cmp-write-small,rb-cmp-read-small --small-runs 500000
```
The pure-Ruby control used the same command with
`THRIFT_BENCHMARK_SKIP_NATIVE=1` and `--small-runs 150000`. Each timing is the
median of three measured trials after one discarded warm-up.
| Scenario | master median (range) | proposed median (range) | Change |
| --- | ---: | ---: | ---: |
| Native Compact write | 2.666s (2.657–2.724s) | 2.523s (2.512–2.585s) |
5.3% faster |
| Native Compact read control | 2.639s (2.612–2.683s) | 2.673s
(2.625–2.778s) | 1.3% slower |
| Pure-Ruby Compact write control | 2.260s (2.233–2.288s) | 2.260s
(2.248–2.355s) | unchanged |
Allocations were measured around the same `rb-cmp-write-small` scenario with
`GC.stat(:total_allocated_objects)`, using 500,000 writes, one discarded
warm-up, and three measured trials.
| Allocation metric | master median (range) | proposed median (range) |
Change |
| --- | ---: | ---: | ---: |
| Allocated objects per 500,000 writes | 16,500,007 (16,500,007–16,500,008)
| 12,000,007 (12,000,007–12,000,008) | 27.3% fewer |
| Allocated objects per small struct | approximately 33 | approximately 24 |
9 fewer |
The benchmark serializes the small `OneOfEach` fixture through
`MemoryBufferTransport` and measures this change as a standalone commit. The
allocation count includes protocol and transport setup but measures objects
rather than allocated bytes or peak memory. These results do not cover large
payloads or imply the earlier aggregate 2× result.
<!-- We recommend you review the checklist/tips before submitting a pull
request. -->
- [x] Did you create an [Apache
Jira](https://issues.apache.org/jira/browse/THRIFT-6153) ticket? ([Request
account here](https://selfserve.apache.org/jira-account.html), not required for
trivial changes)
- [x] If a ticket exists: Does your pull request title follow the pattern
"THRIFT-NNNN: describe my issue"?
- [x] Did you squash your changes to a single commit? (not required, but
preferred)
- [x] Did you do your best to avoid breaking changes? If one was needed,
did you label the Jira ticket with "Breaking-Change"?
- [ ] If your change does not involve any code, include `[skip ci]` anywhere
in the commit message to free up build resources.
<!--
The Contributing Guide at:
https://github.com/apache/thrift/blob/master/CONTRIBUTING.md
has more details and tips for committing properly.
-->
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
