-
Notifications
You must be signed in to change notification settings - Fork 2.3k
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Spark 3.5: Don't cache or reuse manifest entries while rewriting meta…
…data by default (#8935) The action for rewriting manifests caches the manifest entry DF or does an extra shuffle in order to skip reading the actual manifest files twice. We did this assuming it would increase the performance. However, the caching seems to perform poorly for larger tables as it requires substantial cluster resources. In addition, doing a round-robin repartition is expensive as the entries must be written to disk. The extra write is actually more expensive than the extra read required for the range-based shuffle of manifest entries. Therefore, this change disables caching by default and removes the optional round-robin repartition step. Instead, we will read the manifests twice (this step is distributed and scales really well even for tables with huge metadata). The new approach should be both faster and more robust.
- Loading branch information
1 parent
a661663
commit fceea89
Showing
2 changed files
with
32 additions
and
19 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters