fallintoplace opened a new pull request, #1752: URL: https://github.com/apache/iceberg-go/pull/1752
## Summary Equality delete matching currently scans every equality delete entry for every data file. This makes scan planning grow with the product of data and delete file counts. This change builds a small index before creating scan tasks: * unpartitioned equality deletes are kept in a global sequence-sorted list * partitioned equality deletes are grouped by partition spec and normalized partition value * each data file uses binary search to skip delete files with non-applicable sequence numbers Partitioned matching now follows Java's `DeleteFileIndex`: deletes apply within the same spec and partition, while unpartitioned deletes remain global. The common paths with no equality deletes or only global deletes avoid partition-key construction. ## Benchmarks The benchmark includes index construction and matching 10,000 data files. Each data file matches at most the newest delete so output materialization stays bounded. ```text delete files partitions before after speedup 100 1 3.17 ms 1.47 ms 2.2x 100 10 2.89 ms 1.09 ms 2.6x 100 100 2.86 ms 0.95 ms 3.0x 1,000 1 27.16 ms 1.51 ms 18.0x 1,000 10 27.14 ms 1.21 ms 22.4x 1,000 100 27.29 ms 1.12 ms 24.3x 10,000 1 269.21 ms 2.46 ms 109.3x 10,000 10 265.48 ms 2.42 ms 109.8x 10,000 100 265.76 ms 2.34 ms 113.6x ``` Command: ```sh go test ./table -run '^$' -bench '^BenchmarkEqualityDeleteIndex$' -benchtime=200ms -count=3 ``` ## Testing * `go test ./...` * `go test -race ./table ./table/compaction -run 'EqualityDelete|PartitionKey' -count=1` * `golangci-lint run --timeout=10m ./table/...` -- 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]
