[grempe]

Fixes #1243.

newMySQLSequencer and NewAntispam applied MaxIdleConns whenever it was >= 0. A config that left it unset therefore called SetMaxIdleConns(0), and the pool kept no idle connections: every query opened a new MySQL connection and closed it afterwards.

This change applies it whenever it's != 0. An unset value then keeps the database/sql default, which is currently 2, and a negative value still keeps no idle connections, as database/sql does for any n <= 0. MaxOpenConns and MaxIdleConns are now documented on both aws.Config and AntispamOpts.

Testing

• New TestMySQLSequencerIdleConnections and TestAntispamIdleConnections. Each makes 10 sequential queries on a new pool and checks DBStats.MaxIdleClosed:
  • with the setting unset, none of the 10 queries closes its connection; on main, all 10 do;
  • with -1, all 10 do; on main, none do, because main ignores a negative value.
• Mutants: setting either half back to >= 0, or to > 0, fails only that half's test.
• Race tests: go test -race ./storage/aws/... -is_mysql_test_optional=false against MySQL 8.4 passes on Go 1.26.10, and the tests also pass on Go 1.27.3.
• Lint: golangci-lint v2.13.2 reports 0 issues.
• Full suite: it passes on macOS apart from TestConstructAndUseAppender (/tmp is not a directory), which fails the same way on main here.
• Under load: 100,000 entries through the AWS driver to RustFS and MySQL 8.4. The table counts the MySQL server's new connections:

MaxIdleConns	main	this change
unset	1,697–1,774 (3 runs)	13–177, median 20 (5 runs)
-1	23	1,699–1,804 (2 runs)
64	5	5

The spread when unset comes from the default's 2 idle connections: when more queries run at once, the extra connections are closed afterwards.

A behaviour change to note: an explicit MaxIdleConns: 0 used to keep no idle connections; it now keeps the default, and -1 keeps none. The same goes for --db_max_idle_conns=0 in the AWS conformance and migrate binaries. Their help text ("defaults to 2") still holds, so this PR leaves them unchanged.

---

Reviewed and submitted by me, a human. The investigation, the fix and its testing were developed with the assistance of an AI coding agent.

[grempe]

Rebased onto main and moved the new tests to the middle of their files, so this merges cleanly with #1252 and the other open PRs in any order. The change itself is the same.

[grempe]

Thanks! Switched to != 0 with your comments, added your doc comments to both aws.Config and AntispamOpts, and rebased onto main. The tests now also check that -1 keeps no idle connections. I applied the first suggestion by hand: its range ends just above MaxIdleConns int, so committing it as-is would have declared the field twice. As for the conformance and migrate binaries, --db_max_idle_conns=0 now means the default too; their help text still holds, so I've left them alone, but I'm happy to spell out zero and negative values there.