chore: release v0.20.7

internal/db/mysql.go

@@ -134,15 +134,19 @@ func translateTextColumns(s string) string {
 //     server's global max_connections — "Error 1040: Too many connections".
 //   - Too SMALL a pool starves this instance's OWN concurrent UI queries — the live
 //     multi-op sync (revision poll every 2s + a 3-query grid refresh), live_status,
-//     chat, stats, cluster — so the grid stops showing other operators' QSOs. (8 was
-//     too tight and reintroduced this "deadlock-like stall".)
-// 16 is the balance: plenty for the UI's bursts, and ~15 ops still fit under a
-// raised max_connections=300 (advise operators to raise it for a big multi-op).
+//     chat, stats, cluster. Under real multi-op load, 8 and even 16 stalled: hot
+//     paths (WorkedBefore fires ~10 sequential round-trips; every poll uses the
+//     app-lifetime ctx with no per-query timeout) hold a connection for the whole
+//     chain of remote-MySQL latency, so a handful of concurrent UI bursts drained
+//     the pool and "after a while nothing updated" — grid, chat, live_status froze.
+// 40 fits the actual deployment (max 5 ops on a server with max_connections=300 →
+// 40*5 = 200, leaving ~100 for phpMyAdmin/server overhead). This is the per-INSTANCE
+// pool, so keep max_connections >= pool*ops + headroom.
 // Brisk idle reaping returns slots to the server; no max lifetime so a slow first
 // migration on one connection isn't reaped mid-run (drops the selected database).
 func tuneMySQLPool(conn *sql.DB) {
-	conn.SetMaxOpenConns(16)
-	conn.SetMaxIdleConns(6)
+	conn.SetMaxOpenConns(40)
+	conn.SetMaxIdleConns(8)
 	conn.SetConnMaxIdleTime(30 * time.Second)
 	conn.SetConnMaxLifetime(0)
 }