Fix Share stats disappearing regression

api/internal/state/aggregator.go

@@ -7,6 +7,7 @@ package state
 import (
 	"context"
 	"encoding/json"
+	"errors"
 	"log/slog"
 	"strconv"
 	"sync"
@@ -371,6 +372,17 @@ type Aggregator struct {
 	sessionDiffSum       float64 // sum of share difficulties (session)
 	alltimeDiffSum       float64 // sum of share difficulties (alltime)
 	lastShareStatsSave   time.Time
+
+	// loadOnce guards LoadPersisted so main can complete the restore
+	// before starting the ingest goroutines, while Run stays safe to
+	// call on its own.
+	loadOnce sync.Once
+	// statsLoaded reports that the alltime share statistics have been
+	// read back from the store. Persisting before that would write the
+	// zero-valued in-memory counters over the accumulated history — the
+	// first share event triggers a save, so this is the difference
+	// between a restart and permanent data loss.
+	statsLoaded bool
 }
 
 func New(ck *ckpool.Client, rpc *bitcoind.RPC, interval time.Duration, log *slog.Logger) *Aggregator {
@@ -387,6 +399,19 @@ func New(ck *ckpool.Client, rpc *bitcoind.RPC, interval time.Duration, log *slog
 	}
 }
 
+// LoadPersisted restores block history and counters from the store.
+// Run calls it, but callers should invoke it directly before starting
+// any ingest goroutine: those goroutines mutate the same counters and
+// persist them on their first event, so a restore still in flight would
+// be overwritten by — and then read back as — near-empty state. Safe to
+// call repeatedly; only the first call does the work.
+func (a *Aggregator) LoadPersisted() {
+	a.loadOnce.Do(func() {
+		a.loadPersistedBlocks()
+		a.loadPersistedState()
+	})
+}
+
 // Ready returns a channel that's closed once the aggregator has
 // completed its first refresh — used by main to delay HTTP serving
 // until /api/snapshot reflects real state instead of zeros.
@@ -407,8 +432,7 @@ func (a *Aggregator) Run(ctx context.Context, tipEvents <-chan zmqmon.TipEvent)
 	a.mu.Lock()
 	a.zmqEnabled = tipEvents != nil
 	a.mu.Unlock()
-	a.loadPersistedBlocks()
-	a.loadPersistedState()
+	a.LoadPersisted()
 	a.refresh(ctx)
 	t := time.NewTicker(a.Interval)
 	defer t.Stop()
@@ -926,6 +950,50 @@ func (a *Aggregator) IngestShareEvents(ctx context.Context, events <-chan logmon
 	}
 }
 
+// RebuildShareStatsFromLog recounts the all-time share statistics by
+// scanning the whole ckpool log, and adopts the result when it accounts
+// for more shares than the counters currently hold. Recovery for stored
+// stats that were lost or truncated: the log is the only other record of
+// per-share difficulty, and ckpool keeps its full history there.
+//
+// Deliberately never automatic — the scan walks every line of a log that
+// can hold millions, and it was a long startup scan that made the race
+// this repairs so easy to hit in the first place.
+func (a *Aggregator) RebuildShareStatsFromLog() (adopted bool, scanned int64, err error) {
+	if a.LogFilePath == "" {
+		return false, 0, errors.New("no ckpool log path configured")
+	}
+	st, err := logmon.ScanShareStats(a.LogFilePath)
+	if err != nil {
+		return false, 0, err
+	}
+
+	var dist [6]int64
+	for _, d := range st.Diffs {
+		dist[diffBucket(d)]++
+	}
+
+	a.mu.Lock()
+	current := a.alltimeDiffCount
+	if st.Accepted <= current {
+		a.mu.Unlock()
+		a.Log.Info("share stats rebuild: log has no more shares than stored, keeping stored",
+			"log_shares", st.Accepted, "stored_shares", current)
+		return false, st.Accepted, nil
+	}
+	a.alltimeDiffDist = dist
+	a.alltimeDiffCount = st.Accepted
+	a.alltimeDiffSum = st.DiffSum
+	a.alltimeRejectReasons = st.RejectReasons
+	a.statsLoaded = true
+	a.mu.Unlock()
+
+	a.persistShareStats()
+	a.Log.Warn("share stats rebuilt from ckpool log",
+		"shares", st.Accepted, "was", current)
+	return true, st.Accepted, nil
+}
+
 // ResetSessionShareStats clears session-level share statistics. Called
 // when a ckpool restart is detected (pool.Shares decreases).
 func (a *Aggregator) resetSessionShareStats() {
@@ -939,6 +1007,15 @@ func (a *Aggregator) persistShareStats() {
 	if a.Store == nil {
 		return
 	}
+	a.mu.RLock()
+	loaded := a.statsLoaded
+	a.mu.RUnlock()
+	if !loaded {
+		// Belt and braces alongside the load-before-ingest ordering in
+		// main: writing here would replace the accumulated history with
+		// whatever few shares have arrived since startup.
+		return
+	}
 	a.mu.Lock()
 	reasonsCopy := make(map[string]int64, len(a.alltimeRejectReasons))
 	for k, v := range a.alltimeRejectReasons {
@@ -1165,6 +1242,14 @@ func (a *Aggregator) loadPersistedState() {
 		}
 	}
 
+	// From here the alltime counters reflect the store, so persisting
+	// them can no longer erase history.
+	a.mu.Lock()
+	a.statsLoaded = true
+	loadedShares := a.alltimeDiffCount
+	a.mu.Unlock()
+	a.Log.Info("share stats loaded", "alltime_shares", loadedShares)
+
 	cutoff := time.Now().Add(-24 * time.Hour).Unix()
 	if samples, err := a.Store.HashrateSince(cutoff); err != nil {
 		a.Log.Warn("hashrate history load failed", "err", err)