Skip to content

~3x perf drop on GHA "Search+" query at commit 5a382f9 #7310

Description

@philrz

Using the attached synth.json.gz test data, the performance of the following query dropped significantly at commit 5a382f9, which is associated with the merge of #7200.

$ super -version && time super -c "SELECT count() FROM 'synth.json.gz' WHERE grep('in case you have any feedback', this);"
Version: v0.3.0-317-g770db1c36
{count:3}
real	0m0.925s
user	0m3.760s
sys	0m0.247s

$ super -version && time super -c "SELECT count() FROM 'synth.json.gz' WHERE grep('in case you have any feedback', this);"
Version: v0.3.0-318-g5a382f927
{count:3}
real	0m4.970s
user	0m43.189s
sys	0m1.559s

Details

Repro is with super commit 5a382f9, which is associated with the merge of #7200.

The symptom was originally spotted in a run of the GHA benchmark's Search+ query on a reference AWS EC2 m6idn.2xlarge instance where the runtime went from ~17 seconds to ~55 seconds. The simplified repro above was cooked by Claude after I had it do the binary search to confirm the commit where the drop occurred. Claude's full write-up analyzing the root cause and proposing ideas for a fix is available in a Gist.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions