iceberg-kitten opened a new pull request, #2642: URL: https://github.com/apache/iceberg-go/pull/2642
Depends on #2641. The diff against `main` includes that patch. [Snapshot-only diff](https://github.com/iceberg-kitten/iceberg-go/pull/1/files). Parse the common v2 snapshot fields and summaries directly during eager metadata loads. Unusual snapshots fall back individually to `encoding/json`, with the full document's nesting limit preserved. The manifest path and summary strings own their data, so a snapshot can outlive the table without retaining the history buffer. The regression tests cover malformed JSON, duplicate fields, escaped values, the exact depth limit, and input ownership. ## Benchmark | 64-snapshot workload | With #2641 | With this patch | | --- | ---: | ---: | | Warm metadata file load | 13.951 ms | 4.624 ms | | In-memory parse, bytes/op | 294,010 | 124,539 | | In-memory parse, allocs/op | 5,107 | 1,319 | File-load timings are medians from three matched rounds on one pinned core at approximately 0.9 GHz, with Go 1.26.10 and `GOGC=100`. Each load reads the same warm 66,321-byte v2 file and checks the schema width and snapshot sequence. The allocation rows are medians from three runs of the separate in-memory Go benchmark. ```sh GOMAXPROCS=1 GOGC=100 go test ./table -run '^$' \ -bench '^BenchmarkParseMetadataBytes$/^schemas=1$/^specs=1$/^snapshots=64$' \ -benchmem -benchtime=300ms -count=3 ``` Run the same command on the head of #2641 and on this branch. `make test` and `make lint` passed on the full stack. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
