andygrove opened a new pull request, #1595: URL: https://github.com/apache/datafusion-ballista/pull/1595
# Which issue does this PR close? Closes #. # Rationale for this change The user guide does not currently describe how Ballista's shuffle works or that two distinct writer implementations exist. The sort-based shuffle writer and its four `ballista.shuffle.sort_based.*` session-level config keys are only discoverable by reading the source. Anyone trying to tune a shuffle-heavy workload — particularly to reduce the `N × M` file count produced by the default hash writer — has no documentation to point them at the alternative. # What changes are included in this PR? - Add a "Shuffle Implementation" section to the tuning guide that explains the default hash-based writer (one file per `(input_partition, output_partition)` pair, no buffering) and the opt-in sort-based writer (per-output-partition in-memory buffers with spilling, consolidated data + index file per input partition), and lists the four `ballista.shuffle.sort_based.*` keys that tune it. - Add a "Shuffle Settings" table to `configs.md` covering the same sort-shuffle keys plus the existing `ballista.shuffle.max_concurrent_read_requests`, `ballista.shuffle.force_remote_read`, and `ballista.shuffle.remote_read_prefer_flight` keys, which were also previously undocumented in the user guide. # Are there any user-facing changes? Documentation only. No code, behavior, or public API changes. -- 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]
