wangyong9999 commented on issue #305: URL: https://github.com/apache/paimon-cpp/issues/305#issuecomment-5615000909
### Post-merge validation of the option-parsing change A later controlled local-filesystem test isolated the `OptionsUtils` part of #303 from the snapshot-reuse change. Test shape: - Linux x86_64, one process, 32 physical / 64 logical CPUs. - PK table with 2.1M rows and 22 uncompressed Parquet files. - Fixed snapshot, identical wide projection and lookup keys. - Four alternating rounds at C1 and C8. - The compared builds differed only by the option-parsing implementation. | Concurrency | Before AVG | After AVG | Change | |---|---:|---:|---:| | C1 | 4.060 ms | 3.658 ms | -9.9% | | C8 | 5.399 ms | 3.896 ms | -27.8% | An on-CPU `perf` profile (`-F 999 -g --call-graph dwarf`) attributed approximately 5.34% of the relevant point-read stack to `CoreOptions::FromMap` before the change and 1.34% afterward. The missing-option `NotExist` formatting and locale-related subtree disappeared. An independent post-fix C64 saturation profile reported 1.36% `CoreOptions` stack coverage, confirming that option parsing was no longer a dominant cost. This profile result is specific to the measured request path and is not an additive CPU or capacity claim. At C64 the host was already CPU-saturated and snapshot/manifest planning, Parquet page reads, and materialization dominated the remaining samples. Semantics were rechecked for present values, missing required values, missing values with defaults, invalid conversion, and explicitly empty strings. The merged implementation preserves those outcomes while avoiding error construction for an expected absent optional value. -- 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]
