chore: fix more flaky tests on Windows with Postgres (#15629)

hugodutka · web-flow · commit 83c493e832f9 · 2024-11-26T11:56:07.000+01:00
Addresses the following flakes: - coder/internal#222 - coder/internal#223 - coder/internal#224 - coder/internal#225 - coder/internal#226 - coder/internal#227 - coder/internal#228 - coder/internal#229 - coder/internal#230
diff --git a/coderd/database/dbgen/dbgen.go b/coderd/database/dbgen/dbgen.go
@@ -220,16 +220,29 @@ func WorkspaceAgentScriptTimings(t testing.TB, db database.Store, script databas
 }
 
 func WorkspaceAgentScriptTiming(t testing.TB, db database.Store, orig database.WorkspaceAgentScriptTiming) database.WorkspaceAgentScriptTiming {
-	timing, err := db.InsertWorkspaceAgentScriptTimings(genCtx, database.InsertWorkspaceAgentScriptTimingsParams{
-		StartedAt: takeFirst(orig.StartedAt, dbtime.Now()),
-		EndedAt:   takeFirst(orig.EndedAt, dbtime.Now()),
-		Stage:     takeFirst(orig.Stage, database.WorkspaceAgentScriptTimingStageStart),
-		ScriptID:  takeFirst(orig.ScriptID, uuid.New()),
-		ExitCode:  takeFirst(orig.ExitCode, 0),
-		Status:    takeFirst(orig.Status, database.WorkspaceAgentScriptTimingStatusOk),
-	})
-	require.NoError(t, err, "insert workspace agent script")
-	return timing
+	// retry a few times in case of a unique constraint violation
+	for i := 0; i < 10; i++ {
+		timing, err := db.InsertWorkspaceAgentScriptTimings(genCtx, database.InsertWorkspaceAgentScriptTimingsParams{
+			StartedAt: takeFirst(orig.StartedAt, dbtime.Now()),
+			EndedAt:   takeFirst(orig.EndedAt, dbtime.Now()),
+			Stage:     takeFirst(orig.Stage, database.WorkspaceAgentScriptTimingStageStart),
+			ScriptID:  takeFirst(orig.ScriptID, uuid.New()),
+			ExitCode:  takeFirst(orig.ExitCode, 0),
+			Status:    takeFirst(orig.Status, database.WorkspaceAgentScriptTimingStatusOk),
+		})
+		if err == nil {
+			return timing
+		}
+		// Some tests run WorkspaceAgentScriptTiming in a loop and run into
+		// a unique violation - 2 rows get the same started_at value.
+		if (database.IsUniqueViolation(err, database.UniqueWorkspaceAgentScriptTimingsScriptIDStartedAtKey) && orig.StartedAt == time.Time{}) {
+			// Wait 1 millisecond so dbtime.Now() changes
+			time.Sleep(time.Millisecond * 1)
+			continue
+		}
+		require.NoError(t, err, "insert workspace agent script")
+	}
+	panic("failed to insert workspace agent script timing")
 }
 
 func Workspace(t testing.TB, db database.Store, orig database.WorkspaceTable) database.WorkspaceTable {
diff --git a/coderd/database/dbpurge/dbpurge_test.go b/coderd/database/dbpurge/dbpurge_test.go
@@ -54,7 +54,7 @@ func TestPurge(t *testing.T) {
 
 //nolint:paralleltest // It uses LockIDDBPurge.
 func TestDeleteOldWorkspaceAgentStats(t *testing.T) {
-	ctx, cancel := context.WithTimeout(context.Background(), testutil.WaitShort)
+	ctx, cancel := context.WithTimeout(context.Background(), testutil.WaitLong)
 	defer cancel()
 
 	now := dbtime.Now()
diff --git a/enterprise/tailnet/pgcoord_test.go b/enterprise/tailnet/pgcoord_test.go
@@ -921,7 +921,7 @@ func TestPGCoordinatorPropogatedPeerContext(t *testing.T) {
 		t.Skip("test only with postgres")
 	}
 
-	ctx := testutil.Context(t, testutil.WaitShort)
+	ctx := testutil.Context(t, testutil.WaitMedium)
 	store, ps := dbtestutil.NewDB(t)
 	logger := testutil.Logger(t)
 
diff --git a/testutil/duration_windows.go b/testutil/duration_windows.go
@@ -7,18 +7,18 @@ import "time"
 //
 // Windows durations are adjusted for slow CI workers.
 const (
-	WaitShort     = 15 * time.Second
-	WaitMedium    = 20 * time.Second
-	WaitLong      = 35 * time.Second
-	WaitSuperLong = 120 * time.Second
+	WaitShort     = 30 * time.Second
+	WaitMedium    = 40 * time.Second
+	WaitLong      = 70 * time.Second
+	WaitSuperLong = 240 * time.Second
 )
 
 // Constants for delaying repeated operations, e.g. in
 // require.Eventually.
 //
 // Windows durations are adjusted for slow CI workers.
 const (
-	IntervalFast   = 50 * time.Millisecond
-	IntervalMedium = 500 * time.Millisecond
-	IntervalSlow   = 2 * time.Second
+	IntervalFast   = 100 * time.Millisecond
+	IntervalMedium = 1000 * time.Millisecond
+	IntervalSlow   = 4 * time.Second
 )

Original file line number	Diff line number	Diff line change
`@@ -921,7 +921,7 @@ func TestPGCoordinatorPropogatedPeerContext(t *testing.T) {`
`921`	`921`	`t.Skip("test only with postgres")`
`922`	`922`	`}`
`923`	`923`
`924`		`- ctx := testutil.Context(t, testutil.WaitShort)`
	`924`	`+ ctx := testutil.Context(t, testutil.WaitMedium)`
`925`	`925`	`store, ps := dbtestutil.NewDB(t)`
`926`	`926`	`logger := testutil.Logger(t)`
`927`	`927`