fix: add missing flush op timeout (grafana#13679)

Author: grobinson-grafana

docs/sources/shared/configuration.md

@@ -358,7 +358,7 @@ ingester_rf1:
   # The timeout for an individual flush. Will be retried up to
   # `flush-op-backoff-retries` times.
   # CLI flag: -ingester-rf1.flush-op-timeout
-  [flush_op_timeout: <duration> | default = 10m]
+  [flush_op_timeout: <duration> | default = 10s]
 
   # Forget about ingesters having heartbeat timestamps older than
   # `ring.kvstore.heartbeat_timeout`. This is equivalent to clicking on the

pkg/ingester-rf1/flush.go

@@ -96,8 +96,10 @@ func (i *Ingester) flush(l log.Logger, j int, it *wal.PendingSegment) error {
 }
 
 func (i *Ingester) flushSegment(ctx context.Context, j int, w *wal.SegmentWriter) error {
-	start := time.Now()
+	ctx, cancelFunc := context.WithTimeout(ctx, i.cfg.FlushOpTimeout)
+	defer cancelFunc()
 
+	start := time.Now()
 	i.metrics.flushesTotal.Add(1)
 	defer func() { i.metrics.flushDuration.Observe(time.Since(start).Seconds()) }()
 

pkg/ingester-rf1/ingester.go

@@ -110,7 +110,7 @@ func (cfg *Config) RegisterFlags(f *flag.FlagSet) {
 	f.DurationVar(&cfg.FlushOpBackoff.MinBackoff, "ingester-rf1.flush-op-backoff-min-period", 100*time.Millisecond, "Minimum backoff period when a flush fails. Each concurrent flush has its own backoff, see `ingester.concurrent-flushes`.")
 	f.DurationVar(&cfg.FlushOpBackoff.MaxBackoff, "ingester-rf1.flush-op-backoff-max-period", time.Minute, "Maximum backoff period when a flush fails. Each concurrent flush has its own backoff, see `ingester.concurrent-flushes`.")
 	f.IntVar(&cfg.FlushOpBackoff.MaxRetries, "ingester-rf1.flush-op-backoff-retries", 10, "Maximum retries for failed flushes.")
-	f.DurationVar(&cfg.FlushOpTimeout, "ingester-rf1.flush-op-timeout", 10*time.Minute, "The timeout for an individual flush. Will be retried up to `flush-op-backoff-retries` times.")
+	f.DurationVar(&cfg.FlushOpTimeout, "ingester-rf1.flush-op-timeout", 10*time.Second, "The timeout for an individual flush. Will be retried up to `flush-op-backoff-retries` times.")
 	f.DurationVar(&cfg.MaxSegmentAge, "ingester-rf1.max-segment-age", 500*time.Millisecond, "The maximum age of a segment before it should be flushed. Increasing this value allows more time for a segment to grow to max-segment-size, but may increase latency if the write volume is too small.")
 	f.IntVar(&cfg.MaxSegmentSize, "ingester-rf1.max-segment-size", 8*1024*1024, "The maximum size of a segment before it should be flushed. It is not a strict limit, and segments can exceed the maximum size when individual appends are larger than the remaining capacity.")
 	f.IntVar(&cfg.MaxSegments, "ingester-rf1.max-segments", 10, "The maximum number of segments to buffer in-memory. Increasing this value allows for large bursts of writes to be buffered in memory, but may increase latency if the write volume exceeds the rate at which segments can be flushed.")