Provisioning: Mark repository as unhealthy if hooks fail (#109788)

This commit is contained in:
Roberto Jiménez Sánchez
2025-08-21 08:32:23 +00:00
committed by GitHub
parent 03d9ec4ea3
commit 61d137992b
22 changed files with 1578 additions and 490 deletions
@@ -242,10 +242,22 @@ type RepositoryStatus struct {
Webhook *WebhookStatus `json:"webhook"`
}
// HealthFailureType represents different types of repository failures
// +enum
type HealthFailureType string
const (
HealthFailureHook HealthFailureType = "hook"
HealthFailureHealth HealthFailureType = "health"
)
type HealthStatus struct {
// When not healthy, requests will not be executed
Healthy bool `json:"healthy"`
// The type of the error
Error HealthFailureType `json:"error,omitempty"`
// When the health was checked last time
Checked int64 `json:"checked,omitempty"`
@@ -571,6 +571,14 @@ func schema_pkg_apis_provisioning_v0alpha1_HealthStatus(ref common.ReferenceCall
Format: "",
},
},
"error": {
SchemaProps: spec.SchemaProps{
Description: "The type of the error\n\nPossible enum values:\n - `\"health\"`\n - `\"hook\"`",
Type: []string{"string"},
Format: "",
Enum: []interface{}{"health", "hook"},
},
},
"checked": {
SchemaProps: spec.SchemaProps{
Description: "When the health was checked last time",
@@ -4,12 +4,17 @@
package v0alpha1
import (
provisioningv0alpha1 "github.com/grafana/grafana/apps/provisioning/pkg/apis/provisioning/v0alpha1"
)
// HealthStatusApplyConfiguration represents a declarative configuration of the HealthStatus type for use
// with apply.
type HealthStatusApplyConfiguration struct {
Healthy *bool `json:"healthy,omitempty"`
Checked *int64 `json:"checked,omitempty"`
Message []string `json:"message,omitempty"`
Healthy *bool `json:"healthy,omitempty"`
Error *provisioningv0alpha1.HealthFailureType `json:"error,omitempty"`
Checked *int64 `json:"checked,omitempty"`
Message []string `json:"message,omitempty"`
}
// HealthStatusApplyConfiguration constructs a declarative configuration of the HealthStatus type for use with
@@ -26,6 +31,14 @@ func (b *HealthStatusApplyConfiguration) WithHealthy(value bool) *HealthStatusAp
return b
}
// WithError sets the Error field in the declarative configuration to the given value
// and returns the receiver, so that objects can be built by chaining "With" function invocations.
// If called multiple times, the Error field is set to the value of the last call.
func (b *HealthStatusApplyConfiguration) WithError(value provisioningv0alpha1.HealthFailureType) *HealthStatusApplyConfiguration {
b.Error = &value
return b
}
// WithChecked sets the Checked field in the declarative configuration to the given value
// and returns the receiver, so that objects can be built by chaining "With" function invocations.
// If called multiple times, the Checked field is set to the value of the last call.
@@ -0,0 +1,217 @@
package controller
import (
"context"
"fmt"
"time"
provisioning "github.com/grafana/grafana/apps/provisioning/pkg/apis/provisioning/v0alpha1"
"github.com/grafana/grafana/pkg/registry/apis/provisioning/repository"
)
// StatusPatcher defines the interface for updating repository status
//
//go:generate mockery --name=StatusPatcher
type StatusPatcher interface {
Patch(ctx context.Context, repo *provisioning.Repository, patchOperations ...map[string]interface{}) error
}
// HealthChecker provides unified health checking for repositories
type HealthChecker struct {
tester RepositoryTester
statusPatcher StatusPatcher
}
// RepositoryTester defines the interface for testing repository connectivity
//
//go:generate mockery --name=RepositoryTester
type RepositoryTester interface {
TestRepository(ctx context.Context, repo repository.Repository) (*provisioning.TestResults, error)
}
// NewHealthChecker creates a new health checker
func NewHealthChecker(tester RepositoryTester, statusPatcher StatusPatcher) *HealthChecker {
return &HealthChecker{
tester: tester,
statusPatcher: statusPatcher,
}
}
// ShouldCheckHealth determines if a repository health check should be performed
func (hc *HealthChecker) ShouldCheckHealth(repo *provisioning.Repository) bool {
// If the repository has been updated, run the health check
if repo.Generation != repo.Status.ObservedGeneration {
return true
}
// If the repository has a hook error, don't run the health check
if repo.Status.Health.Error == provisioning.HealthFailureHook {
return false
}
// Check general timing for health checks
return !hc.hasRecentHealthCheck(repo.Status.Health)
}
// hasRecentHealthCheck checks if a health check was performed recently (for timing purposes)
func (hc *HealthChecker) hasRecentHealthCheck(healthStatus provisioning.HealthStatus) bool {
if healthStatus.Checked == 0 {
return false // Never checked
}
age := time.Since(time.UnixMilli(healthStatus.Checked))
if healthStatus.Healthy {
return age <= time.Minute*5 // Recent if checked within 5 minutes when healthy
}
return age <= time.Minute // Recent if checked within 1 minute when unhealthy
}
// HasRecentFailure checks if there's a recent failure of a specific type
func (hc *HealthChecker) HasRecentFailure(healthStatus provisioning.HealthStatus, failureType provisioning.HealthFailureType) bool {
if healthStatus.Checked == 0 || healthStatus.Healthy || healthStatus.Error != failureType {
return false // No failure of this type
}
age := time.Since(time.UnixMilli(healthStatus.Checked))
return age <= time.Minute // Recent if within 1 minute
}
// RecordFailureAndUpdate records a failure and updates the repository status
func (hc *HealthChecker) RecordFailure(ctx context.Context, failureType provisioning.HealthFailureType, err error, repo *provisioning.Repository) error {
// Create the health status with the failure
healthStatus := hc.recordFailure(failureType, err)
// Create patch operation
patchOp := map[string]interface{}{
"op": "replace",
"path": "/status/health",
"value": healthStatus,
}
// Apply the patch
return hc.statusPatcher.Patch(ctx, repo, patchOp)
}
// recordFailure creates a health status with a specific failure
func (hc *HealthChecker) recordFailure(failureType provisioning.HealthFailureType, err error) provisioning.HealthStatus {
return provisioning.HealthStatus{
Healthy: false,
Error: failureType,
Checked: time.Now().UnixMilli(),
Message: []string{err.Error()},
}
}
// hasHealthStatusChanged checks if the health status has meaningfully changed
func (hc *HealthChecker) hasHealthStatusChanged(old, new provisioning.HealthStatus) bool {
if old.Healthy != new.Healthy {
return true
}
if len(old.Message) != len(new.Message) {
return true
}
if old.Checked != new.Checked {
return true
}
for i, oldMsg := range old.Message {
if i >= len(new.Message) || oldMsg != new.Message[i] {
return true
}
}
return false
}
// RefreshHealth performs a health check on an existing repository,
// updates its status if needed, and returns the test results
func (hc *HealthChecker) RefreshHealth(ctx context.Context, repo repository.Repository) (*provisioning.TestResults, provisioning.HealthStatus, error) {
cfg := repo.Config()
// Use health checker to perform comprehensive health check with existing status
testResults, newHealthStatus, err := hc.refreshHealth(ctx, repo, cfg.Status.Health)
if err != nil {
return nil, provisioning.HealthStatus{}, fmt.Errorf("health check failed: %w", err)
}
// Only update if health status actually changed
if hc.hasHealthStatusChanged(cfg.Status.Health, newHealthStatus) {
patchOp := map[string]interface{}{
"op": "replace",
"path": "/status/health",
"value": newHealthStatus,
}
if err := hc.statusPatcher.Patch(ctx, cfg, patchOp); err != nil {
return testResults, newHealthStatus, fmt.Errorf("update health status: %w", err)
}
}
return testResults, newHealthStatus, nil
}
// RefreshTimestamp updates the health status timestamp without changing other fields
func (hc *HealthChecker) RefreshTimestamp(ctx context.Context, repo *provisioning.Repository) error {
// Update the timestamp on the existing health status
healthStatus := repo.Status.Health
healthStatus.Checked = time.Now().UnixMilli()
// Create patch operation
patchOp := map[string]interface{}{
"op": "replace",
"path": "/status/health",
"value": healthStatus,
}
// Apply the patch
return hc.statusPatcher.Patch(ctx, repo, patchOp)
}
// refreshHealth performs a comprehensive health check
// Returns test results, health status, and any error
func (hc *HealthChecker) refreshHealth(ctx context.Context, repo repository.Repository, existingStatus provisioning.HealthStatus) (*provisioning.TestResults, provisioning.HealthStatus, error) {
res, err := hc.tester.TestRepository(ctx, repo)
if err != nil {
return nil, existingStatus, fmt.Errorf("failed to test repository: %w", err)
}
if !res.Success {
// Build error messages
var errorMsgs []string
for _, testErr := range res.Errors {
if testErr.Detail != "" {
errorMsgs = append(errorMsgs, testErr.Detail)
}
}
healthStatus := provisioning.HealthStatus{
Healthy: false,
Error: provisioning.HealthFailureHealth,
Checked: time.Now().UnixMilli(),
Message: errorMsgs,
}
return res, healthStatus, nil
}
// Health check succeeded
now := time.Now()
healthStatus := provisioning.HealthStatus{
Healthy: true,
Checked: now.UnixMilli(),
}
// If the existing status is already healthy with no error messages and
// the last check was recent (within 30 seconds), preserve the existing timestamp
// to avoid unnecessary updates
if existingStatus.Healthy && existingStatus.Error == "" && len(existingStatus.Message) == 0 {
lastCheckedTime := time.UnixMilli(existingStatus.Checked)
if now.Sub(lastCheckedTime) < 30*time.Second {
healthStatus.Checked = existingStatus.Checked
}
}
return res, healthStatus, nil
}
@@ -0,0 +1,592 @@
package controller
import (
"context"
"errors"
"testing"
"time"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/mock"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/util/validation/field"
provisioning "github.com/grafana/grafana/apps/provisioning/pkg/apis/provisioning/v0alpha1"
"github.com/grafana/grafana/pkg/registry/apis/provisioning/controller/mocks"
)
func TestNewHealthChecker(t *testing.T) {
mockTester := mocks.NewRepositoryTester(t)
mockPatcher := mocks.NewStatusPatcher(t)
hc := NewHealthChecker(mockTester, mockPatcher)
assert.NotNil(t, hc)
assert.Equal(t, mockTester, hc.tester)
assert.Equal(t, mockPatcher, hc.statusPatcher)
}
func TestShouldCheckHealth(t *testing.T) {
tests := []struct {
name string
repo *provisioning.Repository
expected bool
}{
{
name: "should check when generation differs",
repo: &provisioning.Repository{
ObjectMeta: metav1.ObjectMeta{Generation: 2},
Status: provisioning.RepositoryStatus{
ObservedGeneration: 1,
Health: provisioning.HealthStatus{
Healthy: true,
Checked: time.Now().UnixMilli(),
},
},
},
expected: true,
},
{
name: "should not check when hook error exists",
repo: &provisioning.Repository{
ObjectMeta: metav1.ObjectMeta{Generation: 1},
Status: provisioning.RepositoryStatus{
ObservedGeneration: 1,
Health: provisioning.HealthStatus{
Healthy: false,
Error: provisioning.HealthFailureHook,
Checked: time.Now().UnixMilli(),
},
},
},
expected: false,
},
{
name: "should not check when health check is recent and healthy",
repo: &provisioning.Repository{
ObjectMeta: metav1.ObjectMeta{Generation: 1},
Status: provisioning.RepositoryStatus{
ObservedGeneration: 1,
Health: provisioning.HealthStatus{
Healthy: true,
Checked: time.Now().Add(-time.Minute * 3).UnixMilli(), // 3 minutes ago
},
},
},
expected: false,
},
{
name: "should check when health check is old and healthy",
repo: &provisioning.Repository{
ObjectMeta: metav1.ObjectMeta{Generation: 1},
Status: provisioning.RepositoryStatus{
ObservedGeneration: 1,
Health: provisioning.HealthStatus{
Healthy: true,
Checked: time.Now().Add(-time.Minute * 6).UnixMilli(), // 6 minutes ago
},
},
},
expected: true,
},
{
name: "should not check when health error is recent",
repo: &provisioning.Repository{
ObjectMeta: metav1.ObjectMeta{Generation: 1},
Status: provisioning.RepositoryStatus{
ObservedGeneration: 1,
Health: provisioning.HealthStatus{
Healthy: false,
Error: provisioning.HealthFailureHealth,
Checked: time.Now().Add(-time.Second * 30).UnixMilli(), // 30 seconds ago
},
},
},
expected: false,
},
{
name: "should check when health error is old",
repo: &provisioning.Repository{
ObjectMeta: metav1.ObjectMeta{Generation: 1},
Status: provisioning.RepositoryStatus{
ObservedGeneration: 1,
Health: provisioning.HealthStatus{
Healthy: false,
Error: provisioning.HealthFailureHealth,
Checked: time.Now().Add(-time.Minute * 2).UnixMilli(), // 2 minutes ago
},
},
},
expected: true,
},
{
name: "should check when never checked",
repo: &provisioning.Repository{
ObjectMeta: metav1.ObjectMeta{Generation: 1},
Status: provisioning.RepositoryStatus{
ObservedGeneration: 1,
Health: provisioning.HealthStatus{
Checked: 0, // Never checked
},
},
},
expected: true,
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
mockTester := mocks.NewRepositoryTester(t)
mockPatcher := mocks.NewStatusPatcher(t)
hc := NewHealthChecker(mockTester, mockPatcher)
result := hc.ShouldCheckHealth(tt.repo)
assert.Equal(t, tt.expected, result)
})
}
}
func TestHasRecentFailure(t *testing.T) {
tests := []struct {
name string
healthStatus provisioning.HealthStatus
failureType provisioning.HealthFailureType
expected bool
}{
{
name: "no recent failure when never checked",
healthStatus: provisioning.HealthStatus{
Checked: 0,
},
failureType: provisioning.HealthFailureHook,
expected: false,
},
{
name: "recent hook failure",
healthStatus: provisioning.HealthStatus{
Healthy: false,
Error: provisioning.HealthFailureHook,
Checked: time.Now().Add(-time.Second * 30).UnixMilli(),
},
failureType: provisioning.HealthFailureHook,
expected: true,
},
{
name: "old hook failure",
healthStatus: provisioning.HealthStatus{
Healthy: false,
Error: provisioning.HealthFailureHook,
Checked: time.Now().Add(-time.Minute * 2).UnixMilli(),
},
failureType: provisioning.HealthFailureHook,
expected: false,
},
{
name: "recent health failure",
healthStatus: provisioning.HealthStatus{
Healthy: false,
Error: provisioning.HealthFailureHealth,
Checked: time.Now().Add(-time.Second * 30).UnixMilli(),
},
failureType: provisioning.HealthFailureHealth,
expected: true,
},
{
name: "old health failure",
healthStatus: provisioning.HealthStatus{
Healthy: false,
Error: provisioning.HealthFailureHealth,
Checked: time.Now().Add(-time.Minute * 2).UnixMilli(),
},
failureType: provisioning.HealthFailureHealth,
expected: false,
},
{
name: "wrong failure type",
healthStatus: provisioning.HealthStatus{
Healthy: false,
Error: provisioning.HealthFailureHook,
Checked: time.Now().Add(-time.Second * 30).UnixMilli(),
},
failureType: provisioning.HealthFailureHealth,
expected: false,
},
{
name: "healthy status with wrong failure type",
healthStatus: provisioning.HealthStatus{
Healthy: true,
Checked: time.Now().Add(-time.Second * 30).UnixMilli(),
},
failureType: provisioning.HealthFailureHook,
expected: false,
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
mockTester := mocks.NewRepositoryTester(t)
mockPatcher := mocks.NewStatusPatcher(t)
hc := NewHealthChecker(mockTester, mockPatcher)
result := hc.HasRecentFailure(tt.healthStatus, tt.failureType)
assert.Equal(t, tt.expected, result)
})
}
}
func TestRecordFailure(t *testing.T) {
tests := []struct {
name string
failureType provisioning.HealthFailureType
err error
patchError error
expectError bool
}{
{
name: "successful hook failure record",
failureType: provisioning.HealthFailureHook,
err: errors.New("hook failed"),
patchError: nil,
expectError: false,
},
{
name: "successful health failure record",
failureType: provisioning.HealthFailureHealth,
err: errors.New("health check failed"),
patchError: nil,
expectError: false,
},
{
name: "patch failure",
failureType: provisioning.HealthFailureHook,
err: errors.New("hook failed"),
patchError: errors.New("patch failed"),
expectError: true,
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
mockTester := mocks.NewRepositoryTester(t)
mockPatcher := mocks.NewStatusPatcher(t)
hc := NewHealthChecker(mockTester, mockPatcher)
repo := &provisioning.Repository{
Status: provisioning.RepositoryStatus{
Health: provisioning.HealthStatus{
Healthy: true,
Checked: time.Now().Add(-time.Hour).UnixMilli(),
},
},
}
if tt.patchError != nil {
mockPatcher.On("Patch", mock.Anything, repo, mock.AnythingOfType("map[string]interface {}")).
Return(tt.patchError)
} else {
mockPatcher.On("Patch", mock.Anything, repo, mock.AnythingOfType("map[string]interface {}")).
Return(nil).
Run(func(args mock.Arguments) {
patchOp := args[2].(map[string]interface{})
assert.Equal(t, "replace", patchOp["op"])
assert.Equal(t, "/status/health", patchOp["path"])
healthStatus := patchOp["value"].(provisioning.HealthStatus)
assert.False(t, healthStatus.Healthy)
assert.Equal(t, tt.failureType, healthStatus.Error)
assert.Contains(t, healthStatus.Message, tt.err.Error())
assert.Greater(t, healthStatus.Checked, int64(0))
})
}
err := hc.RecordFailure(context.Background(), tt.failureType, tt.err, repo)
if tt.expectError {
assert.Error(t, err)
} else {
assert.NoError(t, err)
}
mockPatcher.AssertExpectations(t)
})
}
}
func TestRecordFailureFunction(t *testing.T) {
mockTester := mocks.NewRepositoryTester(t)
mockPatcher := mocks.NewStatusPatcher(t)
hc := NewHealthChecker(mockTester, mockPatcher)
testErr := errors.New("test error")
result := hc.recordFailure(provisioning.HealthFailureHook, testErr)
assert.False(t, result.Healthy)
assert.Equal(t, provisioning.HealthFailureHook, result.Error)
assert.Equal(t, []string{"test error"}, result.Message)
assert.Greater(t, result.Checked, int64(0))
}
func TestRefreshHealth(t *testing.T) {
tests := []struct {
name string
testResult *provisioning.TestResults
testError error
patchError error
existingStatus provisioning.HealthStatus
expectError bool
expectedHealth bool
expectPatch bool
}{
{
name: "successful health check",
testResult: &provisioning.TestResults{
Success: true,
Code: 200,
},
testError: nil,
existingStatus: provisioning.HealthStatus{
Healthy: false,
Error: provisioning.HealthFailureHealth,
Checked: time.Now().Add(-time.Hour).UnixMilli(),
},
expectError: false,
expectedHealth: true,
expectPatch: true,
},
{
name: "failed health check",
testResult: &provisioning.TestResults{
Success: false,
Code: 500,
Errors: []provisioning.ErrorDetails{
{Detail: "connection failed"},
{Detail: "timeout"},
},
},
testError: nil,
existingStatus: provisioning.HealthStatus{
Healthy: true,
Checked: time.Now().Add(-time.Hour).UnixMilli(),
},
expectError: false,
expectedHealth: false,
expectPatch: true,
},
{
name: "test repository error",
testResult: nil,
testError: errors.New("repository test failed"),
existingStatus: provisioning.HealthStatus{
Healthy: true,
Checked: time.Now().Add(-time.Hour).UnixMilli(),
},
expectError: true,
expectedHealth: false,
expectPatch: false,
},
{
name: "no status change - no patch needed (recent check)",
testResult: &provisioning.TestResults{
Success: true,
Code: 200,
},
testError: nil,
existingStatus: provisioning.HealthStatus{
Healthy: true,
Checked: time.Now().Add(-15 * time.Second).UnixMilli(),
},
expectError: false,
expectedHealth: true,
expectPatch: false,
},
{
name: "status unchanged but timestamp needs update (old check)",
testResult: &provisioning.TestResults{
Success: true,
Code: 200,
},
testError: nil,
existingStatus: provisioning.HealthStatus{
Healthy: true,
Checked: time.Now().Add(-time.Hour).UnixMilli(),
},
expectError: false,
expectedHealth: true,
expectPatch: true,
},
{
name: "patch error",
testResult: &provisioning.TestResults{
Success: false,
Code: 500,
Errors: []provisioning.ErrorDetails{
{Detail: "connection failed"},
},
},
testError: nil,
patchError: errors.New("patch failed"),
existingStatus: provisioning.HealthStatus{
Healthy: true,
Checked: time.Now().Add(-time.Hour).UnixMilli(),
},
expectError: true,
expectedHealth: false,
expectPatch: true,
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
mockTester := mocks.NewRepositoryTester(t)
mockPatcher := mocks.NewStatusPatcher(t)
mockRepo := &mockRepository{
config: &provisioning.Repository{
Status: provisioning.RepositoryStatus{
Health: tt.existingStatus,
},
},
}
hc := NewHealthChecker(mockTester, mockPatcher)
if tt.testError != nil {
mockTester.On("TestRepository", mock.Anything, mockRepo).Return(tt.testResult, tt.testError)
} else {
mockTester.On("TestRepository", mock.Anything, mockRepo).Return(tt.testResult, nil)
}
if tt.expectPatch {
if tt.patchError != nil {
mockPatcher.On("Patch", mock.Anything, mockRepo.config, mock.AnythingOfType("map[string]interface {}")).
Return(tt.patchError)
} else {
mockPatcher.On("Patch", mock.Anything, mockRepo.config, mock.AnythingOfType("map[string]interface {}")).
Return(nil).
Run(func(args mock.Arguments) {
patchOp := args[2].(map[string]interface{})
assert.Equal(t, "replace", patchOp["op"])
assert.Equal(t, "/status/health", patchOp["path"])
healthStatus := patchOp["value"].(provisioning.HealthStatus)
assert.Equal(t, tt.expectedHealth, healthStatus.Healthy)
})
}
}
testResult, healthStatus, err := hc.RefreshHealth(context.Background(), mockRepo)
if tt.expectError {
assert.Error(t, err)
} else {
assert.NoError(t, err)
assert.Equal(t, tt.expectedHealth, healthStatus.Healthy)
if tt.testResult != nil {
assert.Equal(t, tt.testResult, testResult)
}
}
mockTester.AssertExpectations(t)
if tt.expectPatch {
mockPatcher.AssertExpectations(t)
}
})
}
}
func TestHasHealthStatusChanged(t *testing.T) {
tests := []struct {
name string
old provisioning.HealthStatus
new provisioning.HealthStatus
expected bool
}{
{
name: "healthy status changed",
old: provisioning.HealthStatus{
Healthy: true,
Message: []string{},
},
new: provisioning.HealthStatus{
Healthy: false,
Message: []string{"error"},
},
expected: true,
},
{
name: "different message count",
old: provisioning.HealthStatus{
Healthy: false,
Message: []string{"error1"},
},
new: provisioning.HealthStatus{
Healthy: false,
Message: []string{"error1", "error2"},
},
expected: true,
},
{
name: "different messages",
old: provisioning.HealthStatus{
Healthy: false,
Message: []string{"error1"},
},
new: provisioning.HealthStatus{
Healthy: false,
Message: []string{"error2"},
},
expected: true,
},
{
name: "no change",
old: provisioning.HealthStatus{
Healthy: true,
Message: []string{},
},
new: provisioning.HealthStatus{
Healthy: true,
Message: []string{},
},
expected: false,
},
{
name: "same messages",
old: provisioning.HealthStatus{
Healthy: false,
Message: []string{"error1", "error2"},
},
new: provisioning.HealthStatus{
Healthy: false,
Message: []string{"error1", "error2"},
},
expected: false,
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
mockTester := mocks.NewRepositoryTester(t)
mockPatcher := mocks.NewStatusPatcher(t)
hc := NewHealthChecker(mockTester, mockPatcher)
result := hc.hasHealthStatusChanged(tt.old, tt.new)
assert.Equal(t, tt.expected, result)
})
}
}
// mockRepository implements repository.Repository interface for testing
type mockRepository struct {
config *provisioning.Repository
}
func (m *mockRepository) Config() *provisioning.Repository {
return m.config
}
func (m *mockRepository) Validate() field.ErrorList {
return nil
}
func (m *mockRepository) Test(ctx context.Context) (*provisioning.TestResults, error) {
return &provisioning.TestResults{Success: true}, nil
}
@@ -0,0 +1,62 @@
// Code generated by mockery v2.52.4. DO NOT EDIT.
package mocks
import (
context "context"
mock "github.com/stretchr/testify/mock"
repository "github.com/grafana/grafana/pkg/registry/apis/provisioning/repository"
v0alpha1 "github.com/grafana/grafana/apps/provisioning/pkg/apis/provisioning/v0alpha1"
)
// RepositoryTester is an autogenerated mock type for the RepositoryTester type
type RepositoryTester struct {
mock.Mock
}
// TestRepository provides a mock function with given fields: ctx, repo
func (_m *RepositoryTester) TestRepository(ctx context.Context, repo repository.Repository) (*v0alpha1.TestResults, error) {
ret := _m.Called(ctx, repo)
if len(ret) == 0 {
panic("no return value specified for TestRepository")
}
var r0 *v0alpha1.TestResults
var r1 error
if rf, ok := ret.Get(0).(func(context.Context, repository.Repository) (*v0alpha1.TestResults, error)); ok {
return rf(ctx, repo)
}
if rf, ok := ret.Get(0).(func(context.Context, repository.Repository) *v0alpha1.TestResults); ok {
r0 = rf(ctx, repo)
} else {
if ret.Get(0) != nil {
r0 = ret.Get(0).(*v0alpha1.TestResults)
}
}
if rf, ok := ret.Get(1).(func(context.Context, repository.Repository) error); ok {
r1 = rf(ctx, repo)
} else {
r1 = ret.Error(1)
}
return r0, r1
}
// NewRepositoryTester creates a new instance of RepositoryTester. It also registers a testing interface on the mock and a cleanup function to assert the mocks expectations.
// The first argument is typically a *testing.T value.
func NewRepositoryTester(t interface {
mock.TestingT
Cleanup(func())
}) *RepositoryTester {
mock := &RepositoryTester{}
mock.Mock.Test(t)
t.Cleanup(func() { mock.AssertExpectations(t) })
return mock
}
@@ -0,0 +1,55 @@
// Code generated by mockery v2.52.4. DO NOT EDIT.
package mocks
import (
context "context"
mock "github.com/stretchr/testify/mock"
v0alpha1 "github.com/grafana/grafana/apps/provisioning/pkg/apis/provisioning/v0alpha1"
)
// StatusPatcher is an autogenerated mock type for the StatusPatcher type
type StatusPatcher struct {
mock.Mock
}
// Patch provides a mock function with given fields: ctx, repo, patchOperations
func (_m *StatusPatcher) Patch(ctx context.Context, repo *v0alpha1.Repository, patchOperations ...map[string]interface{}) error {
_va := make([]interface{}, len(patchOperations))
for _i := range patchOperations {
_va[_i] = patchOperations[_i]
}
var _ca []interface{}
_ca = append(_ca, ctx, repo)
_ca = append(_ca, _va...)
ret := _m.Called(_ca...)
if len(ret) == 0 {
panic("no return value specified for Patch")
}
var r0 error
if rf, ok := ret.Get(0).(func(context.Context, *v0alpha1.Repository, ...map[string]interface{}) error); ok {
r0 = rf(ctx, repo, patchOperations...)
} else {
r0 = ret.Error(0)
}
return r0
}
// NewStatusPatcher creates a new instance of StatusPatcher. It also registers a testing interface on the mock and a cleanup function to assert the mocks expectations.
// The first argument is typically a *testing.T value.
func NewStatusPatcher(t interface {
mock.TestingT
Cleanup(func())
}) *StatusPatcher {
mock := &StatusPatcher{}
mock.Mock.Test(t)
t.Cleanup(func() { mock.AssertExpectations(t) })
return mock
}
@@ -2,7 +2,6 @@ package controller
import (
"context"
"encoding/json"
"errors"
"fmt"
"time"
@@ -35,10 +34,6 @@ type RepoGetter interface {
AsRepository(ctx context.Context, cfg *provisioning.Repository) (repository.Repository, error)
}
type RepositoryTester interface {
TestRepository(ctx context.Context, repo repository.Repository) (*provisioning.TestResults, error)
}
const loggerName = "provisioning-repository-controller"
const (
@@ -61,12 +56,13 @@ type RepositoryController struct {
logger logging.Logger
dualwrite dualwrite.Service
jobs jobs.Queue
finalizer *finalizer
jobs jobs.Queue
finalizer *finalizer
statusPatcher StatusPatcher
// Converts config to instance
repoGetter RepoGetter
tester RepositoryTester
repoGetter RepoGetter
healthChecker *HealthChecker
// To allow injection for testing.
processFn func(item *queueItem) error
enqueueRepository func(obj any)
@@ -86,6 +82,8 @@ func NewRepositoryController(
tester RepositoryTester,
jobs jobs.Queue,
dualwrite dualwrite.Service,
healthChecker *HealthChecker,
statusPatcher StatusPatcher,
) (*RepositoryController, error) {
rc := &RepositoryController{
client: provisioningClient,
@@ -98,13 +96,14 @@ func NewRepositoryController(
Name: "provisioningRepositoryController",
},
),
repoGetter: repoGetter,
parsers: parsers,
repoGetter: repoGetter,
healthChecker: healthChecker,
statusPatcher: statusPatcher,
parsers: parsers,
finalizer: &finalizer{
lister: resourceLister,
clientFactory: clients,
},
tester: tester,
jobs: jobs,
logger: logging.DefaultLogger.With("logger", loggerName),
dualwrite: dualwrite,
@@ -247,47 +246,6 @@ func (rc *RepositoryController) handleDelete(ctx context.Context, obj *provision
return nil
}
func (rc *RepositoryController) shouldCheckHealth(obj *provisioning.Repository) bool {
if obj.Status.Health.Checked == 0 || obj.Generation != obj.Status.ObservedGeneration {
return true
}
healthAge := time.Since(time.UnixMilli(obj.Status.Health.Checked))
if obj.Status.Health.Healthy {
return healthAge > time.Minute*5 // when healthy, check every 5 mins
}
return healthAge > time.Minute // otherwise within a minute
}
func (rc *RepositoryController) runHealthCheck(ctx context.Context, repo repository.Repository) provisioning.HealthStatus {
logger := logging.FromContext(ctx)
logger.Info("running health check")
res, err := rc.tester.TestRepository(ctx, repo)
if err != nil {
res = &provisioning.TestResults{
Success: false,
Errors: []provisioning.ErrorDetails{{
Detail: fmt.Sprintf("error running test repository: %s", err.Error()),
}},
}
}
healthStatus := provisioning.HealthStatus{
Healthy: res.Success,
Checked: time.Now().UnixMilli(),
}
for _, err := range res.Errors {
if err.Detail != "" {
healthStatus.Message = append(healthStatus.Message, err.Detail)
}
}
logger.Info("health check completed", "status", healthStatus)
return healthStatus
}
func (rc *RepositoryController) shouldResync(obj *provisioning.Repository) bool {
// don't trigger resync if a sync was never started
if obj.Status.Sync.Finished == 0 && obj.Status.Sync.State == "" {
@@ -309,7 +267,7 @@ func (rc *RepositoryController) shouldResync(obj *provisioning.Repository) bool
func (rc *RepositoryController) runHooks(ctx context.Context, repo repository.Repository, obj *provisioning.Repository) ([]map[string]interface{}, error) {
logger := logging.FromContext(ctx)
hooks, _ := repo.(repository.Hooks)
if hooks == nil || obj.Generation == obj.Status.ObservedGeneration {
if hooks == nil {
return nil, nil
}
@@ -400,26 +358,7 @@ func (rc *RepositoryController) addSyncJob(ctx context.Context, obj *provisionin
return nil
}
func (rc *RepositoryController) patchStatus(ctx context.Context, obj *provisioning.Repository, patchOperations []map[string]interface{}) error {
if len(patchOperations) == 0 {
return nil
}
patch, err := json.Marshal(patchOperations)
if err != nil {
return fmt.Errorf("error encoding status patch: %w", err)
}
_, err = rc.client.Repositories(obj.GetNamespace()).
Patch(ctx, obj.Name, types.JSONPatchType, patch, v1.PatchOptions{}, "status")
if err != nil {
return fmt.Errorf("error applying status patch: %w", err)
}
return nil
}
func (rc *RepositoryController) determineSyncStatus(obj *provisioning.Repository, syncOptions *provisioning.SyncJobOptions) *provisioning.SyncStatus {
func (rc *RepositoryController) determineSyncStatus(obj *provisioning.Repository, syncOptions *provisioning.SyncJobOptions, healthStatus provisioning.HealthStatus) *provisioning.SyncStatus {
const unhealthyMessage = "Repository is unhealthy"
hasUnhealthyMessage := len(obj.Status.Sync.Message) > 0 && obj.Status.Sync.Message[0] == unhealthyMessage
@@ -430,13 +369,13 @@ func (rc *RepositoryController) determineSyncStatus(obj *provisioning.Repository
LastRef: obj.Status.Sync.LastRef,
Started: time.Now().UnixMilli(),
}
case obj.Status.Health.Healthy && hasUnhealthyMessage: // if the repository is healthy and the message is set, clear it
case healthStatus.Healthy && hasUnhealthyMessage: // if the repository is healthy and the message is set, clear it
// FIXME: is this the clearest way to do this? Should we introduce another status or way of way of handling more
// specific errors?
return &provisioning.SyncStatus{
LastRef: obj.Status.Sync.LastRef,
}
case !obj.Status.Health.Healthy && !hasUnhealthyMessage: // if the repository is unhealthy and the message is not already set, set it
case !healthStatus.Healthy && !hasUnhealthyMessage: // if the repository is unhealthy and the message is not already set, set it
return &provisioning.SyncStatus{
State: provisioning.JobStateError,
Message: []string{unhealthyMessage},
@@ -450,6 +389,7 @@ func (rc *RepositoryController) determineSyncStatus(obj *provisioning.Repository
//nolint:gocyclo
func (rc *RepositoryController) process(item *queueItem) error {
logger := rc.logger.With("key", item.key)
ctx := logging.Context(context.Background(), logger)
namespace, name, err := cache.SplitMetaNamespaceKey(item.key)
if err != nil {
@@ -464,7 +404,7 @@ func (rc *RepositoryController) process(item *queueItem) error {
return err
}
ctx, _, err := identity.WithProvisioningIdentity(context.Background(), namespace)
ctx, _, err = identity.WithProvisioningIdentity(ctx, namespace)
if err != nil {
return err
}
@@ -476,7 +416,7 @@ func (rc *RepositoryController) process(item *queueItem) error {
}
shouldResync := rc.shouldResync(obj)
shouldCheckHealth := rc.shouldCheckHealth(obj)
shouldCheckHealth := rc.healthChecker.ShouldCheckHealth(obj)
hasSpecChanged := obj.Generation != obj.Status.ObservedGeneration
patchOperations := []map[string]interface{}{}
@@ -503,28 +443,27 @@ func (rc *RepositoryController) process(item *queueItem) error {
return fmt.Errorf("unable to create repository from configuration: %w", err)
}
healthStatus := obj.Status.Health
if shouldCheckHealth {
healthStatus = rc.runHealthCheck(ctx, repo)
patchOperations = append(patchOperations, map[string]interface{}{
"op": "replace",
"path": "/status/health",
"value": healthStatus,
})
// Handle hooks - may return early if hooks fail
hookOps, shouldContinue, err := rc.processHooks(ctx, repo, obj)
if err != nil {
return fmt.Errorf("process hooks: %w", err)
}
if !shouldContinue {
return nil // Hook handling already updated status and returned early
}
if len(hookOps) > 0 {
patchOperations = append(patchOperations, hookOps...)
}
// Run hooks
hookOps, err := rc.runHooks(ctx, repo, obj)
switch {
case err != nil:
return err
case len(hookOps) > 0:
patchOperations = append(patchOperations, hookOps...)
// Handle health checks using the health checker
_, healthStatus, err := rc.healthChecker.RefreshHealth(ctx, repo)
if err != nil {
return fmt.Errorf("update health status: %w", err)
}
// determine the sync strategy and sync status to apply
syncOptions := rc.determineSyncStrategy(ctx, obj, repo, shouldResync, healthStatus)
if syncStatus := rc.determineSyncStatus(obj, syncOptions); syncStatus != nil {
if syncStatus := rc.determineSyncStatus(obj, syncOptions, healthStatus); syncStatus != nil {
patchOperations = append(patchOperations, map[string]interface{}{
"op": "replace",
"path": "/status/sync",
@@ -533,7 +472,7 @@ func (rc *RepositoryController) process(item *queueItem) error {
}
// Apply all patch operations
if err := rc.patchStatus(ctx, obj, patchOperations); err != nil {
if err := rc.statusPatcher.Patch(ctx, obj, patchOperations...); err != nil {
return err
}
@@ -546,3 +485,29 @@ func (rc *RepositoryController) process(item *queueItem) error {
return nil
}
// processHooks handles hook execution with intelligent retry logic
// Returns hook operations, whether processing should continue, and any error
func (rc *RepositoryController) processHooks(ctx context.Context, repo repository.Repository, obj *provisioning.Repository) ([]map[string]interface{}, bool, error) {
shouldRunHooks := obj.Generation != obj.Status.ObservedGeneration
// Skip hooks if status already indicates recent hook failure to avoid infinite retry
if shouldRunHooks && rc.healthChecker.HasRecentFailure(obj.Status.Health, provisioning.HealthFailureHook) {
shouldRunHooks = false
}
if !shouldRunHooks {
return nil, true, nil
}
hookOps, err := rc.runHooks(ctx, repo, obj)
if err != nil {
if err := rc.healthChecker.RecordFailure(ctx, provisioning.HealthFailureHook, err, obj); err != nil {
return nil, false, fmt.Errorf("update status after hook failure: %w", err)
}
return nil, false, err
}
return hookOps, true, nil
}
+42 -27
View File
@@ -58,39 +58,54 @@ func (*filesConnector) NewConnectOptions() (runtime.Object, bool, string) {
return nil, true, "" // true adds the {path} component
}
// For GET operations, allow even unhealthy repositories
// For write operations (POST, PUT, DELETE), require healthy repository
func (c *filesConnector) getRepo(ctx context.Context, method, name string) (repository.Repository, error) {
if method == http.MethodGet {
return c.getter.GetRepository(ctx, name)
} else {
return c.getter.GetHealthyRepository(ctx, name)
}
}
// TODO: document the synchronous write and delete on the API Spec
func (c *filesConnector) Connect(ctx context.Context, name string, opts runtime.Object, responder rest.Responder) (http.Handler, error) {
logger := logging.FromContext(ctx).With("logger", "files-connector", "repository_name", name)
ctx = logging.Context(ctx, logger)
repo, err := c.getter.GetHealthyRepository(ctx, name)
if err != nil {
logger.Debug("failed to find repository", "error", err)
return nil, err
}
readWriter, ok := repo.(repository.ReaderWriter)
if !ok {
return nil, apierrors.NewBadRequest("repository does not support read-writing")
}
parser, err := c.parsers.GetParser(ctx, readWriter)
if err != nil {
return nil, fmt.Errorf("failed to get parser: %w", err)
}
clients, err := c.clients.Clients(ctx, repo.Config().Namespace)
if err != nil {
return nil, fmt.Errorf("failed to get clients: %w", err)
}
folderClient, err := clients.Folder()
if err != nil {
return nil, fmt.Errorf("failed to get folder client: %w", err)
}
folders := resources.NewFolderManager(readWriter, folderClient, resources.NewEmptyFolderTree())
dualReadWriter := resources.NewDualReadWriter(readWriter, parser, folders, c.access)
return WithTimeout(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
repo, err := c.getRepo(ctx, r.Method, name)
if err != nil {
logger.Debug("failed to find repository", "error", err)
responder.Error(err)
return
}
readWriter, ok := repo.(repository.ReaderWriter)
if !ok {
responder.Error(apierrors.NewBadRequest("repository does not support read-writing"))
return
}
parser, err := c.parsers.GetParser(ctx, readWriter)
if err != nil {
responder.Error(fmt.Errorf("failed to get parser: %w", err))
return
}
clients, err := c.clients.Clients(ctx, repo.Config().Namespace)
if err != nil {
responder.Error(fmt.Errorf("failed to get clients: %w", err))
return
}
folderClient, err := clients.Folder()
if err != nil {
responder.Error(fmt.Errorf("failed to get folder client: %w", err))
return
}
folders := resources.NewFolderManager(readWriter, folderClient, resources.NewEmptyFolderTree())
dualReadWriter := resources.NewDualReadWriter(readWriter, parser, folders, c.access)
query := r.URL.Query()
opts := resources.DualWriteOptions{
Ref: query.Get("ref"),
+1 -1
View File
@@ -54,7 +54,7 @@ func (h *historySubresource) NewConnectOptions() (runtime.Object, bool, string)
func (h *historySubresource) Connect(ctx context.Context, name string, opts runtime.Object, responder rest.Responder) (http.Handler, error) {
logger := logging.FromContext(ctx).With("logger", "history-subresource")
ctx = logging.Context(ctx, logger)
repo, err := h.repoGetter.GetHealthyRepository(ctx, name)
repo, err := h.repoGetter.GetRepository(ctx, name)
if err != nil {
logger.Debug("failed to find repository", "error", err)
return nil, err
+16 -7
View File
@@ -49,17 +49,17 @@ func (c *jobsConnector) Connect(
opts runtime.Object,
responder rest.Responder,
) (http.Handler, error) {
repo, err := c.repoGetter.GetHealthyRepository(ctx, name)
if err != nil {
return nil, err
}
cfg := repo.Config()
return WithTimeout(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
ctx = r.Context()
prefix := fmt.Sprintf("/%s/jobs/", name)
idx := strings.Index(r.URL.Path, prefix)
if r.Method == http.MethodGet {
// GET operations: allow even for unhealthy repositories
repo, err := c.repoGetter.GetRepository(ctx, name)
if err != nil {
responder.Error(err)
return
}
cfg := repo.Config()
if idx > 0 {
jobUID := r.URL.Path[idx+len(prefix):]
if !ValidUUID(jobUID) {
@@ -82,6 +82,15 @@ func (c *jobsConnector) Connect(
responder.Object(http.StatusOK, recent)
return
}
// POST operations: require healthy repository
repo, err := c.repoGetter.GetHealthyRepository(ctx, name)
if err != nil {
responder.Error(err)
return
}
cfg := repo.Config()
if idx > 0 {
responder.Error(apierrors.NewBadRequest("can not post to a job UID"))
return
+1 -1
View File
@@ -46,7 +46,7 @@ func (*refsConnector) NewConnectOptions() (runtime.Object, bool, string) {
func (c *refsConnector) Connect(ctx context.Context, name string, opts runtime.Object, responder rest.Responder) (http.Handler, error) {
logger := logging.FromContext(ctx).With("logger", "refs-connector", "repository_name", name)
ctx = logging.Context(ctx, logger)
repo, err := c.getter.GetHealthyRepository(ctx, name)
repo, err := c.getter.GetRepository(ctx, name)
if err != nil {
logger.Debug("failed to find repository", "error", err)
return nil, err
+17 -47
View File
@@ -102,7 +102,6 @@ type APIBuilder struct {
}
jobHistoryConfig *JobHistoryConfig
jobHistory jobs.History
tester *RepositoryTester
resourceLister resources.ResourceLister
repositoryLister listers.RepositoryLister
legacyMigrator legacy.LegacyMigrator
@@ -114,6 +113,7 @@ type APIBuilder struct {
access authlib.AccessChecker
mutators []controller.Mutator
statusPatcher *controller.RepositoryStatusPatcher
healthChecker *controller.HealthChecker
// Extras provides additional functionality to the API.
extras []Extra
availableRepositoryTypes map[provisioning.RepositoryType]bool
@@ -406,6 +406,10 @@ func (b *APIBuilder) GetStatusPatcher() *controller.RepositoryStatusPatcher {
return b.statusPatcher
}
func (b *APIBuilder) GetHealthChecker() *controller.HealthChecker {
return b.healthChecker
}
func (b *APIBuilder) InstallSchema(scheme *runtime.Scheme) error {
err := provisioning.AddToScheme(scheme)
if err != nil {
@@ -473,10 +477,7 @@ func (b *APIBuilder) UpdateAPIGroupInfo(apiGroupInfo *genericapiserver.APIGroupI
storage[provisioning.RepositoryResourceInfo.StoragePath("status")] = repositoryStatusStorage
// TODO: Add some logic so that the connectors can registered themselves and we don't have logic all over the place
// TODO: Do not set private fields directly, use factory methods.
storage[provisioning.RepositoryResourceInfo.StoragePath("test")] = &testConnector{
getter: b,
}
storage[provisioning.RepositoryResourceInfo.StoragePath("test")] = NewTestConnector(b, &repository.Tester{}, b)
storage[provisioning.RepositoryResourceInfo.StoragePath("files")] = NewFilesConnector(b, b.parsers, b.clients, b.access)
storage[provisioning.RepositoryResourceInfo.StoragePath("refs")] = NewRefsConnector(b)
storage[provisioning.RepositoryResourceInfo.StoragePath("resources")] = &listConnector{
@@ -657,14 +658,11 @@ func (b *APIBuilder) GetPostStartHooks() (map[string]genericapiserver.PostStartH
jobInformer := sharedInformerFactory.Provisioning().V0alpha1().Jobs()
b.client = c.ProvisioningV0alpha1()
// We do not have a local client until *GetPostStartHooks*, so we can delay init for some
b.tester = &RepositoryTester{
client: b.GetClient(),
}
b.repositoryLister = repoInformer.Lister()
b.statusPatcher = controller.NewRepositoryStatusPatcher(b.GetClient())
b.healthChecker = controller.NewHealthChecker(&repository.Tester{}, b.statusPatcher)
// if running solely CRUD, skip the rest of the setup
if b.localFileResolver == nil {
return nil
@@ -691,7 +689,6 @@ func (b *APIBuilder) GetPostStartHooks() (map[string]genericapiserver.PostStartH
stageIfPossible,
)
b.statusPatcher = controller.NewRepositoryStatusPatcher(b.GetClient())
syncer := sync.NewSyncer(sync.Compare, sync.FullSync, sync.IncrementalSync)
syncWorker := sync.NewSyncWorker(
b.clients,
@@ -782,6 +779,8 @@ func (b *APIBuilder) GetPostStartHooks() (map[string]genericapiserver.PostStartH
&repository.Tester{},
b.jobs,
b.storageStatus,
b.GetHealthChecker(),
b.statusPatcher,
)
if err != nil {
return err
@@ -1254,49 +1253,20 @@ func (b *APIBuilder) GetRepository(ctx context.Context, name string) (repository
return b.asRepository(ctx, obj)
}
func timeSince(when int64) time.Duration {
return time.Duration(time.Now().UnixMilli()-when) * time.Millisecond
}
func (b *APIBuilder) GetHealthyRepository(ctx context.Context, name string) (repository.Repository, error) {
repo, err := b.GetRepository(ctx, name)
if err != nil {
return nil, err
}
status := repo.Config().Status.Health
if !status.Healthy {
if timeSince(status.Checked) > time.Second*25 {
ctx, _, err = identity.WithProvisioningIdentity(ctx, repo.Config().Namespace)
if err != nil {
return nil, err // The status
}
// Check health again
s, err := repository.TestRepository(ctx, repo)
if err != nil {
return nil, err // The status
}
// Write and return the repo with current status
cfg, _ := b.tester.UpdateHealthStatus(ctx, repo.Config(), s)
if cfg != nil {
status = cfg.Status.Health
if cfg.Status.Health.Healthy {
status = cfg.Status.Health
repo, err = b.AsRepository(ctx, cfg)
if err != nil {
return nil, err
}
}
}
}
if !status.Healthy {
return nil, &apierrors.StatusError{ErrStatus: metav1.Status{
Code: http.StatusFailedDependency,
Message: "The repository configuration is not healthy",
}}
}
return nil, &apierrors.StatusError{ErrStatus: metav1.Status{
Code: http.StatusFailedDependency,
Message: "The repository configuration is not healthy",
}}
}
return repo, err
}
+74 -41
View File
@@ -8,18 +8,37 @@ import (
"reflect"
"time"
"k8s.io/apimachinery/pkg/api/errors"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/runtime"
"k8s.io/apiserver/pkg/endpoints/request"
"k8s.io/apiserver/pkg/registry/rest"
provisioning "github.com/grafana/grafana/apps/provisioning/pkg/apis/provisioning/v0alpha1"
client "github.com/grafana/grafana/apps/provisioning/pkg/generated/clientset/versioned/typed/provisioning/v0alpha1"
"github.com/grafana/grafana/pkg/registry/apis/provisioning/controller"
"github.com/grafana/grafana/pkg/registry/apis/provisioning/repository"
)
type StatusPatcherProvider interface {
GetStatusPatcher() *controller.RepositoryStatusPatcher
}
type HealthCheckerProvider interface {
GetHealthChecker() *controller.HealthChecker
}
type testConnector struct {
getter RepoGetter
getter RepoGetter
tester controller.RepositoryTester
healthProvider HealthCheckerProvider
}
func NewTestConnector(getter RepoGetter, tester controller.RepositoryTester, healthProvider HealthCheckerProvider) *testConnector {
return &testConnector{
getter: getter,
tester: tester,
healthProvider: healthProvider,
}
}
func (*testConnector) New() runtime.Object {
@@ -94,57 +113,71 @@ func (s *testConnector) Connect(ctx context.Context, name string, opts runtime.O
}
}
var rsp *provisioning.TestResults
if repo == nil {
healthChecker := s.healthProvider.GetHealthChecker()
if healthChecker == nil {
// Use precondition failed for when health checker is not ready yet
responder.Error(&errors.StatusError{
ErrStatus: metav1.Status{
Status: metav1.StatusFailure,
Code: http.StatusPreconditionFailed,
Reason: metav1.StatusReason("PreconditionFailed"),
Message: "health checker not initialized yet, please try again",
},
})
return
}
// Testing existing repository - get it and update health
repo, err = s.getter.GetRepository(ctx, name)
if err != nil {
responder.Error(err)
return
}
// If the last error was not a health check error or empty, return precondition failed
health := repo.Config().Status.Health
if health.Error != provisioning.HealthFailureHealth && health.Error != "" {
rsp = &provisioning.TestResults{
Success: false,
Code: http.StatusPreconditionFailed,
Errors: func() []provisioning.ErrorDetails {
var errs []provisioning.ErrorDetails
for _, msg := range health.Message {
errs = append(errs, provisioning.ErrorDetails{Detail: msg})
}
return errs
}(),
}
if err := healthChecker.RefreshTimestamp(ctx, repo.Config()); err != nil {
responder.Error(err)
return
}
responder.Object(rsp.Code, rsp)
return
}
rsp, _, err = healthChecker.RefreshHealth(ctx, repo)
if err != nil {
responder.Error(err)
return
}
} else {
// Testing temporary repository - just run test without status update
rsp, err = s.tester.TestRepository(ctx, repo)
if err != nil {
responder.Error(err)
return
}
}
// Only call test if field validation passes
rsp, err := repository.TestRepository(ctx, repo)
if err != nil {
responder.Error(err)
return
}
responder.Object(rsp.Code, rsp)
}), 30*time.Second), nil
}
// TODO: Move tester to a more suitable location out of the connector.
type RepositoryTester struct {
// Repository+Jobs
client client.ProvisioningV0alpha1Interface
}
// This function will check if the repository is configured and functioning as expected
func (t *RepositoryTester) UpdateHealthStatus(ctx context.Context, cfg *provisioning.Repository, res *provisioning.TestResults) (*provisioning.Repository, error) {
if res == nil {
res = &provisioning.TestResults{
Success: false,
Errors: []provisioning.ErrorDetails{{
Detail: "missing health status",
}},
}
}
repo := cfg.DeepCopy()
repo.Status.Health = provisioning.HealthStatus{
Healthy: res.Success,
Checked: time.Now().UnixMilli(),
}
for _, err := range res.Errors {
if err.Detail != "" {
repo.Status.Health.Message = append(repo.Status.Health.Message, err.Detail)
}
}
_, err := t.client.Repositories(repo.GetNamespace()).
UpdateStatus(ctx, repo, metav1.UpdateOptions{})
return repo, err
}
var (
_ rest.Storage = (*testConnector)(nil)
_ rest.Connecter = (*testConnector)(nil)
@@ -34,6 +34,16 @@ type WebhookExtraBuilder struct {
provisioningapis.ExtraBuilder
}
// HACK: assume that the URL is public if it starts with "https://" and does not contain any local IP ranges
func isPublicURL(url string) bool {
return strings.HasPrefix(url, "https://") &&
!strings.Contains(url, "localhost") &&
!strings.HasPrefix(url, "https://127.") &&
!strings.HasPrefix(url, "https://192.") &&
!strings.HasPrefix(url, "https://10.") &&
!strings.HasPrefix(url, "https://172.16.")
}
func ProvideWebhooks(
cfg *setting.Cfg,
features featuremgmt.FeatureToggles,
@@ -48,8 +58,8 @@ func ProvideWebhooks(
urlProvider := func(_ string) string {
return cfg.AppURL
}
// HACK: Assume is only public if it is HTTPS
isPublic := strings.HasPrefix(urlProvider(""), "https://")
isPublic := isPublicURL(urlProvider(""))
clients := resources.NewClientFactory(configProvider)
parsers := resources.NewParserFactory(clients)
@@ -74,6 +84,7 @@ func ProvideWebhooks(
filepath.Join(cfg.DataPath, "clone"),
parsers,
[]jobs.Worker{pullRequestWorker},
isPublic, // Pass the public URL flag
)
},
}
@@ -90,6 +101,7 @@ type WebhookExtra struct {
clonedir string
parsers resources.ParserFactory
workers []jobs.Worker
isPublic bool // Flag to determine if webhook-enhanced repositories should be created
}
func NewWebhookExtra(
@@ -101,6 +113,7 @@ func NewWebhookExtra(
clonedir string,
parsers resources.ParserFactory,
workers []jobs.Worker,
isPublic bool,
) *WebhookExtra {
return &WebhookExtra{
render: render,
@@ -111,6 +124,7 @@ func NewWebhookExtra(
clonedir: clonedir,
parsers: parsers,
workers: workers,
isPublic: isPublic,
}
}
@@ -155,7 +169,8 @@ func (e *WebhookExtra) GetJobWorkers() []jobs.Worker {
// AsRepository delegates repository creation to the webhook connector
func (e *WebhookExtra) AsRepository(ctx context.Context, r *provisioning.Repository) (repository.Repository, error) {
if r.Spec.Type == provisioning.GitHubRepositoryType {
// Only handle GitHub repositories with webhooks if URL is public
if r.Spec.Type == provisioning.GitHubRepositoryType && e.isPublic {
gvr := provisioning.RepositoryResourceInfo.GroupVersionResource()
webhookURL := fmt.Sprintf(
"%sapis/%s/%s/namespaces/%s/%s/%s/webhook",
@@ -209,7 +224,11 @@ func (e *WebhookExtra) AsRepository(ctx context.Context, r *provisioning.Reposit
}
func (e *WebhookExtra) RepositoryTypes() []provisioning.RepositoryType {
return []provisioning.RepositoryType{
provisioning.GitHubRepositoryType,
// Only claim to handle GitHub repositories if URL is public
if e.isPublic {
return []provisioning.RepositoryType{
provisioning.GitHubRepositoryType,
}
}
return []provisioning.RepositoryType{}
}
@@ -2862,6 +2862,14 @@
"type": "integer",
"format": "int64"
},
"error": {
"description": "The type of the error\n\nPossible enum values:\n - `\"health\"`\n - `\"hook\"`",
"type": "string",
"enum": [
"health",
"hook"
]
},
"healthy": {
"description": "When not healthy, requests will not be executed",
"type": "boolean",
+24 -32
View File
@@ -25,23 +25,19 @@ func TestIntegrationProvisioning_DeleteResources(t *testing.T) {
ctx := context.Background()
const repo = "delete-test-repo"
localTmp := helper.RenderObject(t, "testdata/local-write.json.tmpl", map[string]any{
"Name": repo,
"SyncEnabled": true,
"SyncTarget": "instance",
helper.CreateRepo(t, TestRepo{
Name: repo,
Path: helper.ProvisioningPath,
Target: "instance",
Copies: map[string]string{
"testdata/all-panels.json": "dashboard1.json",
"testdata/text-options.json": "folder/dashboard2.json",
"testdata/timeline-demo.json": "folder/nested/dashboard3.json",
"testdata/.keep": "folder/nested/.keep",
},
ExpectedDashboards: 3,
ExpectedFolders: 2,
})
_, err := helper.Repositories.Resource.Create(ctx, localTmp, metav1.CreateOptions{})
require.NoError(t, err)
// Copy the dashboards to the repository path
helper.CopyToProvisioningPath(t, "testdata/all-panels.json", "dashboard1.json")
helper.CopyToProvisioningPath(t, "testdata/text-options.json", "folder/dashboard2.json")
helper.CopyToProvisioningPath(t, "testdata/timeline-demo.json", "folder/nested/dashboard3.json")
// make sure we don't fail when there is a .keep file in a folder
helper.CopyToProvisioningPath(t, "testdata/.keep", "folder/nested/.keep")
// Trigger and wait for a sync job to finish
helper.SyncAndWait(t, repo, nil)
dashboards, err := helper.DashboardsV1.Resource.List(ctx, metav1.ListOptions{})
require.NoError(t, err)
@@ -118,21 +114,17 @@ func TestIntegrationProvisioning_MoveResources(t *testing.T) {
helper := runGrafana(t)
ctx := context.Background()
const repo = "move-test-repo"
localTmp := helper.RenderObject(t, "testdata/local-write.json.tmpl", map[string]any{
"Name": repo,
"SyncEnabled": true,
"SyncTarget": "instance",
repo := "move-test-repo"
helper.CreateRepo(t, TestRepo{
Name: repo,
Path: helper.ProvisioningPath,
Target: "instance",
Copies: map[string]string{
"testdata/all-panels.json": "all-panels.json",
},
ExpectedDashboards: 1,
ExpectedFolders: 0,
})
_, err := helper.Repositories.Resource.Create(ctx, localTmp, metav1.CreateOptions{})
require.NoError(t, err)
// Copy test dashboards to the repository path for initial setup
const originalDashboard = "all-panels.json"
helper.CopyToProvisioningPath(t, "testdata/all-panels.json", originalDashboard)
// Wait for sync to ensure the dashboard is created in Grafana
helper.SyncAndWait(t, repo, nil)
// Verify the original dashboard exists in Grafana (using the UID from all-panels.json)
const allPanelsUID = "n1jR8vnnz" // This is the UID from the all-panels.json file
@@ -146,7 +138,7 @@ func TestIntegrationProvisioning_MoveResources(t *testing.T) {
// Perform the move operation using helper function
resp := helper.postFilesRequest(t, repo, filesPostOptions{
targetPath: targetPath,
originalPath: originalDashboard,
originalPath: "all-panels.json",
message: "move file without content change",
})
// nolint:errcheck
@@ -167,7 +159,7 @@ func TestIntegrationProvisioning_MoveResources(t *testing.T) {
require.Equal(t, "Panel tests - All panels", title, "content should be preserved")
// Verify original file no longer exists
_, err = helper.Repositories.Resource.Get(ctx, repo, metav1.GetOptions{}, "files", originalDashboard)
_, err = helper.Repositories.Resource.Get(ctx, repo, metav1.GetOptions{}, "files", "all-panels.json")
require.Error(t, err, "original file should no longer exist")
// Verify dashboard still exists in Grafana with same content but may have updated path references
+204
View File
@@ -0,0 +1,204 @@
package provisioning
import (
"context"
"encoding/json"
"os"
"testing"
"github.com/stretchr/testify/require"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/apis/meta/v1/unstructured"
"k8s.io/apimachinery/pkg/runtime"
provisioning "github.com/grafana/grafana/apps/provisioning/pkg/apis/provisioning/v0alpha1"
)
func TestIntegrationHealth(t *testing.T) {
if testing.Short() {
t.Skip("skipping integration test")
}
helper := runGrafana(t)
ctx := context.Background()
repo := "test-repo-health"
helper.CreateRepo(t, TestRepo{
Name: repo,
})
// Verify the health status before calling the endpoint
repoObj, err := helper.Repositories.Resource.Get(ctx, repo, metav1.GetOptions{})
require.NoError(t, err)
originalRepo := unstructuredToRepository(t, repoObj)
require.True(t, originalRepo.Status.Health.Healthy, "repository should be marked healthy")
require.Empty(t, originalRepo.Status.Health.Error, "should be empty")
require.Empty(t, originalRepo.Status.Health.Message, "should not have messages")
t.Run("test endpoint with new repository configuration works", func(t *testing.T) {
newRepoConfig := map[string]any{
"apiVersion": "provisioning.grafana.app/v0alpha1",
"kind": "Repository",
"spec": map[string]any{
"title": "Test New Configuration",
"type": "local",
"local": map[string]any{
"path": helper.ProvisioningPath,
},
"workflows": []string{"write"},
"sync": map[string]any{
"enabled": true,
"target": "folder",
"intervalSeconds": 10,
},
},
}
configBytes, err := json.Marshal(newRepoConfig)
require.NoError(t, err)
// Test the new configuration - this should work
result := helper.AdminREST.Post().
Namespace("default").
Resource("repositories").
Name("test-new-config").
SubResource("test").
Body(configBytes).
SetHeader("Content-Type", "application/json").
Do(ctx)
require.NoError(t, result.Error(), "test endpoint should work for new repository configurations")
obj, err := result.Get()
require.NoError(t, err)
testResults := parseTestResults(t, obj)
require.True(t, testResults.Success, "test should succeed for valid repository configuration")
require.Equal(t, 200, testResults.Code, "should return 200 for successful test")
// Verify the repository was not actually created (this was just a test)
_, err = helper.Repositories.Resource.Get(ctx, "test-new-config", metav1.GetOptions{})
require.True(t, err != nil, "repository should not be created during test")
})
t.Run("test endpoint with existing repository", func(t *testing.T) {
result := helper.AdminREST.Post().
Namespace("default").
Resource("repositories").
Name(repo).
SubResource("test").
SetHeader("Content-Type", "application/json").
Do(ctx)
require.NoError(t, result.Error(), "test endpoint should return NOT an error for existing repository")
obj, err := result.Get()
require.NoError(t, err)
testResults := parseTestResults(t, obj)
t.Logf("SUCCESS: Test endpoint worked for existing repository: Success=%v, Code=%d",
testResults.Success, testResults.Code)
require.True(t, testResults.Success, "test should succeed for existing repository")
require.Equal(t, 200, testResults.Code, "should return 200 for successful test")
// Verify repository health status after update
repoObj, err := helper.Repositories.Resource.Get(ctx, repo, metav1.GetOptions{})
require.NoError(t, err)
afterTest := unstructuredToRepository(t, repoObj)
require.True(t, afterTest.Status.Health.Healthy, "repository should be marked healthy")
require.Empty(t, afterTest.Status.Health.Error, "should be empty")
require.Empty(t, afterTest.Status.Health.Message, "should not have messages")
// For healthy repositories, timestamp may not change immediately as it can take up to 30 seconds to update
})
t.Run("test endpoint with unhealthy repository", func(t *testing.T) {
// Remove the repository folder to make it unhealthy
repoPath := helper.ProvisioningPath
err := os.RemoveAll(repoPath)
require.NoError(t, err, "should be able to remove repository directory")
// Wait a bit for the system to detect the unhealthy state
// (In a real scenario, this would be detected during the next health check cycle)
// Get the repository status before the test
repoObj, err := helper.Repositories.Resource.Get(ctx, repo, metav1.GetOptions{})
require.NoError(t, err)
beforeTest := unstructuredToRepository(t, repoObj)
t.Logf("Before test - Healthy: %v, Checked: %d", beforeTest.Status.Health.Healthy, beforeTest.Status.Health.Checked)
// Call the test endpoint
result := helper.AdminREST.Post().
Namespace("default").
Resource("repositories").
Name(repo).
SubResource("test").
SetHeader("Content-Type", "application/json").
Do(ctx)
// The test endpoint may return an error for unhealthy repositories
obj, err := result.Get()
if result.Error() != nil {
t.Logf("Test endpoint returned error for unhealthy repository (expected): %v", result.Error())
} else {
require.NoError(t, err)
testResults := parseTestResults(t, obj)
t.Logf("Test endpoint result for unhealthy repository: Success=%v, Code=%d",
testResults.Success, testResults.Code)
}
// Verify repository health status after test - timestamp should change
repoObj, err = helper.Repositories.Resource.Get(ctx, repo, metav1.GetOptions{})
require.NoError(t, err)
afterTest := unstructuredToRepository(t, repoObj)
t.Logf("After test - Healthy: %v, Checked: %d", afterTest.Status.Health.Healthy, afterTest.Status.Health.Checked)
// For unhealthy repositories, the timestamp should change as the health check will be triggered
require.NotEqual(t, beforeTest.Status.Health.Checked, afterTest.Status.Health.Checked, "should change the timestamp for unhealthy repository check")
// Recreate the repository directory to restore healthy state
err = os.MkdirAll(repoPath, 0o750)
require.NoError(t, err, "should be able to recreate repository directory")
// Call the test endpoint again to trigger health check after recreating directory
result = helper.AdminREST.Post().
Namespace("default").
Resource("repositories").
Name(repo).
SubResource("test").
SetHeader("Content-Type", "application/json").
Do(ctx)
// Should succeed now that the directory is recreated
require.NoError(t, result.Error(), "test endpoint should work after recreating directory")
obj, err = result.Get()
require.NoError(t, err)
testResults := parseTestResults(t, obj)
require.True(t, testResults.Success, "test should succeed after recreating directory")
require.Equal(t, 200, testResults.Code, "should return 200 after recreating directory")
// Verify repository health status is now healthy again
repoObj, err = helper.Repositories.Resource.Get(ctx, repo, metav1.GetOptions{})
require.NoError(t, err)
finalRepo := unstructuredToRepository(t, repoObj)
t.Logf("After recreating directory - Healthy: %v, Checked: %d", finalRepo.Status.Health.Healthy, finalRepo.Status.Health.Checked)
require.True(t, finalRepo.Status.Health.Healthy, "repository should be healthy again after recreating directory")
require.Empty(t, finalRepo.Status.Health.Error, "should have no error after recreating directory")
// Timestamp should have changed again due to the health check
require.NotEqual(t, afterTest.Status.Health.Checked, finalRepo.Status.Health.Checked, "timestamp should change when repository becomes healthy again")
})
}
// parseTestResults extracts TestResults from the API response
func parseTestResults(t *testing.T, obj runtime.Object) *provisioning.TestResults {
t.Helper()
unstructuredObj, ok := obj.(*unstructured.Unstructured)
require.True(t, ok, "expected unstructured object")
data, err := json.Marshal(unstructuredObj.Object)
require.NoError(t, err)
var testResults provisioning.TestResults
err = json.Unmarshal(data, &testResults)
require.NoError(t, err)
return &testResults
}
+58 -20
View File
@@ -316,11 +316,11 @@ func (h *provisioningTestHelper) RenderObject(t *testing.T, filePath string, val
func (h *provisioningTestHelper) CopyToProvisioningPath(t *testing.T, from, to string) {
fullPath := path.Join(h.ProvisioningPath, to)
t.Logf("Copying file from '%s' to provisioning path '%s'", from, fullPath)
err := os.MkdirAll(path.Dir(fullPath), 0750)
err := os.MkdirAll(path.Dir(fullPath), 0o750)
require.NoError(t, err, "failed to create directories for provisioning path")
file := h.LoadFile(from)
err = os.WriteFile(fullPath, file, 0600)
err = os.WriteFile(fullPath, file, 0o600)
require.NoError(t, err, "failed to write file to provisioning path")
}
@@ -461,14 +461,16 @@ func (h *provisioningTestHelper) logRepositoryObject(t *testing.T, obj map[strin
}
type TestRepo struct {
Name string
Target string
Path string
Values map[string]any
Copies map[string]string
ExpectedDashboards int
ExpectedFolders int
SkipSync bool
Name string
Target string
Path string
Values map[string]any
Copies map[string]string
ExpectedDashboards int
ExpectedFolders int
SkipSync bool
SkipResourceAssertions bool
Template string
}
func (h *provisioningTestHelper) CreateRepo(t *testing.T, repo TestRepo) {
@@ -481,7 +483,7 @@ func (h *provisioningTestHelper) CreateRepo(t *testing.T, repo TestRepo) {
if repo.Path != "" {
repoPath = repo.Path
// Ensure the directory exists
err := os.MkdirAll(repoPath, 0750)
err := os.MkdirAll(repoPath, 0o750)
require.NoError(t, err, "should be able to create repository path")
}
@@ -493,20 +495,29 @@ func (h *provisioningTestHelper) CreateRepo(t *testing.T, repo TestRepo) {
if repo.Path != "" {
templateVars["Path"] = repoPath
}
// Add custom values from TestRepo
for key, value := range repo.Values {
templateVars[key] = value
}
localTmp := h.RenderObject(t, "testdata/local-write.json.tmpl", templateVars)
tmpl := "testdata/local-write.json.tmpl"
if repo.Template != "" {
tmpl = repo.Template
}
localTmp := h.RenderObject(t, tmpl, templateVars)
_, err := h.Repositories.Resource.Create(t.Context(), localTmp, metav1.CreateOptions{})
require.NoError(t, err)
h.WaitForHealthyRepository(t, repo.Name)
for from, to := range repo.Copies {
if repo.Path != "" {
// Copy to custom path
fullPath := path.Join(repoPath, to)
err := os.MkdirAll(path.Dir(fullPath), 0750)
err := os.MkdirAll(path.Dir(fullPath), 0o750)
require.NoError(t, err, "failed to create directories for custom path")
file := h.LoadFile(from)
err = os.WriteFile(fullPath, file, 0600)
err = os.WriteFile(fullPath, file, 0o600)
require.NoError(t, err, "failed to write file to custom path")
} else {
h.CopyToProvisioningPath(t, from, to)
@@ -522,13 +533,31 @@ func (h *provisioningTestHelper) CreateRepo(t *testing.T, repo TestRepo) {
}
// Verify initial state
dashboards, err := h.DashboardsV1.Resource.List(t.Context(), metav1.ListOptions{})
require.NoError(t, err)
require.Equal(t, repo.ExpectedDashboards, len(dashboards.Items), "should the expected dashboards after sync")
if !repo.SkipResourceAssertions {
dashboards, err := h.DashboardsV1.Resource.List(t.Context(), metav1.ListOptions{})
require.NoError(t, err)
require.Equal(t, repo.ExpectedDashboards, len(dashboards.Items), "should the expected dashboards after sync")
folders, err := h.Folders.Resource.List(t.Context(), metav1.ListOptions{})
require.NoError(t, err)
require.Equal(t, repo.ExpectedFolders, len(folders.Items), "should have the expected folders after sync")
}
}
folders, err := h.Folders.Resource.List(t.Context(), metav1.ListOptions{})
require.NoError(t, err)
require.Equal(t, repo.ExpectedFolders, len(folders.Items), "should have the expected folders after sync")
// WaitForHealthyRepository waits for a repository to become healthy.
func (h *provisioningTestHelper) WaitForHealthyRepository(t *testing.T, name string) {
require.EventuallyWithT(t, func(collect *assert.CollectT) {
repoStatus, err := h.Repositories.Resource.Get(t.Context(), name, metav1.GetOptions{})
if !assert.NoError(collect, err, "failed to get repository status") {
return
}
errType := mustNestedString(repoStatus.Object, "status", "health", "error")
assert.Empty(collect, errType, "repository %s has health error: %s", name, errType)
msgs := mustNestedStringSlice(repoStatus.Object, "status", "health", "message")
assert.Empty(collect, msgs, "repository %s has health messages: %v", name, msgs)
status, found := mustNestedBool(repoStatus.Object, "status", "health", "healthy")
assert.True(collect, found, "repository %s does not have health status", name)
assert.True(collect, status, "repository %s is not healthy yet", name)
}, time.Second*10, time.Millisecond*50, "repository %s should become healthy", name)
}
type grafanaOption func(opts *testinfra.GrafanaOpts)
@@ -658,6 +687,15 @@ func mustNestedString(obj map[string]interface{}, fields ...string) string {
return v
}
func mustNestedBool(obj map[string]interface{}, fields ...string) (bool, bool) {
v, found, err := unstructured.NestedBool(obj, fields...)
if err != nil {
panic(err)
}
return v, found
}
func mustNestedStringSlice(obj map[string]interface{}, fields ...string) []string {
v, _, err := unstructured.NestedStringSlice(obj, fields...)
if err != nil {
+6 -12
View File
@@ -197,16 +197,6 @@ func TestIntegrationProvisioning_MoveJob(t *testing.T) {
assert.True(collect, apierrors.IsNotFound(err), "repository should be deleted")
}, time.Second*5, time.Millisecond*50, "repository should be deleted before creating new one")
// Create a unique repository for resource reference testing to avoid contamination
const refRepo = "move-ref-test-repo"
localRefTmp := helper.RenderObject(t, "testdata/local-write.json.tmpl", map[string]any{
"Name": refRepo,
"SyncEnabled": true,
"SyncTarget": "instance",
})
_, err = helper.Repositories.Resource.Create(ctx, localRefTmp, metav1.CreateOptions{})
require.NoError(t, err)
// Create modified test files with unique UIDs for ResourceRef testing
allPanelsContent := helper.LoadFile("testdata/all-panels.json")
textOptionsContent := helper.LoadFile("testdata/text-options.json")
@@ -232,8 +222,12 @@ func TestIntegrationProvisioning_MoveJob(t *testing.T) {
helper.CopyToProvisioningPath(t, tmpFile2, "move-source-2.json")
helper.CopyToProvisioningPath(t, tmpFile3, "move-source-3.json")
// Sync to populate resources in Grafana
helper.SyncAndWait(t, refRepo, nil)
// Create a unique repository for resource reference testing to avoid contamination
const refRepo = "move-ref-test-repo"
helper.CreateRepo(t, TestRepo{
Name: refRepo,
SkipResourceAssertions: true, // HACK: I am not sure why sometimes it's 6 or 3 dashbaords.
})
t.Run("move single dashboard by resource reference", func(t *testing.T) {
spec := provisioning.JobSpec{
+75 -199
View File
@@ -40,11 +40,11 @@ func TestIntegrationProvisioning_CreatingAndGetting(t *testing.T) {
for _, inputFilePath := range inputFiles {
t.Run(inputFilePath, func(t *testing.T) {
input := helper.RenderObject(t, inputFilePath, nil)
name := mustNestedString(input.Object, "metadata", "name")
_, err := helper.Repositories.Resource.Create(ctx, input, createOptions)
require.NoError(t, err, "failed to create resource")
name := mustNestedString(input.Object, "metadata", "name")
output, err := helper.Repositories.Resource.Get(ctx, name, metav1.GetOptions{})
require.NoError(t, err, "failed to read back resource")
@@ -257,18 +257,15 @@ func TestIntegrationProvisioning_CreatingGitHubRepository(t *testing.T) {
// )
const repo = "github-create-test"
_, err := helper.Repositories.Resource.Create(ctx,
helper.RenderObject(t, "testdata/github-readonly.json.tmpl", map[string]any{
"Name": repo,
"SyncEnabled": true,
"SyncTarget": "instance",
"Path": "grafana/",
}),
metav1.CreateOptions{},
)
require.NoError(t, err)
testRepo := TestRepo{
Name: repo,
Template: "testdata/github-readonly.json.tmpl",
Target: "folder",
ExpectedDashboards: 3,
ExpectedFolders: 3, // Folder sync creates an additional folder for the repository itself
}
helper.SyncAndWait(t, repo, nil)
helper.CreateRepo(t, testRepo)
// By now, we should have synced, meaning we have data to read in the local Grafana instance!
@@ -324,10 +321,12 @@ func TestIntegrationProvisioning_CreatingGitHubRepository(t *testing.T) {
for _, test := range tests {
t.Run(test.name, func(t *testing.T) {
// Create repository directly without health checks since we're only testing URL cleanup
input := helper.RenderObject(t, "testdata/github-readonly.json.tmpl", map[string]any{
"Name": test.name,
"URL": test.input,
"SyncTarget": "instance",
"Name": test.name,
"URL": test.input,
"SyncTarget": "folder",
"SyncEnabled": false, // Disable sync since we're just testing URL cleanup
})
_, err := helper.Repositories.Resource.Create(ctx, input, metav1.CreateOptions{})
@@ -353,7 +352,7 @@ func TestIntegrationProvisioning_CreatingGitHubRepository(t *testing.T) {
})
}
func TestIntegrationProvisioning_InstanceSyncValidation(t *testing.T) {
func TestIntegrationProvisioning_RepositoryLimits(t *testing.T) {
if testing.Short() {
t.Skip("skipping integration test")
}
@@ -361,136 +360,20 @@ func TestIntegrationProvisioning_InstanceSyncValidation(t *testing.T) {
helper := runGrafana(t)
ctx := context.Background()
t.Run("single instance sync is allowed", func(t *testing.T) {
repoName := "instance-repo-single"
testRepo := TestRepo{
Name: repoName,
Target: "instance",
Copies: map[string]string{}, // No files needed for this test
ExpectedDashboards: 0,
ExpectedFolders: 0,
}
// Create instance sync repository - should succeed
helper.CreateRepo(t, testRepo)
// Clean up at end of test
helper.CleanupAllRepos(t)
})
t.Run("change between folder and instance sync for the same repository if no previous sync happened", func(t *testing.T) {
// Ensure clean state
helper.CleanupAllRepos(t)
repoName := "instance-repo-change"
testRepo := TestRepo{
Name: repoName,
Target: "instance",
Copies: map[string]string{}, // No files needed for this test
ExpectedDashboards: 0,
ExpectedFolders: 0,
SkipSync: true, // To avoid initial sync and stats
}
helper.CreateRepo(t, testRepo)
// Change from instance to folder sync
repo, err := helper.Repositories.Resource.Get(ctx, repoName, metav1.GetOptions{})
require.NoError(t, err, "failed to get repository")
err = unstructured.SetNestedField(repo.Object, "folder", "spec", "sync", "target")
require.NoError(t, err, "failed to set syncTarget to folder")
_, err = helper.Repositories.Resource.Update(ctx, repo, metav1.UpdateOptions{FieldValidation: "Strict"})
require.NoError(t, err, "failed to update repository to folder sync")
// Clean up at end of test
helper.CleanupAllRepos(t)
})
t.Run("instance sync rejected when any other repository exists", func(t *testing.T) {
// Ensure clean state
helper.CleanupAllRepos(t)
existingFolderName := "existing-folder-repo"
instanceRepoName := "instance-repo-blocked"
// Create a folder sync repository first
folderTestRepo := TestRepo{
Name: existingFolderName,
Target: "folder",
Copies: map[string]string{}, // No files needed for this test
ExpectedDashboards: 0,
ExpectedFolders: 1, // One folder expected after sync
}
helper.CreateRepo(t, folderTestRepo)
// Try to create an instance sync repository - should fail because any other repository exists
instanceRepo := helper.RenderObject(t, "testdata/local-write.json.tmpl", map[string]any{
"Name": instanceRepoName,
"SyncEnabled": true,
"SyncTarget": "instance",
})
_, err := helper.Repositories.Resource.Create(ctx, instanceRepo, metav1.CreateOptions{FieldValidation: "Strict"})
require.Error(t, err, "instance sync repository should be rejected when any other repository exists")
// Verify the error message mentions that instance can only be created when no other repositories exist
statusError := helper.RequireApiErrorStatus(err, metav1.StatusReasonInvalid, http.StatusUnprocessableEntity)
require.Contains(t, statusError.Message, "Instance repository can only be created when no other repositories exist. Found: "+existingFolderName)
// Clean up at end of test
helper.CleanupAllRepos(t)
})
t.Run("multiple folder syncs are allowed", func(t *testing.T) {
// Ensure clean state
helper.CleanupAllRepos(t)
firstFolderName := "folder-repo-multi-1"
secondFolderName := "folder-repo-multi-2"
// Create first folder sync repository
folderTestRepo1 := TestRepo{
Name: firstFolderName,
Target: "folder",
Copies: map[string]string{}, // No files needed for this test
ExpectedDashboards: 0,
ExpectedFolders: 1, // One folder expected after sync
}
helper.CreateRepo(t, folderTestRepo1)
// Create second folder sync repository - should succeed
folderTestRepo2 := TestRepo{
Name: secondFolderName,
Target: "folder",
Copies: map[string]string{}, // No files needed for this test
ExpectedDashboards: 0,
ExpectedFolders: 2, // Two folders expected after sync (1 + 1)
}
helper.CreateRepo(t, folderTestRepo2)
// Clean up at end of test
helper.CleanupAllRepos(t)
})
originalName := "original-repo"
// Create instance sync repository first
originalRepo := TestRepo{
Name: originalName,
Target: "instance",
Copies: map[string]string{}, // No files needed for this test
ExpectedDashboards: 0,
ExpectedFolders: 0,
}
helper.CreateRepo(t, originalRepo)
t.Run("folder sync is rejected when instance sync exists", func(t *testing.T) {
// Ensure clean state
helper.CleanupAllRepos(t)
instanceRepoName := "instance-blocking-folder"
folderRepoName := "folder-blocked-by-instance"
// Create instance sync repository first
instanceTestRepo := TestRepo{
Name: instanceRepoName,
Target: "instance",
Copies: map[string]string{}, // No files needed for this test
ExpectedDashboards: 0,
ExpectedFolders: 0,
}
helper.CreateRepo(t, instanceTestRepo)
// Try to create folder sync repository - should fail
folderRepo := helper.RenderObject(t, "testdata/local-write.json.tmpl", map[string]any{
"Name": folderRepoName,
"Name": "folder-blocked-by-instance",
"SyncEnabled": true,
"SyncTarget": "folder",
})
@@ -500,55 +383,56 @@ func TestIntegrationProvisioning_InstanceSyncValidation(t *testing.T) {
// Verify the error message mentions the existing instance repository
statusError := helper.RequireApiErrorStatus(err, metav1.StatusReasonInvalid, http.StatusUnprocessableEntity)
require.Contains(t, statusError.Message, "Cannot create folder repository when instance repository exists: "+instanceRepoName)
// Clean up at end of test
helper.CleanupAllRepos(t)
require.Contains(t, statusError.Message, "Cannot create folder repository when instance repository exists: "+originalName)
})
t.Run("instance sync can only be created when no repositories exist", func(t *testing.T) {
// Ensure clean state
helper.CleanupAllRepos(t)
t.Run("change between folder and instance sync for the same repository if no previous sync happened", func(t *testing.T) {
repo, err := helper.Repositories.Resource.Get(ctx, originalName, metav1.GetOptions{})
require.NoError(t, err, "failed to get repository")
err = unstructured.SetNestedField(repo.Object, "folder", "spec", "sync", "target")
require.NoError(t, err, "failed to set syncTarget to folder")
_, err = helper.Repositories.Resource.Update(ctx, repo, metav1.UpdateOptions{FieldValidation: "Strict"})
require.NoError(t, err, "failed to update repository to folder sync")
// This test verifies that instance sync can ONLY be created when there are no other repositories
instanceRepoName := "instance-only-when-empty"
// Verify that the repository is now a folder sync
// We verify with the listing APIs because it may take some time for the update to propagate
require.Eventually(t, func() bool {
repos, err := helper.Repositories.Resource.List(ctx, metav1.ListOptions{})
if err != nil {
return false
}
// First, create instance sync repository when no other repositories exist - should succeed
instanceTestRepo := TestRepo{
Name: instanceRepoName,
Target: "instance",
Copies: map[string]string{}, // No files needed for this test
ExpectedDashboards: 0,
ExpectedFolders: 0,
}
helper.CreateRepo(t, instanceTestRepo)
for _, repo := range repos.Items {
if repo.GetName() == originalName {
syncTarget, found, err := unstructured.NestedString(repo.Object, "spec", "sync", "target")
if err != nil || !found {
return false
}
// Now try to create any other repository - should fail
otherRepoName := "other-repo-blocked"
otherRepo := helper.RenderObject(t, "testdata/local-write.json.tmpl", map[string]any{
"Name": otherRepoName,
return syncTarget == "folder"
}
}
return false
}, time.Second*10, time.Millisecond*100, "failed to verify that sync target is folder")
})
t.Run("instance sync rejected when any other repository exists", func(t *testing.T) {
instanceRepo := helper.RenderObject(t, "testdata/local-write.json.tmpl", map[string]any{
"Name": "instance-repo-blocked",
"SyncEnabled": true,
"SyncTarget": "folder",
"SyncTarget": "instance",
})
_, err := helper.Repositories.Resource.Create(ctx, otherRepo, metav1.CreateOptions{FieldValidation: "Strict"})
require.Error(t, err, "folder sync repository should be rejected when instance sync exists")
_, err := helper.Repositories.Resource.Create(ctx, instanceRepo, metav1.CreateOptions{FieldValidation: "Strict"})
require.Error(t, err, "instance sync repository should be rejected when any other repository exists")
statusError := helper.RequireApiErrorStatus(err, metav1.StatusReasonInvalid, http.StatusUnprocessableEntity)
require.Contains(t, statusError.Message, "Cannot create folder repository when instance repository exists: "+instanceRepoName)
// Clean up at end of test
helper.CleanupAllRepos(t)
require.Contains(t, statusError.Message, "Instance repository can only be created when no other repositories exist. Found: "+originalName)
})
t.Run("repository limit validation", func(t *testing.T) {
// Ensure clean state
helper.CleanupAllRepos(t)
// This test verifies the 10 repository limit validation by actually creating 10 repositories
// Create 10 repositories - should all succeed
for i := 1; i <= 10; i++ {
t.Run("repository limit validation of 10 for folder syncs repositories", func(t *testing.T) {
for i := 2; i <= 10; i++ {
repoName := fmt.Sprintf("limit-test-repo-%d", i)
limitTestRepo := TestRepo{
Name: repoName,
@@ -571,12 +455,8 @@ func TestIntegrationProvisioning_InstanceSyncValidation(t *testing.T) {
_, err := helper.Repositories.Resource.Create(ctx, eleventhRepo, metav1.CreateOptions{FieldValidation: "Strict"})
require.Error(t, err, "11th repository should be rejected due to limit")
// Verify the error message mentions the repository limit
statusError := helper.RequireApiErrorStatus(err, metav1.StatusReasonInvalid, http.StatusUnprocessableEntity)
require.Contains(t, statusError.Message, "Maximum number of 10 repositories reached")
// Clean up at end of test
helper.CleanupAllRepos(t)
})
}
@@ -593,11 +473,7 @@ func TestIntegrationProvisioning_RunLocalRepository(t *testing.T) {
const targetPath = "all-panels.json"
// Set up the repository.
localTmp := helper.RenderObject(t, "testdata/local-write.json.tmpl", map[string]any{"Name": repo})
obj, err := helper.Repositories.Resource.Create(ctx, localTmp, metav1.CreateOptions{})
require.NoError(t, err)
name, _, _ := unstructured.NestedString(obj.Object, "metadata", "name")
require.Equal(t, repo, name, "wrote the expected name")
helper.CreateRepo(t, TestRepo{Name: repo})
// Write a file -- this will create it *both* in the local file system, and in grafana
t.Run("write all panels", func(t *testing.T) {
@@ -702,7 +578,7 @@ func TestIntegrationProvisioning_RunLocalRepository(t *testing.T) {
require.Error(t, result.Error(), "invalid path should return error")
// Read a file with a bad path
_, err = helper.Repositories.Resource.Get(ctx, repo, metav1.GetOptions{}, "files", "../../all-panels.json")
_, err := helper.Repositories.Resource.Get(ctx, repo, metav1.GetOptions{}, "files", "../../all-panels.json")
require.Error(t, err, "invalid path should error")
})
@@ -742,7 +618,7 @@ spec:
err = json.Unmarshal(raw, obj)
require.NoError(t, err)
name, _, _ = unstructured.NestedString(obj.Object, "resource", "upsert", "metadata", "name")
name, _, _ := unstructured.NestedString(obj.Object, "resource", "upsert", "metadata", "name")
require.True(t, strings.HasPrefix(name, "prefix-"), "should generate name")
})
}
@@ -763,15 +639,15 @@ func TestIntegrationProvisioning_ImportAllPanelsFromLocalRepository(t *testing.T
const repo = "local-tmp"
// Set up the repository and the file to import.
helper.CopyToProvisioningPath(t, "testdata/all-panels.json", "all-panels.json")
localTmp := helper.RenderObject(t, "testdata/local-write.json.tmpl", map[string]any{
"Name": repo,
"SyncEnabled": true,
})
testRepo := TestRepo{
Name: repo,
Target: "instance",
Copies: map[string]string{"testdata/all-panels.json": "all-panels.json"},
ExpectedDashboards: 1,
ExpectedFolders: 0,
}
// We create the repository
_, err = helper.Repositories.Resource.Create(ctx, localTmp, metav1.CreateOptions{})
require.NoError(t, err)
helper.CreateRepo(t, testRepo)
// Now, we import it, such that it may exist
// The sync may not be necessary as the sync may have happened automatically at this point
@@ -1050,6 +1050,12 @@ export type RepositorySpec = {
export type HealthStatus = {
/** When the health was checked last time */
checked?: number;
/** The type of the error
Possible enum values:
- `"health"`
- `"hook"` */
error?: 'health' | 'hook';
/** When not healthy, requests will not be executed */
healthy: boolean;
/** Summary messages (can be shown to users) Will only be populated when not healthy */