Added filtering for incorrect slo boosting logs

Author: chrisThePattyEater

pkg/gce-cloud-provider/compute/gce.go

@@ -21,6 +21,7 @@ import (
 	"net/http"
 	"net/url"
 	"os"
+	"regexp"
 	"runtime"
 	"sync"
 	"time"
@@ -441,3 +442,27 @@ func IsGCENotFoundError(err error) bool {
 func IsGCEInvalidError(err error) bool {
 	return IsGCEError(err, "invalid")
 }
+
+// ErrorContainsRegex checks if the error message contains a specific regex pattern.
+func ErrorContainsRegex(err error, regexPattern string) bool {
+	if err == nil {
+		return false
+	}
+	re, compileErr := regexp.Compile(regexPattern)
+	if compileErr != nil {
+		fmt.Printf("Error compiling regex '%s': %v\n", regexPattern, compileErr)
+		return false
+	}
+	errorMessage := err.Error()
+	return re.MatchString(errorMessage)
+}
+
+// IsSnapshotAlreadyExistsError checks if the error is a snapshot already exists error.
+func IsSnapshotAlreadyExistsError(err error) bool {
+	return ErrorContainsRegex(err, "The resource '[^']+' already exists")
+}
+
+// IsGCPOrgViolationError checks if the error is a GCP organization policy violation error.
+func IsGCPOrgViolationError(err error) bool {
+	return ErrorContainsRegex(err, "violates constraint constraints/gcp.")
+}

pkg/gce-cloud-provider/compute/gce_test.go

@@ -30,6 +30,8 @@ import (
 
 	"google.golang.org/api/compute/v1"
 	"google.golang.org/api/googleapi"
+	"google.golang.org/grpc/codes"
+	"google.golang.org/grpc/status"
 )
 
 type mockTokenSource struct{}
@@ -101,6 +103,114 @@ func TestIsGCEError(t *testing.T) {
 	}
 }
 
+func TestErrorContainsRegex(t *testing.T) {
+	testCases := []struct {
+		name           string
+		inputErr       error
+		regex          string
+		expectedResult bool
+	}{
+		{
+			name:           "regex found in error message",
+			inputErr:       errors.New("The resource 'my-resource' already exists"),
+			regex:          "The resource '[^']+' already exists",
+			expectedResult: true,
+		},
+		{
+			name:           "wrapped valid error message",
+			inputErr:       status.Errorf(codes.AlreadyExists, "Snapshot already exists: %v", errors.New("The resource 'my-resource' already exists")),
+			regex:          "The resource '[^']+' already exists",
+			expectedResult: true,
+		},
+		{
+			name:           "regex not found in error message",
+			inputErr:       errors.New("The resource 'my-resource' does not exist"),
+			regex:          "The resource '[^']+' already exists",
+			expectedResult: false,
+		},
+		{
+			name:           "empty",
+			inputErr:       errors.New("  "),
+			regex:          "The resource '[^']+' already exists",
+			expectedResult: false,
+		},
+		{
+			name:           "empty 2",
+			inputErr:       errors.New(""),
+			regex:          "The resource '[^']+' already exists",
+			expectedResult: false,
+		},
+		{
+			name:           "nil",
+			inputErr:       nil,
+			regex:          "The resource '[^']+' already exists",
+			expectedResult: false,
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Logf("Running test: %v", tc.name)
+		result := ErrorContainsRegex(tc.inputErr, tc.regex)
+		if tc.expectedResult != result {
+			t.Fatalf("Got '%t', expected '%t'", result, tc.expectedResult)
+		}
+	}
+}
+
+func TestErrorIsGCPViolationRegex(t *testing.T) {
+	testCases := []struct {
+		name           string
+		inputErr       error
+		expectedResult bool
+	}{
+		{
+			name:           "is gcp org violation error",
+			inputErr:       errors.New("Your api request violates constraint constraints/gcp.resourceLocations"),
+			expectedResult: true,
+		},
+		{
+			name:           "is not gcp org violation error",
+			inputErr:       errors.New("Some incorrect error message"),
+			expectedResult: false,
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Logf("Running test: %v", tc.name)
+		result := IsGCPOrgViolationError(tc.inputErr)
+		if tc.expectedResult != result {
+			t.Fatalf("Got '%t', expected '%t'", result, tc.expectedResult)
+		}
+	}
+}
+
+func TestErrorIsSnapshotExistsError(t *testing.T) {
+	testCases := []struct {
+		name           string
+		inputErr       error
+		expectedResult bool
+	}{
+		{
+			name:           "is gcp org violation error",
+			inputErr:       errors.New("Your api request violates constraint constraints/gcp.resourceLocations"),
+			expectedResult: true,
+		},
+		{
+			name:           "is not gcp org violation error",
+			inputErr:       errors.New("Some incorrect error message"),
+			expectedResult: false,
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Logf("Running test: %v", tc.name)
+		result := IsGCPOrgViolationError(tc.inputErr)
+		if tc.expectedResult != result {
+			t.Fatalf("Got '%t', expected '%t'", result, tc.expectedResult)
+		}
+	}
+}
+
 func TestGetComputeVersion(t *testing.T) {
 	testCases := []struct {
 		name               string

pkg/gce-pd-csi-driver/controller.go

@@ -1668,6 +1668,16 @@ func (gceCS *GCEControllerServer) createPDSnapshot(ctx context.Context, project
 			if gce.IsGCEError(err, "notFound") {
 				return nil, status.Errorf(codes.NotFound, "Could not find volume with ID %v: %v", volKey.String(), err.Error())
 			}
+
+			// Identified as incorrect error handling in https://b.corp.google.com/issues/398008774
+			if gce.IsSnapshotAlreadyExistsError(err) {
+				return nil, status.Errorf(codes.AlreadyExists, "Snapshot already exists: %v", err.Error())
+			}
+
+			// Identified as incorrect error handling in https://b.corp.google.com/issues/398008774
+			if gce.IsGCPOrgViolationError(err) {
+				return nil, status.Errorf(codes.FailedPrecondition, "Violates GCP org policy: %v", err.Error())
+			}
 			return nil, common.LoggedError("Failed to create snapshot: ", err)
 		}
 	}