Kill ckpool on bitcoind failure so miners can failover

When bitcoind is unreachable for 3+ consecutive polls and no block
submission is pending, SIGTERM the ckpool process so miners disconnect
and failover to backup pools. Adds a red dashboard banner when
Bitcoin Core is down.

api/internal/state/aggregator.go

@@ -190,6 +190,12 @@ type Aggregator struct {
 	// refresh. Used by the WebSocket hub to push updates to clients.
 	OnRefresh func(Snapshot)
 
+	// KillCKPool, if set, is called when bitcoind has been unreachable
+	// for several consecutive polls and no block submission is pending.
+	// Killing ckpool disconnects miners so they can failover to other
+	// pools instead of mining stale work.
+	KillCKPool func() error
+
 	mu     sync.RWMutex
 	snap   Snapshot
 	blocks []BlockRecord
@@ -231,6 +237,12 @@ type Aggregator struct {
 	// escalate to WARN.
 	ckFailStreak int
 
+	// btcFailStreak counts consecutive refreshes where bitcoind was
+	// unreachable. After a threshold, if no block submission is pending,
+	// we kill the ckpool process so miners can failover to other pools.
+	btcFailStreak    int
+	ckpoolKilled     bool // true after we sent SIGTERM, reset on bitcoind recovery
+
 	// readyOnce + ready closes the Ready() channel exactly once after
 	// the first refresh completes. main blocks briefly on this so the
 	// HTTP server doesn't serve a never-refreshed (all-zeros) snapshot.
@@ -424,6 +436,11 @@ func (a *Aggregator) refresh(ctx context.Context) {
 	if bi, err := a.RPC.GetBlockchainInfo(ctx); err == nil {
 		next.Chain = bi
 		next.BitcoinOK = true
+		if a.btcFailStreak > 0 {
+			a.Log.Info("bitcoind recovered", "after_failures", a.btcFailStreak)
+		}
+		a.btcFailStreak = 0
+		a.ckpoolKilled = false
 		// Track when the tip height last changed.
 		if bi.Blocks != a.lastTipHeight {
 			a.lastTipHeight = bi.Blocks
@@ -436,7 +453,8 @@ func (a *Aggregator) refresh(ctx context.Context) {
 			next.NetworkHashrateHs = nh
 		}
 	} else {
-		a.Log.Warn("bitcoind getblockchaininfo failed", "err", err)
+		a.btcFailStreak++
+		a.Log.Warn("bitcoind getblockchaininfo failed", "err", err, "streak", a.btcFailStreak)
 		if next.LastError == "" {
 			next.LastError = err.Error()
 		}
@@ -675,6 +693,26 @@ func (a *Aggregator) refresh(ctx context.Context) {
 		cb(pushed)
 	}
 	a.markReady()
+
+	// Kill ckpool when bitcoind is unreachable so miners can failover.
+	// Conditions: 3+ consecutive failures, no pending block submission,
+	// haven't already killed it, and a kill callback is configured.
+	const btcFailThreshold = 3
+	submitGap := next.BlockSubmitAttempts - next.BlockSubmitsConfirmed
+	if a.btcFailStreak >= btcFailThreshold && !a.ckpoolKilled && a.KillCKPool != nil {
+		if submitGap > 0 {
+			a.Log.Warn("bitcoind down but block submission pending, keeping ckpool alive",
+				"submit_gap", submitGap, "streak", a.btcFailStreak)
+		} else {
+			a.Log.Warn("bitcoind unreachable, killing ckpool so miners can failover",
+				"streak", a.btcFailStreak)
+			if err := a.KillCKPool(); err != nil {
+				a.Log.Error("failed to kill ckpool", "err", err)
+			} else {
+				a.ckpoolKilled = true
+			}
+		}
+	}
 }
 
 // AckBestDiff records the current best_diff as acknowledged so the UI