Rename lru cache metrics to the same "dgraph_lru" prefix. (#2714)

* Rename lru cache metrics with a "dgraph_lru" prefix.

* Rename dgraph_evicted_lists_total to dgraph_lru_evicted_total.

* docs: Update metrics section for `dgraph_lru` prefixed metrics.

* Rename EvictedPls variable to LcacheEvicts for consistency with other Lcache vars.

* Add TODO about moving away from Prometheus's expvar collector.

* Reflow metrics docs to 100-columns.

Author: danielmai

posting/lists.go

@@ -137,7 +137,7 @@ func periodicUpdateStats(lc *y.Closer) {
 			inUse := float64(megs)
 
 			stats := lcache.Stats()
-			x.EvictedPls.Set(int64(stats.NumEvicts))
+			x.LcacheEvicts.Set(int64(stats.NumEvicts))
 			x.LcacheSize.Set(int64(stats.Size))
 			x.LcacheLen.Set(int64(stats.Length))
 
@@ -246,10 +246,10 @@ func StopLRUEviction() {
 func Get(key []byte) (rlist *List, err error) {
 	lp := lcache.Get(string(key))
 	if lp != nil {
-		x.CacheHit.Add(1)
+		x.LcacheHit.Add(1)
 		return lp, nil
 	}
-	x.CacheMiss.Add(1)
+	x.LcacheMiss.Add(1)
 
 	// Any initialization for l must be done before PutIfMissing. Once it's added
 	// to the map, any other goroutine can retrieve it.
@@ -260,7 +260,7 @@ func Get(key []byte) (rlist *List, err error) {
 	// We are always going to return lp to caller, whether it is l or not
 	lp = lcache.PutIfMissing(string(key), l)
 	if lp != l {
-		x.CacheRace.Add(1)
+		x.LcacheRace.Add(1)
 	}
 	return lp, nil
 }

wiki/content/deploy/index.md

@@ -1395,11 +1395,14 @@ Install **[Grafana](http://docs.grafana.org/installation/)** to plot the metrics
 
 ## Metrics
 
-Dgraph metrics follow the [metric and label conventions for Prometheus](https://prometheus.io/docs/practices/naming/).
+Dgraph metrics follow the [metric and label conventions for
+Prometheus](https://prometheus.io/docs/practices/naming/).
 
 ### Disk Metrics
 
-The disk metrics let you track the disk activity of the Dgraph process. Dgraph does not interact directly with the filesystem. Instead it relies on [Badger](https://github.com/dgraph-io/badger) to read from and write to disk.
+The disk metrics let you track the disk activity of the Dgraph process. Dgraph does not interact
+directly with the filesystem. Instead it relies on [Badger](https://github.com/dgraph-io/badger) to
+read from and write to disk.
 
  Metrics                          | Description
  -------                          | -----------
@@ -1413,9 +1416,12 @@ The disk metrics let you track the disk activity of the Dgraph process. Dgraph d
 
 ### Memory Metrics
 
-The memory metrics let you track the memory usage of the Dgraph process. The idle and inuse metrics gives you a better sense of the active memory usage of the Dgraph process. The process memory metric shows the memory usage as measured by the operating system.
+The memory metrics let you track the memory usage of the Dgraph process. The idle and inuse metrics
+gives you a better sense of the active memory usage of the Dgraph process. The process memory metric
+shows the memory usage as measured by the operating system.
 
-By looking at all three metrics you can see how much memory a Dgraph process is holding from the operating system and how much is actively in use.
+By looking at all three metrics you can see how much memory a Dgraph process is holding from the
+operating system and how much is actively in use.
 
  Metrics                          | Description
  -------                          | -----------
@@ -1425,21 +1431,28 @@ By looking at all three metrics you can see how much memory a Dgraph process is
 
 ### LRU Cache Metrics
 
-The LRU cache metrics let you track on how well the posting list cache is being used. You can measure the cache capacity with the max posting list size (see data metrics below), which may correlate with `dgraph_evicted_lists_total` is growing.
+The LRU cache metrics let you track on how well the posting list cache is being used.
 
- Metrics                          | Description
- -------                          | -----------
- `dgraph_cache_hits_total`        | Total number of cache hits for posting lists in Dgraph.
- `dgraph_cache_miss_total`        | Total number of cache misses for posting lists in Dgraph.
- `dgraph_cache_race_total`        | Total number of cache races when getting posting lists in Dgraph.
- `dgraph_evicted_lists_total`     | Total number of posting lists evicted from LRU cache. A large number here could indicate a large posting list.
- `dgraph_lcache_capacity_bytes`   | Current size of the LRU cache. The max value should be close to the size specified by `--lru_mb`.
- `dgraph_lcache_keys_total`       | Total number of keys in the LRU cache.
- `dgraph_lcache_size_bytes`       | Size in bytes of the LRU cache.
+You can track `dgraph_lru_capacity_bytes`, `dgraph_lru_evicted_total`, and `dgraph_max_list_bytes`
+(see the [Data Metrics]({{< relref "#data-metrics" >}})) to determine if the cache size should be
+adjusted. A high number of evictions can indicate a large posting list that repeatedly is inserted
+and evicted from the cache due to insufficient sizing. The LRU cache size can be tuned with the option
+`--lru_mb`.
+
+ Metrics                     | Description
+ -------                     | -----------
+ `dgraph_lru_hits_total`     | Total number of cache hits for posting lists in Dgraph.
+ `dgraph_lru_miss_total`     | Total number of cache misses for posting lists in Dgraph.
+ `dgraph_lru_race_total`     | Total number of cache races when getting posting lists in Dgraph.
+ `dgraph_lru_evicted_total`  | Total number of posting lists evicted from LRU cache.
+ `dgraph_lru_capacity_bytes` | Current size of the LRU cache. The max value should be close to the size specified by `--lru_mb`.
+ `dgraph_lru_keys_total`     | Total number of keys in the LRU cache.
+ `dgraph_lru_size_bytes`     | Size in bytes of the LRU cache.
 
 ### Data Metrics
 
-The data metrics let you track the [posting list]({{< ref "/design-concepts/index.md#posting-list" >}}) store.
+The data metrics let you track the [posting list]({{< ref "/design-concepts/index.md#posting-list"
+>}}) store.
 
  Metrics                          | Description
  -------                          | -----------

x/metrics.go

@@ -30,11 +30,11 @@ var (
 	PostingWrites *expvar.Int
 	BytesRead     *expvar.Int
 	BytesWrite    *expvar.Int
-	EvictedPls    *expvar.Int
 	NumQueries    *expvar.Int
-	CacheHit      *expvar.Int
-	CacheMiss     *expvar.Int
-	CacheRace     *expvar.Int
+	LcacheHit     *expvar.Int
+	LcacheMiss    *expvar.Int
+	LcacheRace    *expvar.Int
+	LcacheEvicts  *expvar.Int
 
 	// value at particular point of time
 	PendingQueries   *expvar.Int
@@ -66,24 +66,24 @@ func init() {
 	PendingProposals = expvar.NewInt("dgraph_pending_proposals_total")
 	BytesRead = expvar.NewInt("dgraph_read_bytes_total")
 	BytesWrite = expvar.NewInt("dgraph_written_bytes_total")
-	EvictedPls = expvar.NewInt("dgraph_evicted_lists_total")
 	PendingQueries = expvar.NewInt("dgraph_pending_queries_total")
 	NumQueries = expvar.NewInt("dgraph_num_queries_total")
 	AlphaHealth = expvar.NewInt("dgraph_alpha_health_status")
 	DirtyMapSize = expvar.NewInt("dgraph_dirtymap_keys_total")
-	LcacheSize = expvar.NewInt("dgraph_lcache_size_bytes")
-	LcacheLen = expvar.NewInt("dgraph_lcache_keys_total")
-	LcacheCapacity = expvar.NewInt("dgraph_lcache_capacity_bytes")
 	NumGoRoutines = expvar.NewInt("dgraph_goroutines_total")
 	MemoryInUse = expvar.NewInt("dgraph_memory_inuse_bytes")
 	MemoryIdle = expvar.NewInt("dgraph_memory_idle_bytes")
 	MemoryProc = expvar.NewInt("dgraph_memory_proc_bytes")
 	ActiveMutations = expvar.NewInt("dgraph_active_mutations_total")
 	PredicateStats = expvar.NewMap("dgraph_predicate_stats")
 	Conf = expvar.NewMap("dgraph_config")
-	CacheHit = expvar.NewInt("dgraph_cache_hits_total")
-	CacheMiss = expvar.NewInt("dgraph_cache_miss_total")
-	CacheRace = expvar.NewInt("dgraph_cache_race_total")
+	LcacheHit = expvar.NewInt("dgraph_lru_hits_total")
+	LcacheMiss = expvar.NewInt("dgraph_lru_miss_total")
+	LcacheRace = expvar.NewInt("dgraph_lru_race_total")
+	LcacheEvicts = expvar.NewInt("dgraph_lru_evicted_total")
+	LcacheSize = expvar.NewInt("dgraph_lru_size_bytes")
+	LcacheLen = expvar.NewInt("dgraph_lru_keys_total")
+	LcacheCapacity = expvar.NewInt("dgraph_lru_capacity_bytes")
 	MaxPlSize = expvar.NewInt("dgraph_max_list_bytes")
 	MaxPlLength = expvar.NewInt("dgraph_max_list_length")
 
@@ -102,20 +102,42 @@ func init() {
 		}
 	}()
 
+	// TODO: prometheus.NewExpvarCollector is not production worthy (see godocs). Use a better
+	// way for exporting Prometheus metrics (like an OpenCensus metrics exporter).
 	expvarCollector := prometheus.NewExpvarCollector(map[string]*prometheus.Desc{
-		"dgraph_cache_hits_total": prometheus.NewDesc(
-			"dgraph_cache_hits_total",
-			"dgraph_cache_hits_total",
+		"dgraph_lru_hits_total": prometheus.NewDesc(
+			"dgraph_lru_hits_total",
+			"dgraph_lru_hits_total",
 			nil, nil,
 		),
-		"dgraph_cache_miss_total": prometheus.NewDesc(
-			"dgraph_cache_miss_total",
-			"dgraph_cache_miss_total",
+		"dgraph_lru_miss_total": prometheus.NewDesc(
+			"dgraph_lru_miss_total",
+			"dgraph_lru_miss_total",
 			nil, nil,
 		),
-		"dgraph_cache_race_total": prometheus.NewDesc(
-			"dgraph_cache_race_total",
-			"dgraph_cache_race_total",
+		"dgraph_lru_race_total": prometheus.NewDesc(
+			"dgraph_lru_race_total",
+			"dgraph_lru_race_total",
+			nil, nil,
+		),
+		"dgraph_lru_evicted_total": prometheus.NewDesc(
+			"dgraph_lru_evicted_total",
+			"dgraph_lru_evicted_total",
+			nil, nil,
+		),
+		"dgraph_lru_size_bytes": prometheus.NewDesc(
+			"dgraph_lru_size_bytes",
+			"dgraph_lru_size_bytes",
+			nil, nil,
+		),
+		"dgraph_lru_keys_total": prometheus.NewDesc(
+			"dgraph_lru_keys_total",
+			"dgraph_lru_keys_total",
+			nil, nil,
+		),
+		"dgraph_lru_capacity_bytes": prometheus.NewDesc(
+			"dgraph_lru_capacity_bytes",
+			"dgraph_lru_capacity_bytes",
 			nil, nil,
 		),
 		"dgraph_posting_reads_total": prometheus.NewDesc(
@@ -153,11 +175,6 @@ func init() {
 			"dgraph_written_bytes_total",
 			nil, nil,
 		),
-		"dgraph_evicted_lists_total": prometheus.NewDesc(
-			"dgraph_evicted_lists_total",
-			"dgraph_evicted_lists_total",
-			nil, nil,
-		),
 		"dgraph_pending_queries_total": prometheus.NewDesc(
 			"dgraph_pending_queries_total",
 			"dgraph_pending_queries_total",
@@ -178,21 +195,6 @@ func init() {
 			"dgraph_dirtymap_keys_total",
 			nil, nil,
 		),
-		"dgraph_lcache_size_bytes": prometheus.NewDesc(
-			"dgraph_lcache_size_bytes",
-			"dgraph_lcache_size_bytes",
-			nil, nil,
-		),
-		"dgraph_lcache_keys_total": prometheus.NewDesc(
-			"dgraph_lcache_keys_total",
-			"dgraph_lcache_keys_total",
-			nil, nil,
-		),
-		"dgraph_lcache_capacity_bytes": prometheus.NewDesc(
-			"dgraph_lcache_capacity_bytes",
-			"dgraph_lcache_capacity_bytes",
-			nil, nil,
-		),
 		"dgraph_goroutines_total": prometheus.NewDesc(
 			"dgraph_goroutines_total",
 			"dgraph_goroutines_total",