[deanq]

Problem

test_hydrate_rejects_source_replaced_between_ranges fails intermittently with assert restored == 0 (assert 1 == 0). Observed on Python 3.14 in CI on #665 (run 41127886501, attempts 2 and 3) and on the fix/preserve-empty-endpoint-input branch. It passed in the main runs checked. The test came from #621.

Cause

_file_identity compares only (st_size, st_mtime_ns); it skips the inode on purpose because of FUSE mounts (see its docstring). The test replaces the mirror file with random bytes of the same size, so detection depends only on mtime. sync() copies the source mtime onto the mirror with os.utime, and the replacement is written a few milliseconds later. On a filesystem with coarse timestamps both can land on the same mtime. The swap is then not detected and hydrate publishes the file (restored == 1).

Fix

Test-only. After writing the replacement, set its mtime to the original's plus one second, with a one-line comment. Production identity logic is unchanged: a replacement with the same size and the same mtime remains an accepted blind spot, and widening the identity (for example adding the inode) would reintroduce the FUSE false rejections the docstring describes.

Verification

• Natural flake did not reproduce locally: macOS APFS has nanosecond mtimes. 0/200 failures before the fix, 0/200 after (Python 3.14.6, single test in a loop).
• Mechanism reproduced: with the replacement's mtime forced equal to the original's (before the fix), the test fails with assert 1 == 0, the same assertion as CI. _file_identity of two same-size files with equal mtime compares equal; with mtime +1s it differs.
• Full test file: 118 passed.
• make quality-check: 2181 passed, coverage gate satisfied.

[Copilot]

🟢 Approval recommended

No active review issues are reported for the test-only adjustment that gives the replacement file a distinct modification time.

No open issues 🎉

What changed in this PR

Stabilizes the VolumeCache source-replacement test by ensuring the replacement has a distinct modification time, avoiding intermittent failures on filesystems with coarse timestamps.

Changes:

• Sets the replacement file’s mtime to the original’s plus one second.
• Documents why the explicit timestamp is needed; production identity logic remains unchanged.

File	Description
tests/​test_serverless/​test_utils/​test_rp_volume_cache.py	Forces a distinct replacement mtime in the between-ranges source-replacement test.

_{Review effort: Balanced}