[Flaky test] transparent_decompress_chunk flaky because of difference in sort method

[mkindahl]

Which test is flaky?

transparent_decompress_chunk-13

Since when has the test been flaky?

February 2, 2024

Link to the failed test run

https://github.com/timescale/timescaledb/actions/runs/7746993406/job/21126475048

Log output

diff -u /mnt/tsl/test/shared/expected/transparent_decompress_chunk-13.out /mnt/build_abi/tsl/test/shared/results/transparent_decompress_chunk-13.out
--- /mnt/tsl/test/shared/expected/transparent_decompress_chunk-13.out
+++ /mnt/build_abi/tsl/test/shared/results/transparent_decompress_chunk-13.out
@@ -204,10 +204,10 @@
    ->  Gather Merge (actual rows=10 loops=1)
          Workers Planned: 2
          Workers Launched: 2
-         ->  Sort (actual rows=10 loops=2)
+         ->  Sort (actual rows=5 loops=2)
                Sort Key: _hyper_X_X_chunk."time", _hyper_X_X_chunk.device_id
                Worker 0:  Sort Method: top-N heapsort 
-               Worker 1:  Sort Method: top-N heapsort 
+               Worker 1:  Sort Method: quicksort 
                ->  Custom Scan (DecompressChunk) on _hyper_X_X_chunk (actual rows=8995 loops=2)
                      Filter: (device_id < v1)
                      ->  Parallel Seq Scan on compress_hyper_X_X_chunk (actual rows=10 loops=2)
@@ -351,7 +351,7 @@
    ->  Gather Merge (actual rows=10 loops=1)
          Workers Planned: 2
          Workers Launched: 2
-         ->  Sort (actual rows=10 loops=2)
+         ->  Sort (actual rows=6 loops=2)
                Sort Key: _hyper_X_X_chunk."time", _hyper_X_X_chunk.device_id
                Worker 0:  Sort Method: top-N heapsort 
                Worker 1:  Sort Method: top-N heapsort

Reason for flakiness

Part of the difference is a different sort method picked. This is not really relevant for the correctness of the test and it might be a good idea to eliminate the sort method from the output.

[antekresic]

@akuzm was looking at this test flakiness, not sure if it addresses this exact problem
#6550

[akuzm]

@akuzm was looking at this test flakiness, not sure if it addresses this exact problem #6550

Yeah that's the one that went up to fixes in Postgres :) Trying to fix it uncovers many more broken things.