[cbalan]

Cherry-pick of upstream k3s-io/kine@f805442 ("Fix connection leak in Migrate", Tom Wieczorek), with a conventional, non-fork scope so it passes the master fork policy.

Why

Generic.Migrate opened both COUNT(*) rows up front and returned after the first Scan failed. On any database without the legacy key_value table, the second row was never read and kept its connection checked out forever. When the SQLite statement cache is on (_stmt_cache_size, part of the default DSN on the release lines), that leaked statement also holds a read transaction on the startup snapshot, so every post-compaction PRAGMA wal_checkpoint(FULL) waits out the full 30 s busy timeout. That is the TestCompact failure on release/vcluster/v0.17.2 (ENGCP-1950, blocking #35).

Changes

• pkg/drivers/generic/generic.go: query kine only after the key_value row has been scanned and warrants it.
• pkg/drivers/sqlite/sqlite_test.go (new, cgo only): TestMigrateReleasesItsConnections opens the SQLite backend with a 1-connection pool and checks the DB is still usable after Migrate.

Backport adaptations for master: Migrate keeps plain SQL strings (master predates query.New), and the test uses master's default DSN parameters.

Testing

• TestMigrateReleasesItsConnections fails without the fix (pool exhausted, 5 s timeout) and passes with it.
• CGO_ENABLED=1 go test ./pkg/drivers/... ./pkg/logstructured/... passes; CGO_ENABLED=0 go build ./... passes.

A backport to release/vcluster/v0.17.2 as fix(fork): ... follows in a separate PR.

Refs: ENGCP-1950

🤖 Generated with Claude Code