diff --git a/go/base/context.go b/go/base/context.go index 5a06d0d4d..4d0c3b09d 100644 --- a/go/base/context.go +++ b/go/base/context.go @@ -238,7 +238,7 @@ type MigrationContext struct { AbortError error abortMutex *sync.Mutex - Metrics metrics.MemStatsGaugeEmitter + Metrics metrics.Emitter OriginalTableColumnsOnApplier *sql.ColumnList OriginalTableColumns *sql.ColumnList diff --git a/go/logic/applier.go b/go/logic/applier.go index 3bf32e124..3d88acc0a 100644 --- a/go/logic/applier.go +++ b/go/logic/applier.go @@ -16,6 +16,7 @@ import ( "github.com/github/gh-ost/go/base" "github.com/github/gh-ost/go/binlog" + "github.com/github/gh-ost/go/metrics" "github.com/github/gh-ost/go/sql" "context" @@ -893,8 +894,10 @@ func (apl *Applier) CalculateNextIterationRangeEndValues() (hasFurtherRange bool return hasFurtherRange, err } + queryStartTime := time.Now() rows, err := apl.db.Query(query, explodedArgs...) if err != nil { + metrics.RecordQueryDuration(apl.migrationContext.Metrics, "source", "range_select", time.Since(queryStartTime), err) return hasFurtherRange, err } defer rows.Close() @@ -902,13 +905,16 @@ func (apl *Applier) CalculateNextIterationRangeEndValues() (hasFurtherRange bool iterationRangeMaxValues := sql.NewColumnValues(apl.migrationContext.UniqueKey.Len()) for rows.Next() { if err = rows.Scan(iterationRangeMaxValues.ValuesPointers...); err != nil { + metrics.RecordQueryDuration(apl.migrationContext.Metrics, "source", "range_select", time.Since(queryStartTime), err) return hasFurtherRange, err } hasFurtherRange = true } if err = rows.Err(); err != nil { + metrics.RecordQueryDuration(apl.migrationContext.Metrics, "source", "range_select", time.Since(queryStartTime), err) return hasFurtherRange, err } + metrics.RecordQueryDuration(apl.migrationContext.Metrics, "source", "range_select", time.Since(queryStartTime), nil) if hasFurtherRange { apl.migrationContext.MigrationIterationRangeMaxValues = iterationRangeMaxValues return hasFurtherRange, nil @@ -956,7 +962,9 @@ func (apl *Applier) ApplyIterationInsertQuery() (chunkSize int64, rowsAffected i if _, err := tx.Exec(sessionQuery); err != nil { return nil, err } + queryStartTime := time.Now() result, err := tx.Exec(query, explodedArgs...) + metrics.RecordQueryDuration(apl.migrationContext.Metrics, "target", "chunk_copy", time.Since(queryStartTime), err) if err != nil { return nil, err } @@ -1669,13 +1677,16 @@ func (apl *Applier) ApplyDMLEventQueries(dmlEvents [](*binlog.BinlogDMLEvent)) e // in the batch. SHOW WARNINGS only shows warnings from the last statement in a // multi-statement query, so we interleave SHOW WARNINGS after each DML statement. if apl.migrationContext.PanicOnWarnings { + queryStartTime := time.Now() totalDelta, err = apl.executeBatchWithWarningChecking(ctx, tx, buildResults) + metrics.RecordQueryDuration(apl.migrationContext.Metrics, "target", "binlog_apply", time.Since(queryStartTime), err) if err != nil { return rollback(err) } } else { // Fast path: batch together DML queries into multi-statements to minimize network trips. // We use the raw driver connection to access the rows affected for each statement. + queryStartTime := time.Now() execErr := conn.Raw(func(driverConn any) error { ex := driverConn.(driver.ExecerContext) nvc := driverConn.(driver.NamedValueChecker) @@ -1709,6 +1720,7 @@ func (apl *Applier) ApplyDMLEventQueries(dmlEvents [](*binlog.BinlogDMLEvent)) e return nil }) + metrics.RecordQueryDuration(apl.migrationContext.Metrics, "target", "binlog_apply", time.Since(queryStartTime), execErr) if execErr != nil { return rollback(execErr) } diff --git a/go/logic/inspect.go b/go/logic/inspect.go index 96aadd672..982be1d98 100644 --- a/go/logic/inspect.go +++ b/go/logic/inspect.go @@ -16,6 +16,7 @@ import ( "time" "github.com/github/gh-ost/go/base" + "github.com/github/gh-ost/go/metrics" "github.com/github/gh-ost/go/mysql" "github.com/github/gh-ost/go/sql" @@ -677,13 +678,16 @@ func (isp *Inspector) CountTableRows(ctx context.Context) error { query := fmt.Sprintf(`select /* gh-ost */ count(*) as count_rows from %s.%s`, sql.EscapeName(isp.migrationContext.DatabaseName), sql.EscapeName(isp.migrationContext.OriginalTableName)) var rowsEstimate int64 + queryStartTime := time.Now() if err := conn.QueryRowContext(ctx, query).Scan(&rowsEstimate); err != nil { + metrics.RecordQueryDuration(isp.migrationContext.Metrics, "source", "row_count", time.Since(queryStartTime), err) if errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) { isp.migrationContext.Log.Infof("exact row count cancelled (%s), likely because I'm about to cut over. I'm going to kill that query.", ctx.Err()) return mysql.Kill(isp.db, connectionID) } return err } + metrics.RecordQueryDuration(isp.migrationContext.Metrics, "source", "row_count", time.Since(queryStartTime), nil) // row count query finished. nil out the cancel func, so the main migration thread // doesn't bother calling it after row copy is done. diff --git a/go/logic/migrator.go b/go/logic/migrator.go index 5ed2e2d9a..5c51505e5 100644 --- a/go/logic/migrator.go +++ b/go/logic/migrator.go @@ -1168,7 +1168,7 @@ func (mgtr *Migrator) initiateInspector() (err error) { return nil } -// emitProgressMetrics emits StatsD gauges from a progress snapshot. +// emitProgressMetrics emits StatsD gauges func (mgtr *Migrator) emitProgressMetrics(snap migrationProgressSnapshot) { metrics.EmitProgressGauges( mgtr.migrationContext.Metrics, @@ -1181,6 +1181,13 @@ func (mgtr *Migrator) emitProgressMetrics(snap migrationProgressSnapshot) { snap.applyEventsBacklog, snap.applyEventsCapacity, ) + isThrottled, _, _ := mgtr.migrationContext.IsThrottled() + metrics.EmitLagGauges( + mgtr.migrationContext.Metrics, + snap.replicationLagSeconds, + snap.heartbeatLagSeconds, + isThrottled, + ) } // reportStatus samples progress, emits metrics, and optionally prints status output. diff --git a/go/logic/migrator_test.go b/go/logic/migrator_test.go index 31e161b5e..d939ae97d 100644 --- a/go/logic/migrator_test.go +++ b/go/logic/migrator_test.go @@ -410,11 +410,19 @@ func TestMigratorGetMigrationStateAndETA(t *testing.T) { type progressGaugeSpy struct { names []string values []float64 + tags [][]string } -func (s *progressGaugeSpy) Gauge(name string, value float64, _ ...string) { +func (s *progressGaugeSpy) Gauge(name string, value float64, tags ...string) { s.names = append(s.names, name) s.values = append(s.values, value) + s.tags = append(s.tags, append([]string(nil), tags...)) +} + +func (s *progressGaugeSpy) Count(name string, value int64, tags ...string) { +} + +func (s *progressGaugeSpy) Histogram(name string, value float64, tags ...string) { } func TestReportStatusEmitsProgressGaugesEveryTick(t *testing.T) { @@ -427,7 +435,7 @@ func TestReportStatusEmitsProgressGaugesEveryTick(t *testing.T) { migrator := NewMigrator(ctx, "test") migrator.reportStatus(NoPrintStatusRule, io.Discard) - require.Len(t, spy.names, 6) + require.Len(t, spy.names, 8) assert.Equal(t, []string{ "row_copy.rows_copied", "row_copy.rows_estimate", @@ -435,16 +443,18 @@ func TestReportStatusEmitsProgressGaugesEveryTick(t *testing.T) { "binlog.backlog_size", "binlog.backlog_capacity", "binlog.backlog_utilization", + "lag.replication_seconds", + "lag.heartbeat_seconds", }, spy.names) - assert.Equal(t, []float64{1000, 5000, 42, 0, float64(cap(migrator.applyEventsQueue)), 0}, spy.values) + assert.Equal(t, []float64{1000, 5000, 42, 0, float64(cap(migrator.applyEventsQueue)), 0}, spy.values[:6]) assert.InDelta(t, 20.0, ctx.GetProgressPct(), 0.01) spy.names = nil spy.values = nil atomic.StoreInt64(&ctx.TotalDMLEventsApplied, 100) migrator.reportStatus(NoPrintStatusRule, io.Discard) - require.Len(t, spy.names, 6) - assert.Equal(t, []float64{1000, 5000, 100, 0, float64(cap(migrator.applyEventsQueue)), 0}, spy.values) + require.Len(t, spy.names, 8) + assert.Equal(t, []float64{1000, 5000, 100, 0, float64(cap(migrator.applyEventsQueue)), 0}, spy.values[:6]) } func TestReportStatusEmitsBinlogBacklogGauges(t *testing.T) { @@ -463,7 +473,7 @@ func TestReportStatusEmitsBinlogBacklogGauges(t *testing.T) { migrator.reportStatus(NoPrintStatusRule, io.Discard) capacity := float64(cap(migrator.applyEventsQueue)) - require.Len(t, spy.names, 6) + require.Len(t, spy.names, 8) assert.Equal(t, float64(2), spy.values[3]) assert.Equal(t, capacity, spy.values[4]) assert.InDelta(t, 2/capacity, spy.values[5], 1e-9) @@ -480,7 +490,7 @@ func TestReportStatusEmitsGaugesWhenRowCopyComplete(t *testing.T) { atomic.StoreInt64(&migrator.rowCopyCompleteFlag, 1) migrator.reportStatus(NoPrintStatusRule, io.Discard) - require.Len(t, spy.names, 6) + require.Len(t, spy.names, 8) assert.Equal(t, float64(5000), spy.values[0]) assert.Equal(t, float64(5000), spy.values[1], "rows_estimate tracks rows_copied when row copy is complete") } @@ -500,8 +510,27 @@ func TestReportStatusEmitsGaugesWhenPrintSuppressed(t *testing.T) { require.False(t, migrator.shouldPrintStatus(HeuristicPrintStatusRule, snap.elapsedSeconds, etaDuration)) migrator.reportStatus(HeuristicPrintStatusRule, io.Discard) - require.Len(t, spy.names, 6) - assert.Equal(t, []float64{1000, 5000, 0, 0, float64(cap(migrator.applyEventsQueue)), 0}, spy.values) + require.Len(t, spy.names, 8) + assert.Equal(t, []float64{1000, 5000, 0, 0, float64(cap(migrator.applyEventsQueue)), 0}, spy.values[:6]) +} + +func TestReportStatusEmitsLagGaugesWhenThrottled(t *testing.T) { + spy := &progressGaugeSpy{} + ctx := base.NewMigrationContext() + ctx.Metrics = spy + ctx.SetThrottled(true, "max-lag-millis", base.NoThrottleReasonHint) + atomic.StoreInt64(&ctx.CurrentLag, int64(5*time.Second)) + ctx.SetLastHeartbeatOnChangelogTime(time.Now().Add(-4 * time.Second)) + + migrator := NewMigrator(ctx, "test") + migrator.reportStatus(NoPrintStatusRule, io.Discard) + + require.GreaterOrEqual(t, len(spy.names), 8) + assert.Equal(t, "lag.replication_seconds", spy.names[6]) + assert.Equal(t, "lag.heartbeat_seconds", spy.names[7]) + require.Len(t, spy.tags[6], 1) + assert.Equal(t, "throttled:true", spy.tags[6][0]) + assert.Equal(t, "throttled:true", spy.tags[7][0]) } func TestMigratorShouldPrintStatus(t *testing.T) { diff --git a/go/metrics/binlog_backlog.go b/go/metrics/binlog_backlog.go deleted file mode 100644 index 2d6ea5a4b..000000000 --- a/go/metrics/binlog_backlog.go +++ /dev/null @@ -1,31 +0,0 @@ -/* - Copyright 2026 GitHub Inc. - See https://github.com/github/gh-ost/blob/master/LICENSE -*/ - -package metrics - -// EmitBinlogBacklogGauges emits apply-events queue depth gauges (namespace is applied by the client): -// gh_ost.binlog.backlog_size, gh_ost.binlog.backlog_capacity, gh_ost.binlog.backlog_utilization. -func EmitBinlogBacklogGauges(emit MemStatsGaugeEmitter, backlogSize, backlogCapacity int) { - if emit == nil { - return - } - emit.Gauge("binlog.backlog_size", float64(backlogSize)) - emit.Gauge("binlog.backlog_capacity", float64(backlogCapacity)) - emit.Gauge("binlog.backlog_utilization", binlogBacklogUtilization(backlogSize, backlogCapacity)) -} - -func binlogBacklogUtilization(backlogSize, backlogCapacity int) float64 { - if backlogCapacity <= 0 { - return 0 - } - utilization := float64(backlogSize) / float64(backlogCapacity) - if utilization > 1 { - return 1 - } - if utilization < 0 { - return 0 - } - return utilization -} diff --git a/go/metrics/binlog_backlog_test.go b/go/metrics/binlog_backlog_test.go deleted file mode 100644 index 4f99dc1b1..000000000 --- a/go/metrics/binlog_backlog_test.go +++ /dev/null @@ -1,53 +0,0 @@ -/* - Copyright 2026 GitHub Inc. - See https://github.com/github/gh-ost/blob/master/LICENSE -*/ - -package metrics - -import "testing" - -func TestEmitBinlogBacklogGauges(t *testing.T) { - spy := &gaugeSpy{} - EmitBinlogBacklogGauges(spy, 250, 1000) - - wantNames := []string{ - "binlog.backlog_size", - "binlog.backlog_capacity", - "binlog.backlog_utilization", - } - wantVals := []float64{250, 1000, 0.25} - - if len(spy.names) != len(wantNames) { - t.Fatalf("got %d gauges, want %d", len(spy.names), len(wantNames)) - } - for i := range wantNames { - if spy.names[i] != wantNames[i] || spy.values[i] != wantVals[i] { - t.Fatalf("[%d] got %s=%v want %s=%v", i, spy.names[i], spy.values[i], wantNames[i], wantVals[i]) - } - } -} - -func TestEmitBinlogBacklogGauges_nilSafe(t *testing.T) { - EmitBinlogBacklogGauges(nil, 1, 2) -} - -func TestBinlogBacklogUtilization(t *testing.T) { - tests := []struct { - size, capacity int - want float64 - }{ - {0, 1000, 0}, - {250, 1000, 0.25}, - {1000, 1000, 1}, - {1500, 1000, 1}, - {-1, 1000, 0}, - {10, 0, 0}, - } - for _, tt := range tests { - got := binlogBacklogUtilization(tt.size, tt.capacity) - if got != tt.want { - t.Fatalf("utilization(%d, %d) = %v, want %v", tt.size, tt.capacity, got, tt.want) - } - } -} diff --git a/go/metrics/client.go b/go/metrics/client.go index ed6acc096..1d78df216 100644 --- a/go/metrics/client.go +++ b/go/metrics/client.go @@ -21,6 +21,13 @@ type Client struct { sd *statsd.Client } +// Emitter is implemented by *Client; used for tests without UDP. +type Emitter interface { + Gauge(name string, value float64, tags ...string) + Count(name string, value int64, tags ...string) + Histogram(name string, value float64, tags ...string) +} + // NewClient connects to addr for StatsD. If addr is empty, returns Noop and nil error. // namespace is typically "gh_ost." (metrics are named namespace + short name, e.g. gh_ost.startup). // tags are global tags applied to every metric (repeatable --statsd-tags). @@ -60,6 +67,13 @@ func (c *Client) Count(name string, value int64, tags ...string) { _ = c.sd.Count(name, value, tags, 1.0) } +func (c *Client) Histogram(name string, value float64, tags ...string) { + if c.sd == nil { + return + } + _ = c.sd.Histogram(name, value, tags, 1.0) +} + // Close flushes buffered metrics; safe for Noop. func (c *Client) Close() error { if c.sd == nil { diff --git a/go/metrics/emit.go b/go/metrics/emit.go new file mode 100644 index 000000000..8cd0add83 --- /dev/null +++ b/go/metrics/emit.go @@ -0,0 +1,128 @@ +/* + Copyright 2026 GitHub Inc. + See https://github.com/github/gh-ost/blob/master/LICENSE +*/ + +package metrics + +import ( + "context" + "fmt" + "runtime" + "time" +) + +// EmitProgressGauges emits row-copy and DML progress gauges (namespace is applied by the client): +// gh_ost.row_copy.rows_copied, gh_ost.row_copy.rows_estimate, gh_ost.dml.events_applied. +func EmitProgressGauges(emit Emitter, rowsCopied, rowsEstimate, dmlEventsApplied int64) { + if emit == nil { + return + } + emit.Gauge("row_copy.rows_copied", float64(rowsCopied)) + emit.Gauge("row_copy.rows_estimate", float64(rowsEstimate)) + emit.Gauge("dml.events_applied", float64(dmlEventsApplied)) +} + +// EmitBinlogBacklogGauges emits apply-events queue depth gauges (namespace is applied by the client): +// gh_ost.binlog.backlog_size, gh_ost.binlog.backlog_capacity, gh_ost.binlog.backlog_utilization. +func EmitBinlogBacklogGauges(emit Emitter, backlogSize, backlogCapacity int) { + if emit == nil { + return + } + emit.Gauge("binlog.backlog_size", float64(backlogSize)) + emit.Gauge("binlog.backlog_capacity", float64(backlogCapacity)) + emit.Gauge("binlog.backlog_utilization", binlogBacklogUtilization(backlogSize, backlogCapacity)) +} + +func binlogBacklogUtilization(backlogSize, backlogCapacity int) float64 { + if backlogCapacity <= 0 { + return 0 + } + utilization := float64(backlogSize) / float64(backlogCapacity) + if utilization > 1 { + return 1 + } + if utilization < 0 { + return 0 + } + return utilization +} + +// EmitLagGauges emits replication and heartbeat lag gauges (namespace is applied by the client): +// gh_ost.lag.replication_seconds, gh_ost.lag.heartbeat_seconds, each tagged throttled:true|false. +// +// These are point-in-time readings each status tick (not a distribution), so gauges are used +// rather than histograms; DogStatsD histogram aggregation exposes count/max series that do not +// match the log line lag values in Prometheus/Grafana. +func EmitLagGauges(emit Emitter, replicationLagSeconds, heartbeatLagSeconds float64, throttled bool) { + if emit == nil { + return + } + tags := []string{fmt.Sprintf("throttled:%t", throttled)} + emit.Gauge("lag.replication_seconds", replicationLagSeconds, tags...) + emit.Gauge("lag.heartbeat_seconds", heartbeatLagSeconds, tags...) +} + +// EmitGoRuntimeGauges emits gh_ost.go_runtime.* gauges (namespace is applied by the client). +// m and numGoroutine are typically from runtime.ReadMemStats and runtime.NumGoroutine. +func EmitGoRuntimeGauges(emit Emitter, m *runtime.MemStats, numGoroutine int) { + if emit == nil || m == nil { + return + } + emit.Gauge("go_runtime.alloc_bytes", float64(m.Alloc)) + emit.Gauge("go_runtime.sys_bytes", float64(m.Sys)) + emit.Gauge("go_runtime.heap_inuse_bytes", float64(m.HeapInuse)) + emit.Gauge("go_runtime.num_gc", float64(m.NumGC)) + emit.Gauge("go_runtime.gc_pause_total_ns", float64(m.PauseTotalNs)) + emit.Gauge("go_runtime.goroutines", float64(numGoroutine)) +} + +// StartGoRuntimeReporter periodically samples runtime memory and goroutines and emits gauges +// until ctx is cancelled. It is a no-op when interval <= 0, client is nil, or StatsD is disabled +// (noop client). +func StartGoRuntimeReporter(ctx context.Context, client *Client, interval time.Duration) { + if ctx == nil || client == nil || interval <= 0 || client.sd == nil { + return + } + + emit := func() { + var m runtime.MemStats + runtime.ReadMemStats(&m) + EmitGoRuntimeGauges(client, &m, runtime.NumGoroutine()) + } + + go func() { + ticker := time.NewTicker(interval) + defer ticker.Stop() + + emit() + for { + select { + case <-ctx.Done(): + return + case <-ticker.C: + emit() + } + } + }() +} + +type queryHistogramEmitter interface { + Histogram(name string, value float64, tags ...string) +} + +// RecordQueryDuration emits gh_ost.query.duration_milliseconds with side/kind/outcome tags. +func RecordQueryDuration(emit any, side string, kind string, duration time.Duration, err error) { + if emit == nil || side == "" || kind == "" || duration < 0 { + return + } + histogramEmitter, ok := emit.(queryHistogramEmitter) + if !ok { + return + } + outcome := "ok" + if err != nil { + outcome = "error" + } + histogramEmitter.Histogram("query.duration_milliseconds", float64(duration.Milliseconds()), "side:"+side, "kind:"+kind, "outcome:"+outcome) +} diff --git a/go/metrics/emit_test.go b/go/metrics/emit_test.go new file mode 100644 index 000000000..366f246bb --- /dev/null +++ b/go/metrics/emit_test.go @@ -0,0 +1,225 @@ +/* + Copyright 2026 GitHub Inc. + See https://github.com/github/gh-ost/blob/master/LICENSE +*/ + +package metrics + +import ( + "context" + "errors" + "runtime" + "slices" + "testing" + "time" +) + +type gaugeSpy struct { + names []string + values []float64 + tags [][]string +} + +func (g *gaugeSpy) Gauge(name string, value float64, tags ...string) { + g.names = append(g.names, name) + g.values = append(g.values, value) + g.tags = append(g.tags, append([]string(nil), tags...)) +} + +func (g *gaugeSpy) Count(name string, value int64, tags ...string) { +} + +func (g *gaugeSpy) Histogram(name string, value float64, tags ...string) { +} + +func TestEmitProgressGauges(t *testing.T) { + spy := &gaugeSpy{} + EmitProgressGauges(spy, 1000, 5000, 42) + + wantNames := []string{ + "row_copy.rows_copied", + "row_copy.rows_estimate", + "dml.events_applied", + } + wantVals := []float64{1000, 5000, 42} + + if len(spy.names) != len(wantNames) { + t.Fatalf("got %d gauges, want %d", len(spy.names), len(wantNames)) + } + for i := range wantNames { + if spy.names[i] != wantNames[i] || spy.values[i] != wantVals[i] { + t.Fatalf("[%d] got %s=%v want %s=%v", i, spy.names[i], spy.values[i], wantNames[i], wantVals[i]) + } + } +} + +func TestEmitProgressGauges_nilSafe(t *testing.T) { + EmitProgressGauges(nil, 1, 2, 3) +} + +func TestEmitBinlogBacklogGauges(t *testing.T) { + spy := &gaugeSpy{} + EmitBinlogBacklogGauges(spy, 250, 1000) + + wantNames := []string{ + "binlog.backlog_size", + "binlog.backlog_capacity", + "binlog.backlog_utilization", + } + wantVals := []float64{250, 1000, 0.25} + + if len(spy.names) != len(wantNames) { + t.Fatalf("got %d gauges, want %d", len(spy.names), len(wantNames)) + } + for i := range wantNames { + if spy.names[i] != wantNames[i] || spy.values[i] != wantVals[i] { + t.Fatalf("[%d] got %s=%v want %s=%v", i, spy.names[i], spy.values[i], wantNames[i], wantVals[i]) + } + } +} + +func TestEmitBinlogBacklogGauges_nilSafe(t *testing.T) { + EmitBinlogBacklogGauges(nil, 1, 2) +} + +func TestBinlogBacklogUtilization(t *testing.T) { + tests := []struct { + size, capacity int + want float64 + }{ + {0, 1000, 0}, + {250, 1000, 0.25}, + {1000, 1000, 1}, + {1500, 1000, 1}, + {-1, 1000, 0}, + {10, 0, 0}, + } + for _, tt := range tests { + got := binlogBacklogUtilization(tt.size, tt.capacity) + if got != tt.want { + t.Fatalf("utilization(%d, %d) = %v, want %v", tt.size, tt.capacity, got, tt.want) + } + } +} + +func TestEmitLagGauges_notThrottled(t *testing.T) { + spy := &gaugeSpy{} + EmitLagGauges(spy, 2.5, 1.25, false) + + wantNames := []string{"lag.replication_seconds", "lag.heartbeat_seconds"} + wantVals := []float64{2.5, 1.25} + wantTags := []string{"throttled:false"} + + if len(spy.names) != len(wantNames) { + t.Fatalf("got %d gauges, want %d", len(spy.names), len(wantNames)) + } + for i := range wantNames { + if spy.names[i] != wantNames[i] || spy.values[i] != wantVals[i] { + t.Fatalf("[%d] got %s=%v want %s=%v", i, spy.names[i], spy.values[i], wantNames[i], wantVals[i]) + } + if len(spy.tags[i]) != 1 || spy.tags[i][0] != wantTags[0] { + t.Fatalf("[%d] got tags %v want [%s]", i, spy.tags[i], wantTags[0]) + } + } +} + +func TestEmitLagGauges_throttled(t *testing.T) { + spy := &gaugeSpy{} + EmitLagGauges(spy, 4.0, 3.0, true) + + if len(spy.names) != 2 { + t.Fatalf("got %d gauges, want 2", len(spy.names)) + } + for i := range spy.names { + if len(spy.tags[i]) != 1 || spy.tags[i][0] != "throttled:true" { + t.Fatalf("[%d] got tags %v want [throttled:true]", i, spy.tags[i]) + } + } +} + +func TestEmitLagGauges_nilSafe(t *testing.T) { + EmitLagGauges(nil, 1, 2, false) +} + +func TestEmitGoRuntimeGauges(t *testing.T) { + spy := &gaugeSpy{} + m := &runtime.MemStats{ + Alloc: 100, + Sys: 200, + HeapInuse: 300, + NumGC: 7, + PauseTotalNs: 42, + } + EmitGoRuntimeGauges(spy, m, 123) + + wantNames := []string{ + "go_runtime.alloc_bytes", + "go_runtime.sys_bytes", + "go_runtime.heap_inuse_bytes", + "go_runtime.num_gc", + "go_runtime.gc_pause_total_ns", + "go_runtime.goroutines", + } + wantVals := []float64{100, 200, 300, 7, 42, 123} + + if len(spy.names) != len(wantNames) { + t.Fatalf("got %d gauges, want %d", len(spy.names), len(wantNames)) + } + for i := range wantNames { + if spy.names[i] != wantNames[i] || spy.values[i] != wantVals[i] { + t.Fatalf("[%d] got %s=%v want %s=%v", i, spy.names[i], spy.values[i], wantNames[i], wantVals[i]) + } + } +} + +func TestEmitGoRuntimeGauges_nilSafe(t *testing.T) { + EmitGoRuntimeGauges(nil, &runtime.MemStats{}, 1) + EmitGoRuntimeGauges(&gaugeSpy{}, nil, 1) +} + +func TestStartGoRuntimeReporter_stopsOnCancel(t *testing.T) { + ctx, cancel := context.WithCancel(context.Background()) + c := &Client{} // sd nil, so the reporter should not start. + StartGoRuntimeReporter(ctx, c, time.Millisecond) + cancel() + time.Sleep(20 * time.Millisecond) +} + +type histogramSpy struct { + names []string + values []float64 + tags [][]string +} + +func (h *histogramSpy) Histogram(name string, value float64, tags ...string) { + h.names = append(h.names, name) + h.values = append(h.values, value) + h.tags = append(h.tags, tags) +} + +func TestRecordQueryDuration(t *testing.T) { + spy := &histogramSpy{} + + RecordQueryDuration(spy, "source", "row_count", 1500*time.Millisecond, nil) + RecordQueryDuration(spy, "target", "binlog_apply", 2*time.Second, errors.New("boom")) + + if len(spy.names) != 2 { + t.Fatalf("got %d histograms, want 2", len(spy.names)) + } + if spy.names[0] != "query.duration_milliseconds" || spy.values[0] != 1500 { + t.Fatalf("got %s=%v, want query.duration_milliseconds=1500", spy.names[0], spy.values[0]) + } + if !slices.Equal(spy.tags[0], []string{"side:source", "kind:row_count", "outcome:ok"}) { + t.Fatalf("got tags %#v", spy.tags[0]) + } + if spy.values[1] != 2000 || !slices.Equal(spy.tags[1], []string{"side:target", "kind:binlog_apply", "outcome:error"}) { + t.Fatalf("got second metric value=%v tags=%#v", spy.values[1], spy.tags[1]) + } +} + +func TestRecordQueryDurationNilSafe(t *testing.T) { + RecordQueryDuration(nil, "source", "row_count", time.Second, nil) + RecordQueryDuration(&histogramSpy{}, "", "row_count", time.Second, nil) + RecordQueryDuration(&histogramSpy{}, "source", "", time.Second, nil) + RecordQueryDuration(&histogramSpy{}, "source", "row_count", -time.Second, nil) +} diff --git a/go/metrics/go_runtime.go b/go/metrics/go_runtime.go deleted file mode 100644 index 24ae2c6b5..000000000 --- a/go/metrics/go_runtime.go +++ /dev/null @@ -1,61 +0,0 @@ -/* - Copyright 2022 GitHub Inc. - See https://github.com/github/gh-ost/blob/master/LICENSE -*/ - -package metrics - -import ( - "context" - "runtime" - "time" -) - -// MemStatsGaugeEmitter is implemented by *Client; used for tests without UDP. -type MemStatsGaugeEmitter interface { - Gauge(name string, value float64, tags ...string) -} - -// EmitGoRuntimeGauges emits gh_ost.go_runtime.* gauges (namespace is applied by the client). -// m and numGoroutine are typically from runtime.ReadMemStats and runtime.NumGoroutine. -func EmitGoRuntimeGauges(emit MemStatsGaugeEmitter, m *runtime.MemStats, numGoroutine int) { - if emit == nil || m == nil { - return - } - emit.Gauge("go_runtime.alloc_bytes", float64(m.Alloc)) - emit.Gauge("go_runtime.sys_bytes", float64(m.Sys)) - emit.Gauge("go_runtime.heap_inuse_bytes", float64(m.HeapInuse)) - emit.Gauge("go_runtime.num_gc", float64(m.NumGC)) - emit.Gauge("go_runtime.gc_pause_total_ns", float64(m.PauseTotalNs)) - emit.Gauge("go_runtime.goroutines", float64(numGoroutine)) -} - -// StartGoRuntimeReporter periodically samples runtime memory and goroutines and emits gauges -// until ctx is cancelled. It is a no-op when interval <= 0, client is nil, or StatsD is disabled -// (noop client). -func StartGoRuntimeReporter(ctx context.Context, client *Client, interval time.Duration) { - if ctx == nil || client == nil || interval <= 0 || client.sd == nil { - return - } - - emit := func() { - var m runtime.MemStats - runtime.ReadMemStats(&m) - EmitGoRuntimeGauges(client, &m, runtime.NumGoroutine()) - } - - go func() { - ticker := time.NewTicker(interval) - defer ticker.Stop() - - emit() - for { - select { - case <-ctx.Done(): - return - case <-ticker.C: - emit() - } - } - }() -} diff --git a/go/metrics/go_runtime_test.go b/go/metrics/go_runtime_test.go deleted file mode 100644 index 24811206b..000000000 --- a/go/metrics/go_runtime_test.go +++ /dev/null @@ -1,67 +0,0 @@ -/* - Copyright 2022 GitHub Inc. - See https://github.com/github/gh-ost/blob/master/LICENSE -*/ - -package metrics - -import ( - "context" - "runtime" - "testing" - "time" -) - -type gaugeSpy struct { - names []string - values []float64 -} - -func (g *gaugeSpy) Gauge(name string, value float64, _ ...string) { - g.names = append(g.names, name) - g.values = append(g.values, value) -} - -func TestEmitGoRuntimeGauges(t *testing.T) { - spy := &gaugeSpy{} - m := &runtime.MemStats{ - Alloc: 100, - Sys: 200, - HeapInuse: 300, - NumGC: 7, - PauseTotalNs: 42, - } - EmitGoRuntimeGauges(spy, m, 123) - - wantNames := []string{ - "go_runtime.alloc_bytes", - "go_runtime.sys_bytes", - "go_runtime.heap_inuse_bytes", - "go_runtime.num_gc", - "go_runtime.gc_pause_total_ns", - "go_runtime.goroutines", - } - wantVals := []float64{100, 200, 300, 7, 42, 123} - - if len(spy.names) != len(wantNames) { - t.Fatalf("got %d gauges, want %d", len(spy.names), len(wantNames)) - } - for i := range wantNames { - if spy.names[i] != wantNames[i] || spy.values[i] != wantVals[i] { - t.Fatalf("[%d] got %s=%v want %s=%v", i, spy.names[i], spy.values[i], wantNames[i], wantVals[i]) - } - } -} - -func TestEmitGoRuntimeGauges_nilSafe(t *testing.T) { - EmitGoRuntimeGauges(nil, &runtime.MemStats{}, 1) - EmitGoRuntimeGauges(&gaugeSpy{}, nil, 1) -} - -func TestStartGoRuntimeReporter_stopsOnCancel(t *testing.T) { - ctx, cancel := context.WithCancel(context.Background()) - c := &Client{} // sd nil — should not start - StartGoRuntimeReporter(ctx, c, time.Millisecond) - cancel() - time.Sleep(20 * time.Millisecond) -} diff --git a/go/metrics/progress.go b/go/metrics/progress.go deleted file mode 100644 index b6b3361b6..000000000 --- a/go/metrics/progress.go +++ /dev/null @@ -1,17 +0,0 @@ -/* - Copyright 2022 GitHub Inc. - See https://github.com/github/gh-ost/blob/master/LICENSE -*/ - -package metrics - -// EmitProgressGauges emits row-copy and DML progress gauges (namespace is applied by the client): -// gh_ost.row_copy.rows_copied, gh_ost.row_copy.rows_estimate, gh_ost.dml.events_applied. -func EmitProgressGauges(emit MemStatsGaugeEmitter, rowsCopied, rowsEstimate, dmlEventsApplied int64) { - if emit == nil { - return - } - emit.Gauge("row_copy.rows_copied", float64(rowsCopied)) - emit.Gauge("row_copy.rows_estimate", float64(rowsEstimate)) - emit.Gauge("dml.events_applied", float64(dmlEventsApplied)) -} diff --git a/go/metrics/progress_test.go b/go/metrics/progress_test.go deleted file mode 100644 index 4a6ce5d3b..000000000 --- a/go/metrics/progress_test.go +++ /dev/null @@ -1,33 +0,0 @@ -/* - Copyright 2022 GitHub Inc. - See https://github.com/github/gh-ost/blob/master/LICENSE -*/ - -package metrics - -import "testing" - -func TestEmitProgressGauges(t *testing.T) { - spy := &gaugeSpy{} - EmitProgressGauges(spy, 1000, 5000, 42) - - wantNames := []string{ - "row_copy.rows_copied", - "row_copy.rows_estimate", - "dml.events_applied", - } - wantVals := []float64{1000, 5000, 42} - - if len(spy.names) != len(wantNames) { - t.Fatalf("got %d gauges, want %d", len(spy.names), len(wantNames)) - } - for i := range wantNames { - if spy.names[i] != wantNames[i] || spy.values[i] != wantVals[i] { - t.Fatalf("[%d] got %s=%v want %s=%v", i, spy.names[i], spy.values[i], wantNames[i], wantVals[i]) - } - } -} - -func TestEmitProgressGauges_nilSafe(t *testing.T) { - EmitProgressGauges(nil, 1, 2, 3) -}