Optimize recompression for non-segmentby chunks #7632
Conversation
Enables the segmentwise recompression flow to be used for chunks without segmentby columns. This should be more performant than doing a full recompression.
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #7632 +/- ##
==========================================
+ Coverage 80.06% 81.30% +1.23%
==========================================
Files 190 240 +50
Lines 37181 44696 +7515
Branches 9450 11159 +1709
==========================================
+ Hits 29770 36340 +6570
- Misses 2997 3967 +970
+ Partials 4414 4389 -25 ☔ View full report in Codecov by Sentry. |
| @@ -168,6 +168,10 @@ recompress_chunk_segmentwise_impl(Chunk *uncompressed_chunk) | |||
|
|
|||
| CompressedSegmentInfo *current_segment = palloc0(sizeof(CompressedSegmentInfo) * n_keys); | |||
|
|
|||
| // For chunks with no segmentby settings, we can still do segmentwise recompression | |||
| // The entire chunk is treated as a single segment | |||
| elog(ts_guc_debug_compression_path_info ? INFO : DEBUG1, "using non-segmentby index for recompression") ; | |||
There was a problem hiding this comment.
This will log every time that you are using non-segmentby index but thats not true. You should log the index name you are using instead (its easy to check what index is being used that way).
| @@ -483,7 +483,7 @@ select compressed_chunk_name as compressed_chunk_name_after_recompression from c | |||
| select :'compressed_chunk_name_before_recompression' as before_recompression, :'compressed_chunk_name_after_recompression' as after_recompression; | |||
| before_recompression | after_recompression | |||
| ----------------------------+---------------------------- | |||
| compress_hyper_13_14_chunk | compress_hyper_13_15_chunk | |||
| compress_hyper_13_14_chunk | compress_hyper_13_14_chunk | |||
There was a problem hiding this comment.
Comment above on line 475 needs updating since it is incorrect with this change.
| @@ -291,6 +291,23 @@ insert into nullseg_many values (:'start_time', 1, NULL, NULL); | |||
| SELECT compress_chunk(:'chunk_to_compress'); | |||
| select * from :compressed_chunk_name; | |||
|
|
|||
| -- Test behaviour when no segmentby columns are present | |||
| SET timescaledb.debug_compression_path_info TO ON; | |||
There was a problem hiding this comment.
Lets enable this GUC for the complete test so we can verify that the correct index is being used for each recompression.
Enables the segmentwise recompression flow to be used for chunks without segmentby columns.
This should be more performant than doing a full recompression.