Merge #153270 #153275 #153278 #153288

153270: sql/inspect: skip flaky statement timeout test under race r=spilchen a=spilchen

The TestInspectJobImplicitTxnSemantics statement_timeout subtest was failing intermittently under race conditions. The test expects an INSPECT job to start and succeed even when the triggering EXPERIMENTAL SCRUB statement times out after 1 second, but on slow machines with race detection enabled, the job creation may not complete before the timeout.

Fixes #153164

Release note: None

Epic: None

153275: roachtest: fix roachperf metric bugs r=golgeek a=DarrylWong

Release note: none
Epic: none
Fixes: none

153278: roachtest: fix live migration error message r=golgeek a=DarrylWong

Previously the error reporting would list host error vms, instead of live migration vms. This change fixes that.

Epic: none
Fixes: none
Release note: none

153288: roachtest: increase timeout for hibernate test r=spilchen a=spilchen

We noticed that on s390x the hibernate test can timeout after 5 hours. There are also several runs that succeeded but came close to timing out. We are going to bump the timeout to give that platform more chance to finish the test.

Fixes #153029

Release note: none

Epic: none

Co-authored-by: Matt Spilchen <[email protected]>
Co-authored-by: DarrylWong <[email protected]>

Author: 3 people

pkg/cmd/roachtest/roachtestutil/utils.go

@@ -198,7 +198,7 @@ func GetMeanOverLastN(n int, items []float64) float64 {
 // Usage: ROACHTEST_PERF_WORKLOAD_DURATION="5m".
 const EnvWorkloadDurationFlag = "ROACHTEST_PERF_WORKLOAD_DURATION"
 
-var workloadDurationRegex = regexp.MustCompile(`^\d+[mhsMHS]$`)
+var workloadDurationRegex = regexp.MustCompile(`^\d+[mhs]$`)
 
 // GetEnvWorkloadDurationValueOrDefault validates EnvWorkloadDurationFlag and
 // returns value set if valid else returns default duration.

pkg/cmd/roachtest/test_runner.go

@@ -1261,7 +1261,7 @@ func (r *testRunner) runTest(
 				if liveMigrationVMNames != "" {
 					failureMsg = fmt.Sprintf("VMs had live migrations during the test run: %s\n\n**Other Failures:**\n%s", liveMigrationVMNames, failureMsg)
 					t.resetFailures()
-					t.Error(liveMigrationError(hostErrorVMNames))
+					t.Error(liveMigrationError(liveMigrationVMNames))
 				}
 
 				output := fmt.Sprintf("%s\ntest artifacts and logs in: %s", failureMsg, t.ArtifactsDir())

pkg/cmd/roachtest/tests/cdc_bench.go

@@ -78,6 +78,9 @@ func registerCDCBench(r registry.Registry) {
 		if len(histograms.Summaries) != 1 {
 			return nil, errors.Errorf("expected exactly 1 histogram summary, got %d", len(histograms.Summaries))
 		}
+		// CDC scan tests export the scan rate (rows per second) as a single
+		// duration value in nanoseconds, set as the upper bound of the histogram.
+		// This convoluted scheme was cargo-culted from the restore test.
 		scanRate = float64(histograms.Summaries[0].HighestTrackableValue) / float64(time.Second)
 
 		// Add scan rate metric (higher is better)

pkg/cmd/roachtest/tests/hibernate.go

@@ -251,7 +251,7 @@ func registerHibernate(r registry.Registry, opt hibernateOptions) {
 		NativeLibs:       registry.LibGEOS,
 		CompatibleClouds: registry.AllExceptAWS,
 		Suites:           registry.Suites(registry.Nightly, registry.ORM),
-		Timeout:          5 * time.Hour,
+		Timeout:          6 * time.Hour,
 		Run: func(ctx context.Context, t test.Test, c cluster.Cluster) {
 			runHibernate(ctx, t, c)
 		},

pkg/sql/inspect/BUILD.bazel

@@ -79,6 +79,7 @@ go_test(
         "//pkg/sql/sessiondata",
         "//pkg/testutils",
         "//pkg/testutils/serverutils",
+        "//pkg/testutils/skip",
         "//pkg/testutils/sqlutils",
         "//pkg/testutils/testcluster",
         "//pkg/util/leaktest",

pkg/sql/inspect/inspect_job_test.go

@@ -16,6 +16,7 @@ import (
 	"github.com/cockroachdb/cockroach/pkg/sql"
 	"github.com/cockroachdb/cockroach/pkg/testutils"
 	"github.com/cockroachdb/cockroach/pkg/testutils/serverutils"
+	"github.com/cockroachdb/cockroach/pkg/testutils/skip"
 	"github.com/cockroachdb/cockroach/pkg/testutils/sqlutils"
 	"github.com/cockroachdb/cockroach/pkg/util/leaktest"
 	"github.com/cockroachdb/cockroach/pkg/util/log"
@@ -83,14 +84,18 @@ func TestInspectJobImplicitTxnSemantics(t *testing.T) {
 		onStartError      error
 		expectedErrRegex  string
 		expectedJobStatus string
+		skipUnderRace     bool
 	}{
 		{desc: "inspect success", expectedJobStatus: "succeeded"},
 		{desc: "inspect failure", onStartError: errors.Newf("inspect validation error"),
 			expectedErrRegex: "inspect validation error", expectedJobStatus: "failed"},
 		// Note: avoiding small statement timeouts, as this can impact the ability to reset.
 		{desc: "statement timeout", setupSQL: "SET statement_timeout = '1s'", tearDownSQL: "RESET statement_timeout",
-			pauseAtStart: true, expectedErrRegex: "canceled", expectedJobStatus: "succeeded"},
+			pauseAtStart: true, expectedErrRegex: "canceled", expectedJobStatus: "succeeded", skipUnderRace: true},
 	} {
+		if tc.skipUnderRace {
+			skip.UnderRace(t, "timing dependent")
+		}
 		t.Run(tc.desc, func(t *testing.T) {
 			// Run in a closure so that we run teardown before verifying job status
 			func() {