Alerting: Feature toggle to disallow sending alerts externally (#87982)
* Define feature toggle * Implement feature toggle
This commit is contained in:
@@ -16,6 +16,7 @@ import (
|
||||
"github.com/grafana/grafana/pkg/api/datasource"
|
||||
"github.com/grafana/grafana/pkg/infra/log"
|
||||
"github.com/grafana/grafana/pkg/services/datasources"
|
||||
"github.com/grafana/grafana/pkg/services/featuremgmt"
|
||||
"github.com/grafana/grafana/pkg/services/ngalert/api/tooling/definitions"
|
||||
"github.com/grafana/grafana/pkg/services/ngalert/models"
|
||||
"github.com/grafana/grafana/pkg/services/ngalert/notifier"
|
||||
@@ -48,11 +49,12 @@ type AlertsRouter struct {
|
||||
|
||||
datasourceService datasources.DataSourceService
|
||||
secretService secrets.Service
|
||||
featureManager featuremgmt.FeatureToggles
|
||||
}
|
||||
|
||||
func NewAlertsRouter(multiOrgNotifier *notifier.MultiOrgAlertmanager, store store.AdminConfigurationStore,
|
||||
clk clock.Clock, appURL *url.URL, disabledOrgs map[int64]struct{}, configPollInterval time.Duration,
|
||||
datasourceService datasources.DataSourceService, secretService secrets.Service) *AlertsRouter {
|
||||
datasourceService datasources.DataSourceService, secretService secrets.Service, featureManager featuremgmt.FeatureToggles) *AlertsRouter {
|
||||
d := &AlertsRouter{
|
||||
logger: log.New("ngalert.sender.router"),
|
||||
clock: clk,
|
||||
@@ -71,13 +73,14 @@ func NewAlertsRouter(multiOrgNotifier *notifier.MultiOrgAlertmanager, store stor
|
||||
|
||||
datasourceService: datasourceService,
|
||||
secretService: secretService,
|
||||
featureManager: featureManager,
|
||||
}
|
||||
return d
|
||||
}
|
||||
|
||||
// SyncAndApplyConfigFromDatabase looks for the admin configuration in the database
|
||||
// and adjusts the sender(s) and alert handling mechanism accordingly.
|
||||
func (d *AlertsRouter) SyncAndApplyConfigFromDatabase() error {
|
||||
func (d *AlertsRouter) SyncAndApplyConfigFromDatabase(ctx context.Context) error {
|
||||
cfgs, err := d.adminConfigStore.GetAdminConfigurations()
|
||||
if err != nil {
|
||||
return err
|
||||
@@ -85,6 +88,8 @@ func (d *AlertsRouter) SyncAndApplyConfigFromDatabase() error {
|
||||
|
||||
d.logger.Debug("Attempting to sync admin configs", "count", len(cfgs))
|
||||
|
||||
disableExternal := d.featureManager.IsEnabled(ctx, featuremgmt.FlagAlertingDisableSendAlertsExternal)
|
||||
|
||||
orgsFound := make(map[int64]struct{}, len(cfgs))
|
||||
d.adminConfigMtx.Lock()
|
||||
for _, cfg := range cfgs {
|
||||
@@ -93,6 +98,11 @@ func (d *AlertsRouter) SyncAndApplyConfigFromDatabase() error {
|
||||
continue
|
||||
}
|
||||
|
||||
if disableExternal && cfg.SendAlertsTo != models.InternalAlertmanager {
|
||||
d.logger.Warn("Alertmanager choice in configuration will be ignored due to feature flags", "org", cfg.OrgID, "choice", cfg.SendAlertsTo)
|
||||
cfg.SendAlertsTo = models.InternalAlertmanager
|
||||
}
|
||||
|
||||
// Update the Alertmanagers choice for the organization.
|
||||
d.sendAlertsTo[cfg.OrgID] = cfg.SendAlertsTo
|
||||
|
||||
@@ -363,7 +373,7 @@ func (d *AlertsRouter) Run(ctx context.Context) error {
|
||||
for {
|
||||
select {
|
||||
case <-time.After(d.adminConfigPollInterval):
|
||||
if err := d.SyncAndApplyConfigFromDatabase(); err != nil {
|
||||
if err := d.SyncAndApplyConfigFromDatabase(ctx); err != nil {
|
||||
d.logger.Error("Unable to sync admin configuration", "error", err)
|
||||
}
|
||||
case <-ctx.Done():
|
||||
|
||||
@@ -63,14 +63,14 @@ func TestIntegrationSendingToExternalAlertmanager(t *testing.T) {
|
||||
}),
|
||||
}
|
||||
alertsRouter := NewAlertsRouter(moa, fakeAdminConfigStore, mockedClock, appUrl, map[int64]struct{}{}, 10*time.Minute,
|
||||
&fake_ds.FakeDataSourceService{DataSources: []*datasources.DataSource{&ds1}}, fake_secrets.NewFakeSecretsService())
|
||||
&fake_ds.FakeDataSourceService{DataSources: []*datasources.DataSource{&ds1}}, fake_secrets.NewFakeSecretsService(), featuremgmt.WithFeatures())
|
||||
|
||||
mockedGetAdminConfigurations.Return([]*models.AdminConfiguration{
|
||||
{OrgID: ruleKey.OrgID, SendAlertsTo: models.AllAlertmanagers},
|
||||
}, nil)
|
||||
// Make sure we sync the configuration at least once before the evaluation happens to guarantee the sender is running
|
||||
// when the first alert triggers.
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase())
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
require.Equal(t, 1, len(alertsRouter.externalAlertmanagers))
|
||||
require.Equal(t, 1, len(alertsRouter.externalAlertmanagersCfgHash))
|
||||
|
||||
@@ -93,7 +93,7 @@ func TestIntegrationSendingToExternalAlertmanager(t *testing.T) {
|
||||
// Now, let's remove the Alertmanager from the admin configuration.
|
||||
mockedGetAdminConfigurations.Return(nil, nil)
|
||||
// Again, make sure we sync and verify the externalAlertmanagers.
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase())
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
require.Equal(t, 0, len(alertsRouter.externalAlertmanagers))
|
||||
require.Equal(t, 0, len(alertsRouter.externalAlertmanagersCfgHash))
|
||||
|
||||
@@ -134,7 +134,7 @@ func TestIntegrationSendingToExternalAlertmanager_WithMultipleOrgs(t *testing.T)
|
||||
}
|
||||
fakeDs := &fake_ds.FakeDataSourceService{DataSources: []*datasources.DataSource{&ds1}}
|
||||
alertsRouter := NewAlertsRouter(moa, fakeAdminConfigStore, mockedClock, appUrl, map[int64]struct{}{}, 10*time.Minute,
|
||||
fakeDs, fake_secrets.NewFakeSecretsService())
|
||||
fakeDs, fake_secrets.NewFakeSecretsService(), featuremgmt.WithFeatures())
|
||||
|
||||
mockedGetAdminConfigurations.Return([]*models.AdminConfiguration{
|
||||
{OrgID: ruleKey1.OrgID, SendAlertsTo: models.AllAlertmanagers},
|
||||
@@ -142,7 +142,7 @@ func TestIntegrationSendingToExternalAlertmanager_WithMultipleOrgs(t *testing.T)
|
||||
|
||||
// Make sure we sync the configuration at least once before the evaluation happens to guarantee the sender is running
|
||||
// when the first alert triggers.
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase())
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
require.Equal(t, 1, len(alertsRouter.externalAlertmanagers))
|
||||
require.Equal(t, 1, len(alertsRouter.externalAlertmanagersCfgHash))
|
||||
|
||||
@@ -167,7 +167,7 @@ func TestIntegrationSendingToExternalAlertmanager_WithMultipleOrgs(t *testing.T)
|
||||
}, nil)
|
||||
|
||||
// If we sync again, new externalAlertmanagers must have spawned.
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase())
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
require.Equal(t, 2, len(alertsRouter.externalAlertmanagers))
|
||||
require.Equal(t, 2, len(alertsRouter.externalAlertmanagersCfgHash))
|
||||
|
||||
@@ -216,7 +216,7 @@ func TestIntegrationSendingToExternalAlertmanager_WithMultipleOrgs(t *testing.T)
|
||||
currentHash := alertsRouter.externalAlertmanagersCfgHash[ruleKey2.OrgID]
|
||||
|
||||
// Now, sync again.
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase())
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
|
||||
// The hash for org two should not be the same and we should still have two externalAlertmanagers.
|
||||
require.NotEqual(t, alertsRouter.externalAlertmanagersCfgHash[ruleKey2.OrgID], currentHash)
|
||||
@@ -236,7 +236,7 @@ func TestIntegrationSendingToExternalAlertmanager_WithMultipleOrgs(t *testing.T)
|
||||
currentHash = alertsRouter.externalAlertmanagersCfgHash[ruleKey1.OrgID]
|
||||
|
||||
// Now, sync again.
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase())
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
|
||||
// The old configuration should not be running.
|
||||
require.NotEqual(t, alertsRouter.externalAlertmanagersCfgHash[ruleKey1.OrgID], currentHash)
|
||||
@@ -249,13 +249,13 @@ func TestIntegrationSendingToExternalAlertmanager_WithMultipleOrgs(t *testing.T)
|
||||
{OrgID: ruleKey2.OrgID},
|
||||
}, nil)
|
||||
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase())
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
require.NotEqual(t, alertsRouter.externalAlertmanagersCfgHash[ruleKey1.OrgID], currentHash)
|
||||
|
||||
// Finally, remove everything.
|
||||
mockedGetAdminConfigurations.Return([]*models.AdminConfiguration{}, nil)
|
||||
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase())
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
|
||||
require.Equal(t, 0, len(alertsRouter.externalAlertmanagers))
|
||||
require.Equal(t, 0, len(alertsRouter.externalAlertmanagersCfgHash))
|
||||
@@ -293,14 +293,14 @@ func TestChangingAlertmanagersChoice(t *testing.T) {
|
||||
}),
|
||||
}
|
||||
alertsRouter := NewAlertsRouter(moa, fakeAdminConfigStore, mockedClock, appUrl, map[int64]struct{}{},
|
||||
10*time.Minute, &fake_ds.FakeDataSourceService{DataSources: []*datasources.DataSource{&ds}}, fake_secrets.NewFakeSecretsService())
|
||||
10*time.Minute, &fake_ds.FakeDataSourceService{DataSources: []*datasources.DataSource{&ds}}, fake_secrets.NewFakeSecretsService(), featuremgmt.WithFeatures())
|
||||
|
||||
mockedGetAdminConfigurations.Return([]*models.AdminConfiguration{
|
||||
{OrgID: ruleKey.OrgID, SendAlertsTo: models.AllAlertmanagers},
|
||||
}, nil)
|
||||
// Make sure we sync the configuration at least once before the evaluation happens to guarantee the sender is running
|
||||
// when the first alert triggers.
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase())
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
require.Equal(t, 1, len(alertsRouter.externalAlertmanagers))
|
||||
require.Equal(t, 1, len(alertsRouter.externalAlertmanagersCfgHash))
|
||||
require.Equal(t, models.AllAlertmanagers, alertsRouter.sendAlertsTo[ruleKey.OrgID])
|
||||
@@ -325,7 +325,7 @@ func TestChangingAlertmanagersChoice(t *testing.T) {
|
||||
{OrgID: ruleKey.OrgID, SendAlertsTo: models.ExternalAlertmanagers},
|
||||
}, nil)
|
||||
// Again, make sure we sync and verify the externalAlertmanagers.
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase())
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
require.Equal(t, 1, len(alertsRouter.externalAlertmanagers))
|
||||
require.Equal(t, 1, len(alertsRouter.externalAlertmanagersCfgHash))
|
||||
|
||||
@@ -339,7 +339,7 @@ func TestChangingAlertmanagersChoice(t *testing.T) {
|
||||
|
||||
// Again, make sure we sync and verify the externalAlertmanagers.
|
||||
// externalAlertmanagers should be running even though alerts are being handled externally.
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase())
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
require.Equal(t, 1, len(alertsRouter.externalAlertmanagers))
|
||||
require.Equal(t, 1, len(alertsRouter.externalAlertmanagersCfgHash))
|
||||
|
||||
@@ -356,6 +356,86 @@ func TestChangingAlertmanagersChoice(t *testing.T) {
|
||||
require.Len(t, actualAlerts, len(expected))
|
||||
}
|
||||
|
||||
func TestAlertmanagersChoiceWithDisableExternalFeatureToggle(t *testing.T) {
|
||||
ruleKey := models.GenerateRuleKey(1)
|
||||
|
||||
fakeAM := NewFakeExternalAlertmanager(t)
|
||||
defer fakeAM.Close()
|
||||
|
||||
fakeAdminConfigStore := &store.AdminConfigurationStoreMock{}
|
||||
mockedGetAdminConfigurations := fakeAdminConfigStore.EXPECT().GetAdminConfigurations()
|
||||
|
||||
mockedClock := clock.NewMock()
|
||||
mockedClock.Set(time.Now())
|
||||
|
||||
moa := createMultiOrgAlertmanager(t, []int64{1})
|
||||
|
||||
appUrl := &url.URL{
|
||||
Scheme: "http",
|
||||
Host: "localhost",
|
||||
}
|
||||
|
||||
ds := datasources.DataSource{
|
||||
URL: fakeAM.Server.URL,
|
||||
OrgID: ruleKey.OrgID,
|
||||
Type: datasources.DS_ALERTMANAGER,
|
||||
JsonData: simplejson.NewFromAny(map[string]any{
|
||||
"handleGrafanaManagedAlerts": true,
|
||||
"implementation": "prometheus",
|
||||
}),
|
||||
}
|
||||
|
||||
var expected []*models2.PostableAlert
|
||||
alerts := definitions.PostableAlerts{}
|
||||
for i := 0; i < rand.Intn(5)+1; i++ {
|
||||
alert := generatePostableAlert(t, mockedClock)
|
||||
expected = append(expected, &alert)
|
||||
alerts.PostableAlerts = append(alerts.PostableAlerts, alert)
|
||||
}
|
||||
|
||||
alertsRouter := NewAlertsRouter(moa, fakeAdminConfigStore, mockedClock, appUrl, map[int64]struct{}{},
|
||||
10*time.Minute, &fake_ds.FakeDataSourceService{DataSources: []*datasources.DataSource{&ds}},
|
||||
fake_secrets.NewFakeSecretsService(), featuremgmt.WithFeatures(featuremgmt.FlagAlertingDisableSendAlertsExternal))
|
||||
|
||||
// Test that we only send to the internal Alertmanager even though the configuration specifies AllAlertmanagers.
|
||||
|
||||
mockedGetAdminConfigurations.Return([]*models.AdminConfiguration{
|
||||
{OrgID: ruleKey.OrgID, SendAlertsTo: models.AllAlertmanagers},
|
||||
}, nil)
|
||||
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
require.Equal(t, 0, len(alertsRouter.externalAlertmanagers))
|
||||
require.Equal(t, 0, len(alertsRouter.externalAlertmanagersCfgHash))
|
||||
require.Equal(t, models.InternalAlertmanager, alertsRouter.sendAlertsTo[ruleKey.OrgID])
|
||||
|
||||
alertsRouter.Send(context.Background(), ruleKey, alerts)
|
||||
|
||||
am, err := moa.AlertmanagerFor(ruleKey.OrgID)
|
||||
require.NoError(t, err)
|
||||
actualAlerts, err := am.GetAlerts(context.Background(), true, true, true, nil, "")
|
||||
require.NoError(t, err)
|
||||
require.Len(t, actualAlerts, len(expected))
|
||||
|
||||
// Test that we still only send to the internal alertmanager even though the configuration specifies ExternalAlertmanagers.
|
||||
|
||||
mockedGetAdminConfigurations.Return([]*models.AdminConfiguration{
|
||||
{OrgID: ruleKey.OrgID, SendAlertsTo: models.ExternalAlertmanagers},
|
||||
}, nil)
|
||||
|
||||
require.NoError(t, alertsRouter.SyncAndApplyConfigFromDatabase(context.Background()))
|
||||
require.Equal(t, 0, len(alertsRouter.externalAlertmanagers))
|
||||
require.Equal(t, 0, len(alertsRouter.externalAlertmanagersCfgHash))
|
||||
require.Equal(t, models.InternalAlertmanager, alertsRouter.sendAlertsTo[ruleKey.OrgID])
|
||||
|
||||
alertsRouter.Send(context.Background(), ruleKey, alerts)
|
||||
|
||||
am, err = moa.AlertmanagerFor(ruleKey.OrgID)
|
||||
require.NoError(t, err)
|
||||
actualAlerts, err = am.GetAlerts(context.Background(), true, true, true, nil, "")
|
||||
require.NoError(t, err)
|
||||
require.Len(t, actualAlerts, len(expected))
|
||||
}
|
||||
|
||||
func assertAlertmanagersStatusForOrg(t *testing.T, alertsRouter *AlertsRouter, orgID int64, active, dropped int) {
|
||||
t.Helper()
|
||||
require.Eventuallyf(t, func() bool {
|
||||
|
||||
Reference in New Issue
Block a user