// MakeMemMetrics instantiates the metric objects for an SQL endpoint.
func MakeMemMetrics(endpoint string) MemoryMetrics {
prefix := "sql.mon." + endpoint
MetaMemMaxBytes := metric.Metadata{Name: prefix + ".max"}
MetaMemCurBytes := metric.Metadata{Name: prefix + ".cur"}
MetaMemMaxTxnBytes := metric.Metadata{Name: prefix + ".txn.max"}
MetaMemTxnCurBytes := metric.Metadata{Name: prefix + ".txn.cur"}
MetaMemMaxSessionBytes := metric.Metadata{Name: prefix + ".session.max"}
MetaMemSessionCurBytes := metric.Metadata{Name: prefix + ".session.cur"}
return MemoryMetrics{
MaxBytesHist: metric.NewHistogram(MetaMemMaxBytes, time.Minute, log10int64times1000, 3),
CurBytesCount: metric.NewCounter(MetaMemCurBytes),
TxnMaxBytesHist: metric.NewHistogram(MetaMemMaxTxnBytes, time.Minute, log10int64times1000, 3),
TxnCurBytesCount: metric.NewCounter(MetaMemTxnCurBytes),
SessionMaxBytesHist: metric.NewHistogram(MetaMemMaxSessionBytes, time.Minute, log10int64times1000, 3),
SessionCurBytesCount: metric.NewCounter(MetaMemSessionCurBytes),
}
}
// MakeTxnMetrics returns a TxnMetrics struct that contains metrics whose
// windowed portions retain data for approximately sampleInterval.
func MakeTxnMetrics(sampleInterval time.Duration) TxnMetrics {
return TxnMetrics{
Aborts: metric.NewCounterWithRates(metaAbortsRates),
Commits: metric.NewCounterWithRates(metaCommitsRates),
Commits1PC: metric.NewCounterWithRates(metaCommits1PCRates),
Abandons: metric.NewCounterWithRates(metaAbandonsRates),
Durations: metric.NewLatency(metaDurationsHistograms, sampleInterval),
Restarts: metric.NewHistogram(metaRestartsHistogram, sampleInterval, 100, 3),
}
}
// TestMetricsRecorder verifies that the metrics recorder properly formats the
// statistics from various registries, both for Time Series and for Status
// Summaries.
func TestMetricsRecorder(t *testing.T) {
defer leaktest.AfterTest(t)()
// ========================================
// Construct a series of fake descriptors for use in test.
// ========================================
nodeDesc := roachpb.NodeDescriptor{
NodeID: roachpb.NodeID(1),
}
storeDesc1 := roachpb.StoreDescriptor{
StoreID: roachpb.StoreID(1),
Capacity: roachpb.StoreCapacity{
Capacity: 100,
Available: 50,
},
}
storeDesc2 := roachpb.StoreDescriptor{
StoreID: roachpb.StoreID(2),
Capacity: roachpb.StoreCapacity{
Capacity: 200,
Available: 75,
},
}
// ========================================
// Create registries and add them to the recorder (two node-level, two
// store-level).
// ========================================
reg1 := metric.NewRegistry()
store1 := fakeStore{
storeID: roachpb.StoreID(1),
desc: storeDesc1,
registry: metric.NewRegistry(),
}
store2 := fakeStore{
storeID: roachpb.StoreID(2),
desc: storeDesc2,
registry: metric.NewRegistry(),
}
manual := hlc.NewManualClock(100)
recorder := NewMetricsRecorder(hlc.NewClock(manual.UnixNano, time.Nanosecond))
recorder.AddStore(store1)
recorder.AddStore(store2)
recorder.AddNode(reg1, nodeDesc, 50)
// Ensure the metric system's view of time does not advance during this test
// as the test expects time to not advance too far which would age the actual
// data (e.g. in histogram's) unexpectedly.
defer metric.TestingSetNow(func() time.Time {
return time.Unix(0, manual.UnixNano()).UTC()
})()
// ========================================
// Generate Metrics Data & Expected Results
// ========================================
// Flatten the four registries into an array for ease of use.
regList := []struct {
reg *metric.Registry
prefix string
source int64
isNode bool
}{
{
reg: reg1,
prefix: "one.",
source: 1,
isNode: true,
},
{
reg: reg1,
prefix: "two.",
source: 1,
isNode: true,
},
{
reg: store1.registry,
prefix: "",
source: int64(store1.storeID),
isNode: false,
},
{
reg: store2.registry,
prefix: "",
source: int64(store2.storeID),
isNode: false,
},
}
// Every registry will have a copy of the following metrics.
metricNames := []struct {
name string
typ string
val int64
}{
{"testGauge", "gauge", 20},
{"testGaugeFloat64", "floatgauge", 20},
{"testCounter", "counter", 5},
{"testCounterWithRates", "counterwithrates", 2},
{"testHistogram", "histogram", 10},
{"testLatency", "latency", 10},
// Stats needed for store summaries.
{"ranges", "counter", 1},
{"replicas.leaders", "gauge", 1},
{"replicas.leaseholders", "gauge", 1},
{"ranges", "gauge", 1},
{"ranges.available", "gauge", 1},
}
// Add the metrics to each registry and set their values. At the same time,
// generate expected time series results and status summary metric values.
var expected []tspb.TimeSeriesData
expectedNodeSummaryMetrics := make(map[string]float64)
expectedStoreSummaryMetrics := make(map[string]float64)
// addExpected generates expected data for a single metric data point.
addExpected := func(prefix, name string, source, time, val int64, isNode bool) {
// Generate time series data.
tsPrefix := "cr.node."
if !isNode {
tsPrefix = "cr.store."
}
expect := tspb.TimeSeriesData{
Name: tsPrefix + prefix + name,
Source: strconv.FormatInt(source, 10),
Datapoints: []tspb.TimeSeriesDatapoint{
{
TimestampNanos: time,
Value: float64(val),
},
},
}
expected = append(expected, expect)
// Generate status summary data.
if isNode {
expectedNodeSummaryMetrics[prefix+name] = float64(val)
} else {
// This can overwrite the previous value, but this is expected as
// all stores in our tests have identical values; when comparing
// status summaries, the same map is used as expected data for all
// stores.
expectedStoreSummaryMetrics[prefix+name] = float64(val)
}
}
for _, reg := range regList {
for _, data := range metricNames {
switch data.typ {
case "gauge":
g := metric.NewGauge(metric.Metadata{Name: reg.prefix + data.name})
reg.reg.AddMetric(g)
g.Update(data.val)
addExpected(reg.prefix, data.name, reg.source, 100, data.val, reg.isNode)
case "floatgauge":
g := metric.NewGaugeFloat64(metric.Metadata{Name: reg.prefix + data.name})
reg.reg.AddMetric(g)
g.Update(float64(data.val))
addExpected(reg.prefix, data.name, reg.source, 100, data.val, reg.isNode)
case "counter":
c := metric.NewCounter(metric.Metadata{Name: reg.prefix + data.name})
reg.reg.AddMetric(c)
c.Inc((data.val))
addExpected(reg.prefix, data.name, reg.source, 100, data.val, reg.isNode)
case "counterwithrates":
r := metric.NewCounterWithRates(metric.Metadata{Name: reg.prefix + data.name})
reg.reg.AddMetric(r)
r.Inc(data.val)
addExpected(reg.prefix, data.name, reg.source, 100, data.val, reg.isNode)
case "histogram":
h := metric.NewHistogram(metric.Metadata{Name: reg.prefix + data.name}, time.Second, 1000, 2)
reg.reg.AddMetric(h)
h.RecordValue(data.val)
for _, q := range recordHistogramQuantiles {
addExpected(reg.prefix, data.name+q.suffix, reg.source, 100, data.val, reg.isNode)
}
case "latency":
l := metric.NewLatency(metric.Metadata{Name: reg.prefix + data.name}, time.Hour)
reg.reg.AddMetric(l)
l.RecordValue(data.val)
// Latency is simply three histograms (at different resolution
// time scales).
for _, q := range recordHistogramQuantiles {
addExpected(reg.prefix, data.name+q.suffix, reg.source, 100, data.val, reg.isNode)
}
default:
t.Fatalf("unexpected: %+v", data)
}
}
}
// ========================================
// Verify time series data
// ========================================
actual := recorder.GetTimeSeriesData()
// Actual comparison is simple: sort the resulting arrays by time and name,
// and use reflect.DeepEqual.
sort.Sort(byTimeAndName(actual))
sort.Sort(byTimeAndName(expected))
if a, e := actual, expected; !reflect.DeepEqual(a, e) {
t.Errorf("recorder did not yield expected time series collection; diff:\n %v", pretty.Diff(e, a))
}
// ========================================
// Verify node summary generation
// ========================================
expectedNodeSummary := &NodeStatus{
Desc: nodeDesc,
BuildInfo: build.GetInfo(),
StartedAt: 50,
UpdatedAt: 100,
Metrics: expectedNodeSummaryMetrics,
StoreStatuses: []StoreStatus{
{
Desc: storeDesc1,
Metrics: expectedStoreSummaryMetrics,
},
{
Desc: storeDesc2,
Metrics: expectedStoreSummaryMetrics,
},
},
}
nodeSummary := recorder.GetStatusSummary()
if nodeSummary == nil {
t.Fatalf("recorder did not return nodeSummary")
}
sort.Sort(byStoreDescID(nodeSummary.StoreStatuses))
if a, e := nodeSummary, expectedNodeSummary; !reflect.DeepEqual(a, e) {
t.Errorf("recorder did not produce expected NodeSummary; diff:\n %v", pretty.Diff(e, a))
}
}
func newStoreMetrics(sampleInterval time.Duration) *StoreMetrics {
storeRegistry := metric.NewRegistry()
sm := &StoreMetrics{
registry: storeRegistry,
// Replica metrics.
ReplicaCount: metric.NewCounter(metaReplicaCount),
ReservedReplicaCount: metric.NewCounter(metaReservedReplicaCount),
RaftLeaderCount: metric.NewGauge(metaRaftLeaderCount),
RaftLeaderNotLeaseHolderCount: metric.NewGauge(metaRaftLeaderNotLeaseHolderCount),
LeaseHolderCount: metric.NewGauge(metaLeaseHolderCount),
QuiescentCount: metric.NewGauge(metaQuiescentCount),
// Replica CommandQueue metrics.
MaxCommandQueueSize: metric.NewGauge(metaMaxCommandQueueSize),
MaxCommandQueueWriteCount: metric.NewGauge(metaMaxCommandQueueWriteCount),
MaxCommandQueueReadCount: metric.NewGauge(metaMaxCommandQueueReadCount),
MaxCommandQueueTreeSize: metric.NewGauge(metaMaxCommandQueueTreeSize),
MaxCommandQueueOverlaps: metric.NewGauge(metaMaxCommandQueueOverlaps),
CombinedCommandQueueSize: metric.NewGauge(metaCombinedCommandQueueSize),
CombinedCommandWriteCount: metric.NewGauge(metaCombinedCommandWriteCount),
CombinedCommandReadCount: metric.NewGauge(metaCombinedCommandReadCount),
// Range metrics.
RangeCount: metric.NewGauge(metaRangeCount),
UnavailableRangeCount: metric.NewGauge(metaUnavailableRangeCount),
UnderReplicatedRangeCount: metric.NewGauge(metaUnderReplicatedRangeCount),
// Lease request metrics.
LeaseRequestSuccessCount: metric.NewCounter(metaLeaseRequestSuccessCount),
LeaseRequestErrorCount: metric.NewCounter(metaLeaseRequestErrorCount),
// Storage metrics.
LiveBytes: metric.NewGauge(metaLiveBytes),
KeyBytes: metric.NewGauge(metaKeyBytes),
ValBytes: metric.NewGauge(metaValBytes),
IntentBytes: metric.NewGauge(metaIntentBytes),
LiveCount: metric.NewGauge(metaLiveCount),
KeyCount: metric.NewGauge(metaKeyCount),
ValCount: metric.NewGauge(metaValCount),
IntentCount: metric.NewGauge(metaIntentCount),
IntentAge: metric.NewGauge(metaIntentAge),
GcBytesAge: metric.NewGauge(metaGcBytesAge),
LastUpdateNanos: metric.NewGauge(metaLastUpdateNanos),
Capacity: metric.NewGauge(metaCapacity),
Available: metric.NewGauge(metaAvailable),
Reserved: metric.NewCounter(metaReserved),
SysBytes: metric.NewGauge(metaSysBytes),
SysCount: metric.NewGauge(metaSysCount),
// RocksDB metrics.
RdbBlockCacheHits: metric.NewGauge(metaRdbBlockCacheHits),
RdbBlockCacheMisses: metric.NewGauge(metaRdbBlockCacheMisses),
RdbBlockCacheUsage: metric.NewGauge(metaRdbBlockCacheUsage),
RdbBlockCachePinnedUsage: metric.NewGauge(metaRdbBlockCachePinnedUsage),
RdbBloomFilterPrefixChecked: metric.NewGauge(metaRdbBloomFilterPrefixChecked),
RdbBloomFilterPrefixUseful: metric.NewGauge(metaRdbBloomFilterPrefixUseful),
RdbMemtableHits: metric.NewGauge(metaRdbMemtableHits),
RdbMemtableMisses: metric.NewGauge(metaRdbMemtableMisses),
RdbMemtableTotalSize: metric.NewGauge(metaRdbMemtableTotalSize),
RdbFlushes: metric.NewGauge(metaRdbFlushes),
RdbCompactions: metric.NewGauge(metaRdbCompactions),
RdbTableReadersMemEstimate: metric.NewGauge(metaRdbTableReadersMemEstimate),
RdbReadAmplification: metric.NewGauge(metaRdbReadAmplification),
RdbNumSSTables: metric.NewGauge(metaRdbNumSSTables),
// Range event metrics.
RangeSplits: metric.NewCounter(metaRangeSplits),
RangeAdds: metric.NewCounter(metaRangeAdds),
RangeRemoves: metric.NewCounter(metaRangeRemoves),
RangeSnapshotsGenerated: metric.NewCounter(metaRangeSnapshotsGenerated),
RangeSnapshotsNormalApplied: metric.NewCounter(metaRangeSnapshotsNormalApplied),
RangeSnapshotsPreemptiveApplied: metric.NewCounter(metaRangeSnapshotsPreemptiveApplied),
// Raft processing metrics.
RaftTicks: metric.NewCounter(metaRaftTicks),
RaftWorkingDurationNanos: metric.NewCounter(metaRaftWorkingDurationNanos),
RaftTickingDurationNanos: metric.NewCounter(metaRaftTickingDurationNanos),
// Raft message metrics.
RaftRcvdMsgProp: metric.NewCounter(metaRaftRcvdProp),
RaftRcvdMsgApp: metric.NewCounter(metaRaftRcvdApp),
RaftRcvdMsgAppResp: metric.NewCounter(metaRaftRcvdAppResp),
RaftRcvdMsgVote: metric.NewCounter(metaRaftRcvdVote),
RaftRcvdMsgVoteResp: metric.NewCounter(metaRaftRcvdVoteResp),
RaftRcvdMsgPreVote: metric.NewCounter(metaRaftRcvdPreVote),
RaftRcvdMsgPreVoteResp: metric.NewCounter(metaRaftRcvdPreVoteResp),
RaftRcvdMsgSnap: metric.NewCounter(metaRaftRcvdSnap),
RaftRcvdMsgHeartbeat: metric.NewCounter(metaRaftRcvdHeartbeat),
RaftRcvdMsgHeartbeatResp: metric.NewCounter(metaRaftRcvdHeartbeatResp),
RaftRcvdMsgTransferLeader: metric.NewCounter(metaRaftRcvdTransferLeader),
RaftRcvdMsgTimeoutNow: metric.NewCounter(metaRaftRcvdTimeoutNow),
RaftRcvdMsgDropped: metric.NewCounter(metaRaftRcvdDropped),
raftRcvdMessages: make(map[raftpb.MessageType]*metric.Counter, len(raftpb.MessageType_name)),
RaftEnqueuedPending: metric.NewGauge(metaRaftEnqueuedPending),
// This Gauge measures the number of heartbeats queued up just before
// the queue is cleared, to avoid flapping wildly.
RaftCoalescedHeartbeatsPending: metric.NewGauge(metaRaftCoalescedHeartbeatsPending),
// Replica queue metrics.
GCQueueSuccesses: metric.NewCounter(metaGCQueueSuccesses),
GCQueueFailures: metric.NewCounter(metaGCQueueFailures),
GCQueuePending: metric.NewGauge(metaGCQueuePending),
GCQueueProcessingNanos: metric.NewCounter(metaGCQueueProcessingNanos),
RaftLogQueueSuccesses: metric.NewCounter(metaRaftLogQueueSuccesses),
RaftLogQueueFailures: metric.NewCounter(metaRaftLogQueueFailures),
RaftLogQueuePending: metric.NewGauge(metaRaftLogQueuePending),
RaftLogQueueProcessingNanos: metric.NewCounter(metaRaftLogQueueProcessingNanos),
ConsistencyQueueSuccesses: metric.NewCounter(metaConsistencyQueueSuccesses),
ConsistencyQueueFailures: metric.NewCounter(metaConsistencyQueueFailures),
ConsistencyQueuePending: metric.NewGauge(metaConsistencyQueuePending),
ConsistencyQueueProcessingNanos: metric.NewCounter(metaConsistencyQueueProcessingNanos),
ReplicaGCQueueSuccesses: metric.NewCounter(metaReplicaGCQueueSuccesses),
ReplicaGCQueueFailures: metric.NewCounter(metaReplicaGCQueueFailures),
ReplicaGCQueuePending: metric.NewGauge(metaReplicaGCQueuePending),
ReplicaGCQueueProcessingNanos: metric.NewCounter(metaReplicaGCQueueProcessingNanos),
ReplicateQueueSuccesses: metric.NewCounter(metaReplicateQueueSuccesses),
ReplicateQueueFailures: metric.NewCounter(metaReplicateQueueFailures),
ReplicateQueuePending: metric.NewGauge(metaReplicateQueuePending),
ReplicateQueueProcessingNanos: metric.NewCounter(metaReplicateQueueProcessingNanos),
ReplicateQueuePurgatory: metric.NewGauge(metaReplicateQueuePurgatory),
SplitQueueSuccesses: metric.NewCounter(metaSplitQueueSuccesses),
SplitQueueFailures: metric.NewCounter(metaSplitQueueFailures),
SplitQueuePending: metric.NewGauge(metaSplitQueuePending),
SplitQueueProcessingNanos: metric.NewCounter(metaSplitQueueProcessingNanos),
TimeSeriesMaintenanceQueueSuccesses: metric.NewCounter(metaTimeSeriesMaintenanceQueueFailures),
TimeSeriesMaintenanceQueueFailures: metric.NewCounter(metaTimeSeriesMaintenanceQueueSuccesses),
TimeSeriesMaintenanceQueuePending: metric.NewGauge(metaTimeSeriesMaintenanceQueuePending),
TimeSeriesMaintenanceQueueProcessingNanos: metric.NewCounter(metaTimeSeriesMaintenanceQueueProcessingNanos),
// GCInfo cumulative totals.
GCNumKeysAffected: metric.NewCounter(metaGCNumKeysAffected),
GCIntentsConsidered: metric.NewCounter(metaGCIntentsConsidered),
GCIntentTxns: metric.NewCounter(metaGCIntentTxns),
GCTransactionSpanScanned: metric.NewCounter(metaGCTransactionSpanScanned),
GCTransactionSpanGCAborted: metric.NewCounter(metaGCTransactionSpanGCAborted),
GCTransactionSpanGCCommitted: metric.NewCounter(metaGCTransactionSpanGCCommitted),
GCTransactionSpanGCPending: metric.NewCounter(metaGCTransactionSpanGCPending),
GCAbortSpanScanned: metric.NewCounter(metaGCAbortSpanScanned),
GCAbortSpanConsidered: metric.NewCounter(metaGCAbortSpanConsidered),
GCAbortSpanGCNum: metric.NewCounter(metaGCAbortSpanGCNum),
GCPushTxn: metric.NewCounter(metaGCPushTxn),
GCResolveTotal: metric.NewCounter(metaGCResolveTotal),
GCResolveSuccess: metric.NewCounter(metaGCResolveSuccess),
// Mutex timing.
//
// TODO(tschottdorf): Histograms don't work very well as they were
// inherently built in a windowed (i.e. events-discarding) way, which
// is not at all the correct way. Discard at one-minute interval which
// gives sane (though mathematically nonsensical) results when exposed
// at the moment.
MuReplicaNanos: metric.NewHistogram(
metaMuReplicaNanos, sampleInterval,
time.Second.Nanoseconds(), 1,
),
MuCommandQueueNanos: metric.NewHistogram(
metaMuCommandQueueNanos, sampleInterval,
time.Second.Nanoseconds(), 1,
),
MuRaftNanos: metric.NewHistogram(
metaMuRaftNanos, sampleInterval,
time.Second.Nanoseconds(), 1,
),
MuStoreNanos: metric.NewHistogram(
metaMuStoreNanos, sampleInterval,
time.Second.Nanoseconds(), 1,
),
MuSchedulerNanos: metric.NewHistogram(
metaMuSchedulerNanos, time.Minute,
time.Second.Nanoseconds(), 1,
),
}
sm.raftRcvdMessages[raftpb.MsgProp] = sm.RaftRcvdMsgProp
sm.raftRcvdMessages[raftpb.MsgApp] = sm.RaftRcvdMsgApp
sm.raftRcvdMessages[raftpb.MsgAppResp] = sm.RaftRcvdMsgAppResp
sm.raftRcvdMessages[raftpb.MsgVote] = sm.RaftRcvdMsgVote
sm.raftRcvdMessages[raftpb.MsgVoteResp] = sm.RaftRcvdMsgVoteResp
sm.raftRcvdMessages[raftpb.MsgPreVote] = sm.RaftRcvdMsgPreVote
sm.raftRcvdMessages[raftpb.MsgPreVoteResp] = sm.RaftRcvdMsgPreVoteResp
sm.raftRcvdMessages[raftpb.MsgSnap] = sm.RaftRcvdMsgSnap
sm.raftRcvdMessages[raftpb.MsgHeartbeat] = sm.RaftRcvdMsgHeartbeat
sm.raftRcvdMessages[raftpb.MsgHeartbeatResp] = sm.RaftRcvdMsgHeartbeatResp
sm.raftRcvdMessages[raftpb.MsgTransferLeader] = sm.RaftRcvdMsgTransferLeader
sm.raftRcvdMessages[raftpb.MsgTimeoutNow] = sm.RaftRcvdMsgTimeoutNow
storeRegistry.AddMetricStruct(sm)
return sm
}