bvolpato opened a new pull request, #40862:
URL: https://github.com/apache/beam/pull/40862
DirectRunner extracts ParDo schema information and side-input mappings for
every bundle. Those helpers translate the full ParDo payload each time.
Profiling Nexmark's streaming queries showed repeated translation and
serialization on the worker threads.
Cache the serialized metadata per applied transform. Deserialize a fresh
copy for each evaluator because schema conversion functions have mutable state,
and clear the cache during cleanup.
In matched warm local runs on a busy shared host, Nexmark's streaming Query
7 took 174 seconds before and 26 seconds after, excluding Gradle setup. The
input, assertions, JDK and runner settings were unchanged.
The full Nexmark suite and focused DirectRunner tests covering schemas, side
inputs, state and lifecycle passed locally. Formatting and compilation checks
passed as well.
<!--
Thank you for your contribution! Follow this checklist to help us
incorporate your contribution quickly and easily:
- [ ] If this contribution is large, please file an Apache [Individual
Contributor License Agreement](https://www.apache.org/licenses/icla.pdf).
- [x] Replace these instructions with a description of your change and why
you have made it.
- [ ] Mention the appropriate issue in your description (for example:
`addresses #123` or `fixes #456`).
- [ ] Update `CHANGES.md` with noteworthy changes.
See [CONTRIBUTING.md]() for more details.
-->
--
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]
