[10.2][cache] aggWriteDone error cleanup uses a byte offset for del_dir, so failed-write dir entries are never removed

Branch: 10.2.x (10.2.1, commit 31f1f2f3b) · Severity: medium
Location: src/iocore/cache/StripeSM.cc:761

What's wrong

On an aggregation disk-write failure, aggWriteDone tries to remove every directory entry it inserted for the fragments in the failed buffer. It builds the search key with dir_set_offset(&del_dir, directory.header->write_pos + done) — a raw byte position. But directory entries store a vol/block offset: the insert paths in the same file use dir_set_offset(&vc->dir, this->offset_to_vol_offset(doc_offset)) (StripeSM.cc:950, :967). Because the offsets are in different units, directory.remove() searches for a key that never matches, so the entries for the failed write are left dangling in the directory.

Fix

Convert the byte position to a vol/block offset, matching the insert paths:

dir_set_offset(&del_dir, this->offset_to_vol_offset(directory.header->write_pos + done));

PR attached.

Testing

Compiles cleanly on 10.2.x (libinkcache.a builds). clang-format clean.

From an automated multi-lens audit of the 10.2.x branch. Full report on branch claude/codebase-audit-review-9nw7vz (CODEBASE_AUDIT_10.2.md).

Start at src/iocore/cache/StripeSM.cc:761 in aggWriteDone, then compare the directory-entry insertion paths at lines 950 and 967 to confirm the offset units. Build libinkcache.a on the 10.2.x branch and verify that failed aggregation writes clean up their directory entries.