Merge pull request #1532 from ydb-platform/ci-run-slo-on-push-master

Run SLO Tests on push master

Author: asmyasnikov

.github/workflows/slo.yml

@@ -1,6 +1,9 @@
 name: slo
 
 on:
+  push:
+    branches:
+      - master
   pull_request:
     branches:
       - master
@@ -81,8 +84,11 @@ jobs:
           ./tests/slo/.bin/${{matrix.sdk.id}}_linux_amd64 create grpc://localhost:2135 /Root/testdb
           ./tests/slo/.bin/${{matrix.sdk.id}}_linux_amd64 run grpc://localhost:2135 /Root/testdb \
             -prom-pgw localhost:9091 \
-            -report-period 500 \
+            -report-period 250 \
             -time ${{inputs.slo_workload_duration_seconds || 600}} \
             -read-rps ${{inputs.slo_workload_read_max_rps || 1000}} \
-            -write-rps ${{inputs.slo_workload_write_max_rps || 100}}
+            -write-rps ${{inputs.slo_workload_write_max_rps || 100}} \
+            -read-timeout 10000 \
+            -write-timeout 10000 \
+            -shutdown-time 30
           ./tests/slo/.bin/${{matrix.sdk.id}}_linux_amd64 cleanup grpc://localhost:2135 /Root/testdb

tests/slo/internal/metrics/metrics.go

@@ -32,7 +32,6 @@ type (
 		retriesFailureTotal *prometheus.CounterVec
 
 		pendingOperations *prometheus.GaugeVec
-		// TODO:
 		// sdk_cpu_usage_seconds_total *prometheus.CounterVec
 		// sdk_memory_usage_bytes *prometheus.GaugeVec
 		// sdk_connections_open *prometheus.GaugeVec
@@ -54,39 +53,39 @@ func New(url, ref, label, jobName string) (*Metrics, error) {
 	}
 
 	m.errorsTotal = prometheus.NewCounterVec(
-		prometheus.CounterOpts{ //nolint:promlinter
+		prometheus.CounterOpts{
 			Name: "sdk_errors_total",
 			Help: "Total number of errors encountered, categorized by error type.",
 		},
 		[]string{"error_type"},
 	)
 
 	m.operationsTotal = prometheus.NewCounterVec(
-		prometheus.CounterOpts{ //nolint:promlinter
+		prometheus.CounterOpts{
 			Name: "sdk_operations_total",
 			Help: "Total number of operations, categorized by type attempted by the SDK.",
 		},
 		[]string{"operation_type"},
 	)
 
 	m.operationsSuccessTotal = prometheus.NewCounterVec(
-		prometheus.CounterOpts{ //nolint:promlinter
+		prometheus.CounterOpts{
 			Name: "sdk_operations_success_total",
 			Help: "Total number of successful operations, categorized by type.",
 		},
 		[]string{"operation_type"},
 	)
 
 	m.operationsFailureTotal = prometheus.NewCounterVec(
-		prometheus.CounterOpts{ //nolint:promlinter
+		prometheus.CounterOpts{
 			Name: "sdk_operations_failure_total",
 			Help: "Total number of failed operations, categorized by type.",
 		},
 		[]string{"operation_type"},
 	)
 
 	m.operationLatencySeconds = prometheus.NewHistogramVec(
-		prometheus.HistogramOpts{ //nolint:promlinter
+		prometheus.HistogramOpts{
 			Name: "sdk_operation_latency_seconds",
 			Help: "Latency of operations performed by the SDK in seconds, categorized by type and status.",
 			Buckets: []float64{
@@ -109,31 +108,31 @@ func New(url, ref, label, jobName string) (*Metrics, error) {
 	)
 
 	m.retryAttemptsTotal = prometheus.NewCounterVec(
-		prometheus.CounterOpts{ //nolint:promlinter
+		prometheus.CounterOpts{
 			Name: "sdk_retry_attempts_total",
 			Help: "Total number of retry attempts, categorized by operation type.",
 		},
 		[]string{"operation_type"},
 	)
 
 	m.retriesSuccessTotal = prometheus.NewCounterVec(
-		prometheus.CounterOpts{ //nolint:promlinter
+		prometheus.CounterOpts{
 			Name: "sdk_retries_success_total",
 			Help: "Total number of successful retries, categorized by operation type.",
 		},
 		[]string{"operation_type"},
 	)
 
 	m.retriesFailureTotal = prometheus.NewCounterVec(
-		prometheus.CounterOpts{ //nolint:promlinter
+		prometheus.CounterOpts{
 			Name: "sdk_retries_failure_total",
 			Help: "Total number of failed retries, categorized by operation type.",
 		},
 		[]string{"operation_type"},
 	)
 
 	m.pendingOperations = prometheus.NewGaugeVec(
-		prometheus.GaugeOpts{ //nolint:promlinter
+		prometheus.GaugeOpts{
 			Name: "sdk_pending_operations",
 			Help: "Current number of pending operations, categorized by type.",
 		},