fix(ci): keep benchmark stacks comparable

The before stack is a worktree at an arbitrary older SHA, so its compose
file need not gate databases at all. With an empty COMPOSE_PROFILES only
the head side was constrained, leaving the before run carrying MariaDB
and MongoDB as idle load and the after run without them -- the reported
delta would include two fewer containers rather than the code change.

Start the same superset on both sides.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
loks0nandClaude Opus 5 committed 2026-07-29 14:15:54 +01:00
1 parent d0c183b93c
commit 01a8df8d3d
1 file changed
+4 -1
+4 -1
View File
@@ -705,7 +705,10 @@ jobs:
pull-requests: write
packages: read
env:
COMPOSE_PROFILES: ''
# The before stack is a worktree at an arbitrary older SHA, so its compose file
# may not gate databases at all. Start the superset on both sides, otherwise the
# two runs measure different amounts of idle container load rather than the diff.
COMPOSE_PROFILES: mariadb,mongodb
steps:
- name: Checkout repository
uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2