Cleanup obsolete local files for alertmanager. (#3910)

* Cleanup obsolete local files for alertmanager.

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* CHANGELOG.md

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* Comment.

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* Don't ignore directories. Log error when deletion fails instead.

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* Address review feedback.

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* Move per-tenant state into tenant directory to simplify cleanup.

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* Move migration to separate function.
Add test for migration.
Fix test for deletion of unused dirs.

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* Store templates to correct place.

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* CHANGELOG.md

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* Verify that templates are stored properly into correct location.

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* Comments.

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* Comments.

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* Apply suggestions from code review

Co-authored-by: Marco Pracucci <marco@pracucci.com>
Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

* Review feedback.

Signed-off-by: Peter Štibraný <peter.stibrany@grafana.com>

Co-authored-by: Marco Pracucci <marco@pracucci.com>

Author: pstibrany

CHANGELOG.md

@@ -2,6 +2,8 @@
 
 ## master / unreleased
 
+* [CHANGE] Alertmanager now removes local files after Alertmanager is no longer running for removed or resharded user. #3910
+* [CHANGE] Alertmanager now stores local files in per-tenant folders. Files stored by Alertmanager previously are migrated to new hierarchy. Support for this migration will be removed in Cortex 1.10. #3910
 * [ENHANCEMENT] Ruler: optimized `<prefix>/api/v1/rules` and `<prefix>/api/v1/alerts` when ruler sharding is enabled. #3916
 * [ENHANCEMENT] Ruler: added the following metrics when ruler sharding is enabled: #3916
   * `cortex_ruler_clients`

development/tsdb-blocks-storage-s3/config/rules.yaml

@@ -4,3 +4,13 @@ groups:
     rules:
       - record: up:count
         expr: count(up)
+
+  - name: example2
+    rules:
+      - alert: TooManyServices
+        expr: count(up) > 1
+        for: 1m
+        labels:
+          severity: page
+        annotations:
+          summary: Too many services

pkg/alertmanager/alertmanager.go

@@ -46,19 +46,28 @@ import (
 	"github.com/cortexproject/cortex/pkg/util/services"
 )
 
-const notificationLogMaintenancePeriod = 15 * time.Minute
+const (
+	// MaintenancePeriod is used for periodic storing of silences and notifications to local file.
+	maintenancePeriod = 15 * time.Minute
+
+	// Filenames used within tenant-directory
+	notificationLogSnapshot = "notifications"
+	silencesSnapshot        = "silences"
+	templatesDir            = "templates"
+)
 
 // Config configures an Alertmanager.
 type Config struct {
-	UserID string
-	// Used to persist notification logs and silences on disk.
-	DataDir     string
+	UserID      string
 	Logger      log.Logger
 	Peer        *cluster.Peer
 	PeerTimeout time.Duration
 	Retention   time.Duration
 	ExternalURL *url.URL
 
+	// Tenant-specific local directory where AM can store its state (notifications, silences, templates). When AM is stopped, entire dir is removed.
+	TenantDataDir string
+
 	ShardingEnabled    bool
 	ReplicationFactor  int
 	ReplicateStateFunc func(context.Context, string, *clusterpb.Part) error
@@ -118,6 +127,10 @@ type State interface {
 
 // New creates a new Alertmanager.
 func New(cfg *Config, reg *prometheus.Registry) (*Alertmanager, error) {
+	if cfg.TenantDataDir == "" {
+		return nil, fmt.Errorf("directory for tenant-specific AlertManager is not configured")
+	}
+
 	am := &Alertmanager{
 		cfg:    cfg,
 		logger: log.With(cfg.Logger, "user", cfg.UserID),
@@ -153,12 +166,11 @@ func New(cfg *Config, reg *prometheus.Registry) (*Alertmanager, error) {
 	}
 
 	am.wg.Add(1)
-	nflogID := fmt.Sprintf("nflog:%s", cfg.UserID)
 	var err error
 	am.nflog, err = nflog.New(
 		nflog.WithRetention(cfg.Retention),
-		nflog.WithSnapshot(filepath.Join(cfg.DataDir, nflogID)),
-		nflog.WithMaintenance(notificationLogMaintenancePeriod, am.stop, am.wg.Done),
+		nflog.WithSnapshot(filepath.Join(cfg.TenantDataDir, notificationLogSnapshot)),
+		nflog.WithMaintenance(maintenancePeriod, am.stop, am.wg.Done),
 		nflog.WithMetrics(am.registry),
 		nflog.WithLogger(log.With(am.logger, "component", "nflog")),
 	)
@@ -171,9 +183,9 @@ func New(cfg *Config, reg *prometheus.Registry) (*Alertmanager, error) {
 
 	am.marker = types.NewMarker(am.registry)
 
-	silencesID := fmt.Sprintf("silences:%s", cfg.UserID)
+	silencesFile := filepath.Join(cfg.TenantDataDir, silencesSnapshot)
 	am.silences, err = silence.New(silence.Options{
-		SnapshotFile: filepath.Join(cfg.DataDir, silencesID),
+		SnapshotFile: silencesFile,
 		Retention:    cfg.Retention,
 		Logger:       log.With(am.logger, "component", "silences"),
 		Metrics:      am.registry,
@@ -189,7 +201,7 @@ func New(cfg *Config, reg *prometheus.Registry) (*Alertmanager, error) {
 
 	am.wg.Add(1)
 	go func() {
-		am.silences.Maintenance(15*time.Minute, filepath.Join(cfg.DataDir, silencesID), am.stop)
+		am.silences.Maintenance(maintenancePeriod, silencesFile, am.stop)
 		am.wg.Done()
 	}()
 
@@ -249,7 +261,7 @@ func (am *Alertmanager) ApplyConfig(userID string, conf *config.Config, rawCfg s
 	templateFiles := make([]string, len(conf.Templates))
 	if len(conf.Templates) > 0 {
 		for i, t := range conf.Templates {
-			templateFiles[i] = filepath.Join(am.cfg.DataDir, "templates", userID, t)
+			templateFiles[i] = filepath.Join(am.cfg.TenantDataDir, templatesDir, t)
 		}
 	}
 

pkg/alertmanager/api.go

@@ -153,14 +153,14 @@ func validateUserConfig(logger log.Logger, cfg alertspb.AlertConfigDesc) error {
 	// not to configured data dir, and on the flipside, it'll fail if we can't write
 	// to tmpDir. Ignoring both cases for now as they're ultra rare but will revisit if
 	// we see this in the wild.
-	tmpDir, err := ioutil.TempDir("", "validate-config")
+	userTempDir, err := ioutil.TempDir("", "validate-config-"+cfg.User)
 	if err != nil {
 		return err
 	}
-	defer os.RemoveAll(tmpDir)
+	defer os.RemoveAll(userTempDir)
 
 	for _, tmpl := range cfg.Templates {
-		_, err := createTemplateFile(tmpDir, cfg.User, tmpl.Filename, tmpl.Body)
+		_, err := storeTemplateFile(userTempDir, tmpl.Filename, tmpl.Body)
 		if err != nil {
 			level.Error(logger).Log("msg", "unable to create template file", "err", err, "user", cfg.User)
 			return fmt.Errorf("unable to create template file '%s'", tmpl.Filename)
@@ -169,7 +169,7 @@ func validateUserConfig(logger log.Logger, cfg alertspb.AlertConfigDesc) error {
 
 	templateFiles := make([]string, len(amCfg.Templates))
 	for i, t := range amCfg.Templates {
-		templateFiles[i] = filepath.Join(tmpDir, "templates", cfg.User, t)
+		templateFiles[i] = filepath.Join(userTempDir, t)
 	}
 
 	_, err = template.FromGlobs(templateFiles...)