fix(archive): Finish forced archive when the database directory has leftover files

[regdocs]

Problem

A forced Archive Site job can fail at the database drop:

pymysql.err.OperationalError: (1010, 'Error dropping database (can't rmdir './_xxxxxxxx', errno: 39 "Directory not empty")')

DROP DATABASE only removes files MariaDB recognises as table files. Any other file in the schema directory stops the final rmdir. One example is a MyISAM auto-recovery backup (__global_search.MYD-<timestamp>.BAK), which database servers write under myisam-recover-options = BACKUP. bench drop-site --force only tolerates a failed backup, so it stops at this point. The site's database user is never dropped, the site directory never moves to archived, and the job fails. If this happens in the archive step of a Site Migration, the whole migration fails after the site has already been restored on the destination.

Change

In Bench.bench_archive_site, when force is set and drop-site fails with this error (1010 / errno 39 on the site's own schema directory):

• every table is already dropped, so the error is treated as non-fatal;
• every account named after the site's database user is dropped (mysql.user lookup), as drop-site would have done;
• the site directory moves to sites/archived/<site>, using drop-site's numeric suffix when a site with that name was archived before.

Without force, or for any other drop-site error, the archive fails as before.

Trade-off

The emptied schema directory, holding only the stray files, stays on the database server, and the schema still appears in SHOW DATABASES. Removing it needs access to the database host, which the app server's agent doesn't have.

Tests

Added to agent/tests/test_site.py:

• forced archive with the leftover-directory error drops the users and archives the site (including the name-collision suffix);
• the same error without force still fails;
• other drop-site errors with force still fail.

🤖 Generated with Claude Code

[mergify]

This pull request does not currently match the merge queue conditions, so it cannot be queued from here. The box comes back if it matches again.

[greptile-apps]

Confidence Score: 4/5

[High impact] Adds fallback logic to complete site archival when database cleanup fails.

Fix cleanup for sites with a separate db_user before merging.

_{Reviews (1) · Last reviewed commit: "fix(archive): Finish forced archive when..." · Reviewed by Greptile}

[greptile-apps]

Archived site's user remains

When db_user differs from db_name, the fallback drops accounts named after the database instead of the site's actual user. It then archives the site and reports success while leaving that user behind. Pass Site.user separately and use it for both account queries.

Prompt To Fix With AI

This is a comment left during a code review.
Path: agent/bench.py
Line: 466-468

Comment:
**Archived site's user remains**

When `db_user` differs from `db_name`, the fallback drops accounts named after the database instead of the site's actual user. It then archives the site and reports success while leaving that user behind. Pass `Site.user` separately and use it for both account queries.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.