fix: export prometheus metrics for cache GC triggers (#9815)

Bonus change to use channel to serialize triggers,
instead of using atomic variables. More efficient
mechanism for synchronization.

Co-authored-by: Nitish Tiwari <nitish@minio.io>

cmd/disk-cache-backend.go

@@ -125,12 +125,14 @@ func (m *cacheMeta) ToObjectInfo(bucket, object string) (o ObjectInfo) {
 
 // represents disk cache struct
 type diskCache struct {
-	gcCounter uint64 // ref: https://golang.org/pkg/sync/atomic/#pkg-note-BUG
 	// is set to 0 if drive is offline
-	online uint32
+	online       uint32 // ref: https://golang.org/pkg/sync/atomic/#pkg-note-BUG
+	purgeRunning int32
 
-	dir           string // caching directory
+	triggerGC     chan struct{}
-	quotaPct      int    // max usage in %
+	dir           string         // caching directory
+	stats         CacheDiskStats // disk cache stats for prometheus
+	quotaPct      int            // max usage in %
 	pool          sync.Pool
 	after         int // minimum accesses before an object is cached.
 	lowWatermark  int
@@ -142,12 +144,14 @@ type diskCache struct {
 }
 
 // Inits the disk cache dir if it is not initialized already.
-func newDiskCache(dir string, quotaPct, after, lowWatermark, highWatermark int) (*diskCache, error) {
+func newDiskCache(ctx context.Context, dir string, quotaPct, after, lowWatermark, highWatermark int) (*diskCache, error) {
 	if err := os.MkdirAll(dir, 0777); err != nil {
 		return nil, fmt.Errorf("Unable to initialize '%s' dir, %w", dir, err)
 	}
 	cache := diskCache{
 		dir:           dir,
+		triggerGC:     make(chan struct{}),
+		stats:         CacheDiskStats{Dir: dir},
 		quotaPct:      quotaPct,
 		after:         after,
 		lowWatermark:  lowWatermark,
@@ -161,6 +165,8 @@ func newDiskCache(dir string, quotaPct, after, lowWatermark, highWatermark int)
 		},
 		nsMutex: newNSLock(false),
 	}
+	go cache.purgeWait(ctx)
+	cache.diskUsageHigh() // update if cache usage is already high.
 	cache.NewNSLockFn = func(ctx context.Context, cachePath string) RWLocker {
 		return cache.nsMutex.NewNSLock(ctx, nil, cachePath, "")
 	}
@@ -181,7 +187,12 @@ func (c *diskCache) diskUsageLow() bool {
 		return false
 	}
 	usedPercent := (di.Total - di.Free) * 100 / di.Total
-	return int(usedPercent) < gcStopPct
+	low := int(usedPercent) < gcStopPct
+	atomic.StoreUint64(&c.stats.UsagePercent, usedPercent)
+	if low {
+		atomic.StoreInt32(&c.stats.UsageState, 0)
+	}
+	return low
 }
 
 // Returns if the disk usage reaches high water mark w.r.t the configured cache quota.
@@ -196,7 +207,12 @@ func (c *diskCache) diskUsageHigh() bool {
 		return false
 	}
 	usedPercent := (di.Total - di.Free) * 100 / di.Total
-	return int(usedPercent) >= gcTriggerPct
+	high := int(usedPercent) >= gcTriggerPct
+	atomic.StoreUint64(&c.stats.UsagePercent, usedPercent)
+	if high {
+		atomic.StoreInt32(&c.stats.UsageState, 1)
+	}
+	return high
 }
 
 // Returns if size space can be allocated without exceeding
@@ -230,24 +246,36 @@ var (
 	errDoneForNow = errors.New("done for now")
 )
 
+func (c *diskCache) purgeWait(ctx context.Context) {
+	for {
+		select {
+		case <-ctx.Done():
+		case <-c.triggerGC: // wait here until someone triggers.
+			c.purge(ctx)
+		}
+	}
+}
+
 // Purge cache entries that were not accessed.
 func (c *diskCache) purge(ctx context.Context) {
-	if c.diskUsageLow() {
+	if atomic.LoadInt32(&c.purgeRunning) == 1 || c.diskUsageLow() {
 		return
 	}
+
 	toFree := c.toClear()
 	if toFree == 0 {
 		return
 	}
+
+	atomic.StoreInt32(&c.purgeRunning, 1) // do not run concurrent purge()
+	defer atomic.StoreInt32(&c.purgeRunning, 0)
+
 	// expiry for cleaning up old cache.json files that
 	// need to be cleaned up.
 	expiry := UTCNow().Add(-cacheExpiryDays)
 	// defaulting max hits count to 100
-	scorer, err := newFileScorer(toFree, time.Now().Unix(), 100)
+	// ignore error we know what value we are passing.
-	if err != nil {
+	scorer, _ := newFileScorer(toFree, time.Now().Unix(), 100)
-		logger.LogIf(ctx, err)
-		return
-	}
 
 	// this function returns FileInfo for cached range files and cache data file.
 	fiStatFn := func(ranges map[string]string, dataFile, pathPrefix string) map[string]os.FileInfo {
@@ -326,27 +354,20 @@ func (c *diskCache) purge(ctx context.Context) {
 		return
 	}
 
-	for _, path := range scorer.fileNames() {
+	scorer.purgeFunc(func(qfile queuedFile) {
-		removeAll(path)
+		fileName := qfile.name
-		slashIdx := strings.LastIndex(path, SlashSeparator)
+		removeAll(fileName)
-		pathPrefix := path[0:slashIdx]
+		slashIdx := strings.LastIndex(fileName, SlashSeparator)
-		fname := path[slashIdx+1:]
+		if slashIdx >= 0 {
-		if fname == cacheDataFile {
+			fileNamePrefix := fileName[0:slashIdx]
-			removeAll(pathPrefix)
+			fname := fileName[slashIdx+1:]
+			if fname == cacheDataFile {
+				removeAll(fileNamePrefix)
+			}
 		}
-	}
+	})
-}
 
-func (c *diskCache) incGCCounter() {
+	scorer.reset()
-	atomic.AddUint64(&c.gcCounter, 1)
-}
-
-func (c *diskCache) resetGCCounter() {
-	atomic.StoreUint64(&c.gcCounter, 0)
-}
-
-func (c *diskCache) gcCount() uint64 {
-	return atomic.LoadUint64(&c.gcCounter)
 }
 
 // sets cache drive status
@@ -630,7 +651,7 @@ func newCacheEncryptMetadata(bucket, object string, metadata map[string]string)
 // Caches the object to disk
 func (c *diskCache) Put(ctx context.Context, bucket, object string, data io.Reader, size int64, rs *HTTPRangeSpec, opts ObjectOptions, incHitsOnly bool) error {
 	if c.diskUsageHigh() {
-		c.incGCCounter()
+		c.triggerGC <- struct{}{}
 		io.Copy(ioutil.Discard, data)
 		return errDiskFull
 	}

cmd/disk-cache-stats.go

@@ -1,5 +1,5 @@
 /*
- * MinIO Cloud Storage, (C) 2019 MinIO, Inc.
+ * MinIO Cloud Storage, (C) 2019, 2020 MinIO, Inc.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -16,14 +16,27 @@
 
 package cmd
 
-import "sync/atomic"
+import (
+	"sync/atomic"
+)
+
+// CacheDiskStats represents cache disk statistics
+// such as current disk usage and available.
+type CacheDiskStats struct {
+	// indicates if usage is high or low, if high value is '1', if low its '0'
+	UsageState int32
+	// indicates the current usage percentage of this cache disk
+	UsagePercent uint64
+	Dir          string
+}
 
 // CacheStats - represents bytes served from cache,
 // cache hits and cache misses.
 type CacheStats struct {
-	BytesServed uint64
+	BytesServed  uint64
-	Hits        uint64
+	Hits         uint64
-	Misses      uint64
+	Misses       uint64
+	GetDiskStats func() []CacheDiskStats
 }
 
 // Increase total bytes served from cache

cmd/disk-cache-utils.go

@@ -439,6 +439,14 @@ func (f *fileScorer) fileObjInfos() []ObjectInfo {
 	return res
 }
 
+func (f *fileScorer) purgeFunc(p func(qfile queuedFile)) {
+	e := f.queue.Front()
+	for e != nil {
+		p(e.Value.(queuedFile))
+		e = e.Next()
+	}
+}
+
 // fileNames returns all queued file names.
 func (f *fileScorer) fileNames() []string {
 	res := make([]string, 0, f.queue.Len())

cmd/disk-cache.go

@@ -24,6 +24,7 @@ import (
 	"net/http"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/minio/minio/cmd/config/cache"
@@ -282,10 +283,11 @@ func (c *cacheObjects) GetObjectNInfo(ctx context.Context, bucket, object string
 
 	// Reaching here implies cache miss
 	c.cacheStats.incMiss()
+
 	// Since we got here, we are serving the request from backend,
 	// and also adding the object to the cache.
 	if dcache.diskUsageHigh() {
-		dcache.incGCCounter()
+		dcache.triggerGC <- struct{}{} // this is non-blocking
 	}
 
 	bkReader, bkErr := c.GetObjectNInfoFn(ctx, bucket, object, rs, h, lockType, opts)
@@ -544,7 +546,7 @@ func newCache(config cache.Config) ([]*diskCache, bool, error) {
 		if quota == 0 {
 			quota = config.Quota
 		}
-		cache, err := newDiskCache(dir, quota, config.After, config.WatermarkLow, config.WatermarkHigh)
+		cache, err := newDiskCache(ctx, dir, quota, config.After, config.WatermarkLow, config.WatermarkHigh)
 		if err != nil {
 			return nil, false, err
 		}
@@ -677,7 +679,17 @@ func newServerCacheObjects(ctx context.Context, config cache.Config) (CacheObjec
 			return newObjectLayerFn().CopyObject(ctx, srcBucket, srcObject, destBucket, destObject, srcInfo, srcOpts, dstOpts)
 		},
 	}
-
+	c.cacheStats.GetDiskStats = func() []CacheDiskStats {
+		cacheDiskStats := make([]CacheDiskStats, len(c.cache))
+		for i := range c.cache {
+			cacheDiskStats[i] = CacheDiskStats{
+				Dir: c.cache[i].stats.Dir,
+			}
+			atomic.StoreInt32(&cacheDiskStats[i].UsageState, atomic.LoadInt32(&c.cache[i].stats.UsageState))
+			atomic.StoreUint64(&cacheDiskStats[i].UsagePercent, atomic.LoadUint64(&c.cache[i].stats.UsagePercent))
+		}
+		return cacheDiskStats
+	}
 	if migrateSw {
 		go c.migrateCacheFromV1toV2(ctx)
 	}
@@ -697,19 +709,9 @@ func (c *cacheObjects) gc(ctx context.Context) {
 			if c.migrating {
 				continue
 			}
-			var wg sync.WaitGroup
 			for _, dcache := range c.cache {
-				if dcache.gcCount() == 0 {
+				dcache.triggerGC <- struct{}{}
-					continue
-				}
-				wg.Add(1)
-				go func(d *diskCache) {
-					defer wg.Done()
-					d.resetGCCounter()
-					d.purge(ctx)
-				}(dcache)
 			}
-			wg.Wait()
 		}
 	}
 }

cmd/metrics.go

@@ -260,6 +260,27 @@ func cacheMetricsPrometheus(ch chan<- prometheus.Metric) {
 		prometheus.CounterValue,
 		float64(cacheObjLayer.CacheStats().getBytesServed()),
 	)
+	for _, cdStats := range cacheObjLayer.CacheStats().GetDiskStats() {
+		// Cache disk usage percentage
+		ch <- prometheus.MustNewConstMetric(
+			prometheus.NewDesc(
+				prometheus.BuildFQName("cache", "usage", "percent"),
+				"Total percentage cache usage",
+				[]string{"disk"}, nil),
+			prometheus.GaugeValue,
+			float64(cdStats.UsagePercent),
+			cdStats.Dir,
+		)
+		ch <- prometheus.MustNewConstMetric(
+			prometheus.NewDesc(
+				prometheus.BuildFQName("cache", "usage", "high"),
+				"Indicates cache usage is high or low, relative to current cache 'quota' settings",
+				[]string{"disk"}, nil),
+			prometheus.GaugeValue,
+			float64(cdStats.UsageState),
+			cdStats.Dir,
+		)
+	}
 }
 
 // collects http metrics for MinIO server in Prometheus specific format

docs/metrics/prometheus/README.md

@@ -110,7 +110,7 @@ These are the new set of metrics which will be in effect after `RELEASE.2019-10-
     - Metrics that records the http statistics and latencies are labeled to their respective APIs (putobject,getobject etc).
     - Disk usage metrics are distributed and labeled to the respective disk paths.
 
-For more details, please check the `Migration guide for the new set of metrics`. 
+For more details, please check the `Migration guide for the new set of metrics`.
 
 The list of metrics and its definition are as follows. (NOTE: instance here is one MinIO node)
 
@@ -178,14 +178,23 @@ All metrics are labeled by `bucket`, each metric is displayed per bucket. `bucke
 
 MinIO Gateway instances enabled with Disk-Caching expose caching related metrics.
 
-| name                 | description                             |
+#### Global cache metrics
-|:---------------------|:----------------------------------------|
+| name                 | description                                       |
-| `cache_data_served`  | Total number of bytes served from cache |
+|:---------------------|:--------------------------------------------------|
-| `cache_hits_total`   | Total number of cache hits              |
+| `cache_hits_total`   | Total number of cache hits                        |
-| `cache_misses_total` | Total number of cache misses            |
+| `cache_misses_total` | Total number of cache misses                      |
+| `cache_data_served`  | Total number of bytes served from cache           |
 
-### Gateway & Cache specific metrics
+#### Per disk cache metrics
+| `cache_usage_percent` | Total percentage cache usage                                                     |
+| `cache_usage_state`   | Indicates cache usage is high or low, relative to current cache 'quota' settings |
 
+`cache_usage_state` holds only two states
+
+- '1' indicates high disk usage
+- '0' indicates low disk usage
+
+### Gateway specific metrics
 MinIO Gateway instance exposes metrics related to Gateway communication with the cloud backend (S3, Azure & GCS Gateway).
 
 `<gateway_type>` changes based on the gateway in use can be 's3', 'gcs' or 'azure'. Other metrics are labeled with `method` that identifies HTTP GET, HEAD, PUT and POST requests to the backend.