Gate agent behind a feature flag, valide mode flags (#9620)

Signed-off-by: Julien Pivotto <roidelapluie@inuits.eu>

cmd/prometheus/main.go

@@ -132,7 +132,7 @@ func extractFlagName(pc *kingpin.ParseContext) string {
 		if !ok {
 			continue
 		}
-		return fc.Model().Name
+		return "--" + fc.Model().Name
 	}
 	panic("extractFlagName not called from a kingpin PreAction. This is a bug, please report to Prometheus.")
 }
@@ -200,6 +200,9 @@ func (c *flagConfig) setFeatureListOptions(logger log.Logger) error {
 			case "new-service-discovery-manager":
 				c.enableNewSDManager = true
 				level.Info(logger).Log("msg", "Experimental service discovery manager")
+			case "agent":
+				agentMode = true
+				level.Info(logger).Log("msg", "Experimental agent mode enabled.")
 			case "":
 				continue
 			default:
@@ -238,8 +241,6 @@ func main() {
 
 	a.HelpFlag.Short('h')
 
-	a.Flag("agent", "Agent mode.").BoolVar(&agentMode)
-
 	a.Flag("config.file", "Prometheus configuration file path.").
 		Default("prometheus.yml").StringVar(&cfg.configFile)
 
@@ -335,16 +336,16 @@ func main() {
 		PreAction(agentOnlySetting()).
 		Default("data-agent/").StringVar(&cfg.localStoragePath)
 
-	a.Flag("storage.agent.segment-size",
+	a.Flag("storage.agent.wal-segment-size",
 		"Size at which to split WAL segment files. Example: 100MB").
 		PreAction(agentOnlySetting()).
 		Hidden().PlaceHolder("<bytes>").BytesVar(&cfg.agent.WALSegmentSize)
 
-	a.Flag("storage.agent.compression", "Compress the agent WAL.").
+	a.Flag("storage.agent.wal-compression", "Compress the agent WAL.").
 		PreAction(agentOnlySetting()).
 		Default("true").BoolVar(&cfg.agent.WALCompression)
 
-	a.Flag("storage.agent.truncate-frequency",
+	a.Flag("storage.agent.wal-truncate-frequency",
 		"The frequency at which to truncate the WAL and remove old data.").
 		PreAction(agentOnlySetting()).
 		Hidden().PlaceHolder("<duration>").SetValue(&cfg.agent.TruncateFrequency)
@@ -434,6 +435,16 @@ func main() {
 		os.Exit(1)
 	}
 
+	if agentMode && len(serverOnlyFlags) > 0 {
+		fmt.Fprintf(os.Stderr, "The following flag(s) can not be used in agent mode: %q", serverOnlyFlags)
+		os.Exit(3)
+	}
+
+	if !agentMode && len(agentOnlyFlags) > 0 {
+		fmt.Fprintf(os.Stderr, "The following flag(s) can only be used in agent mode: %q", agentOnlyFlags)
+		os.Exit(3)
+	}
+
 	cfg.web.ExternalURL, err = computeExternalURL(cfg.prometheusURL, cfg.web.ListenAddress)
 	if err != nil {
 		fmt.Fprintln(os.Stderr, errors.Wrapf(err, "parse external URL %q", cfg.prometheusURL))
@@ -1006,7 +1017,7 @@ func main() {
 				level.Info(logger).Log("msg", "Starting WAL storage ...")
 				if cfg.agent.WALSegmentSize != 0 {
 					if cfg.agent.WALSegmentSize < 10*1024*1024 || cfg.agent.WALSegmentSize > 256*1024*1024 {
-						return errors.New("flag 'storage.agent.segment-size' must be set between 10MB and 256MB")
+						return errors.New("flag 'storage.agent.wal-segment-size' must be set between 10MB and 256MB")
 					}
 				}
 				db, err := agent.Open(

cmd/prometheus/main_test.go

@@ -350,7 +350,7 @@ func getCurrentGaugeValuesFor(t *testing.T, reg prometheus.Gatherer, metricNames
 }
 
 func TestAgentSuccessfulStartup(t *testing.T) {
-	prom := exec.Command(promPath, "-test.main", "--agent", "--config.file="+agentConfig)
+	prom := exec.Command(promPath, "-test.main", "--enable-feature=agent", "--config.file="+agentConfig)
 	err := prom.Start()
 	require.NoError(t, err)
 
@@ -370,7 +370,7 @@ func TestAgentSuccessfulStartup(t *testing.T) {
 }
 
 func TestAgentStartupWithInvalidConfig(t *testing.T) {
-	prom := exec.Command(promPath, "-test.main", "--agent", "--config.file="+promConfig)
+	prom := exec.Command(promPath, "-test.main", "--enable-feature=agent", "--config.file="+promConfig)
 	err := prom.Start()
 	require.NoError(t, err)
 
@@ -388,3 +388,66 @@ func TestAgentStartupWithInvalidConfig(t *testing.T) {
 	}
 	require.Equal(t, expectedExitStatus, actualExitStatus)
 }
+
+func TestModeSpecificFlags(t *testing.T) {
+	if testing.Short() {
+		t.Skip("skipping test in short mode.")
+	}
+
+	testcases := []struct {
+		mode       string
+		arg        string
+		exitStatus int
+	}{
+		{"agent", "--storage.agent.path", 0},
+		{"server", "--storage.tsdb.path", 0},
+		{"server", "--storage.agent.path", 3},
+		{"agent", "--storage.tsdb.path", 3},
+	}
+
+	for _, tc := range testcases {
+		t.Run(fmt.Sprintf("%s mode with option %s", tc.mode, tc.arg), func(t *testing.T) {
+			args := []string{"-test.main", tc.arg, t.TempDir()}
+
+			if tc.mode == "agent" {
+				args = append(args, "--enable-feature=agent", "--config.file="+agentConfig)
+			} else {
+				args = append(args, "--config.file="+promConfig)
+			}
+
+			prom := exec.Command(promPath, args...)
+
+			// Log stderr in case of failure.
+			stderr, err := prom.StderrPipe()
+			require.NoError(t, err)
+			go func() {
+				slurp, _ := ioutil.ReadAll(stderr)
+				t.Log(string(slurp))
+			}()
+
+			err = prom.Start()
+			require.NoError(t, err)
+
+			if tc.exitStatus == 0 {
+				done := make(chan error, 1)
+				go func() { done <- prom.Wait() }()
+				select {
+				case err := <-done:
+					t.Errorf("prometheus should be still running: %v", err)
+				case <-time.After(5 * time.Second):
+					prom.Process.Kill()
+				}
+				return
+			}
+
+			err = prom.Wait()
+			require.Error(t, err)
+			if exitError, ok := err.(*exec.ExitError); ok {
+				status := exitError.Sys().(syscall.WaitStatus)
+				require.Equal(t, tc.exitStatus, status.ExitStatus())
+			} else {
+				t.Errorf("unable to retrieve the exit status for prometheus: %v", err)
+			}
+		})
+	}
+}

docs/feature_flags.md

@@ -86,3 +86,13 @@ issues upstream.
 
 In future releases, this new service discovery manager will become the default and
 this feature flag will be ignored.
+
+## Prometheus agent
+
+`--enable-feature=agent`
+
+When enabled, Prometheus runs in agent mode. The agent mode is limited to
+discovery, scrape and remote write.
+
+This is useful when you do not need to query the Prometheus data locally, but
+only from a central [remote endpoint](https://prometheus.io/docs/operating/integrations/#remote-endpoints-and-storage).