parquet: Add option to cache file metadata #12548
Closed
Conversation
This file contains hidden or 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
Inspired by datafusion-examples/examples/advanced_parquet_index.rs
github-actions
bot
added
core
Contributor
Author
|
Closing in favor of #12593, which actually works and doesn't require extensive changes to Datafusion. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Inspired by datafusion-examples/examples/advanced_parquet_index.rs
Which issue does this PR close?
This was an attempt to solve #12547, but did not achieve it, and I am not sure it is the right approach.
Rationale for this change
On every query on Parquet ables, Datafusion re-opens every file, and parses its metadata. This takes a significant time for short queries (in my use case, there is usually a single hit in the Page Index).
My goal with to make these queries near-instant. Unfortunately, I realized after writing this code that the Page Index still needs to be parsed every time, because file metadata is lost through the
listinglayer (as mentioned in #9964).So this does spare some (negligible?) time parsing metadata. I'm not sure it's worth the extra complexity, especially in
ParquetFormat. What do you think?What changes are included in this PR?
ParquetFormatcarry state (it probably deserves a renaming then...)CachedParquetFileReaderFactoryas an alternative toDefaultParquetFileReaderFactory, and made it usable through a config optionAre these changes tested?
no
Are there any user-facing changes?
Added
datafusion.execution.parquet.cache_metadata