Make more robust to SNMP failures

.gitignore

@@ -1 +1,3 @@
 config.toml
+apc2mqtt
+.direnv

flake.lock

@@ -1,5 +1,26 @@
 {
   "nodes": {
+    "devshell": {
+      "inputs": {
+        "flake-utils": [
+          "flake-utils"
+        ],
+        "nixpkgs": "nixpkgs"
+      },
+      "locked": {
+        "lastModified": 1649691969,
+        "narHash": "sha256-nY1aUWIyh3TcGVo3sn+3vyCh+tOiEZL4JtMX3aOZSeY=",
+        "owner": "numtide",
+        "repo": "devshell",
+        "rev": "e22633b05fec2fe196888c593d4d9b3f4f648a25",
+        "type": "github"
+      },
+      "original": {
+        "owner": "numtide",
+        "repo": "devshell",
+        "type": "github"
+      }
+    },
     "flake-utils": {
       "locked": {
         "lastModified": 1637014545,
@@ -17,11 +38,27 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1637156900,
+        "lastModified": 1643381941,
-        "narHash": "sha256-nusyaSsL1RLyUEWufUUywDrGKMXw+4ugSSZ3ss8TSuw=",
+        "narHash": "sha256-pHTwvnN4tTsEKkWlXQ8JMY423epos8wUOhthpwJjtpc=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "12fc0f19fefa9dff68bc3e0938b815ab8d89df90",
+        "rev": "5efc8ca954272c4376ac929f4c5ffefcc20551d5",
+        "type": "github"
+      },
+      "original": {
+        "owner": "NixOS",
+        "ref": "nixpkgs-unstable",
+        "repo": "nixpkgs",
+        "type": "github"
+      }
+    },
+    "nixpkgs_2": {
+      "locked": {
+        "lastModified": 1675153841,
+        "narHash": "sha256-EWvU3DLq+4dbJiukfhS7r6sWZyJikgXn6kNl7eHljW8=",
+        "owner": "NixOS",
+        "repo": "nixpkgs",
+        "rev": "ea692c2ad1afd6384e171eabef4f0887d2b882d3",
         "type": "github"
       },
       "original": {
@@ -31,8 +68,9 @@
     },
     "root": {
       "inputs": {
+        "devshell": "devshell",
         "flake-utils": "flake-utils",
-        "nixpkgs": "nixpkgs"
+        "nixpkgs": "nixpkgs_2"
       }
     }
   },

flake.nix

@@ -2,10 +2,18 @@
   description = "Samw's APC PDU MQTT bridge";
 
   inputs.flake-utils.url = "github:numtide/flake-utils";
+  inputs.devshell = {
+    url = "github:numtide/devshell";
+    inputs.flake-utils.follows = "flake-utils";
+  };
 
-  outputs = { self, nixpkgs, flake-utils }:
+  outputs = { self, nixpkgs, flake-utils, devshell }:
     flake-utils.lib.eachDefaultSystem (system:
-      let pkgs = import nixpkgs { inherit system; };
+      let
+        pkgs = import nixpkgs {
+          inherit system;
+          overlays = [ devshell.overlay ];
+        };
       in rec {
         packages.apc2mqtt = pkgs.buildGoModule {
           name = "apc2mqtt";
@@ -16,5 +24,7 @@
           '';
         };
         defaultPackage = packages.apc2mqtt;
+        devShell =
+          pkgs.devshell.mkShell { packages = with pkgs; [ go gopls mosquitto ]; };
       });
 }

main.go

@@ -35,6 +35,8 @@ type config struct {
 	MQTT struct {
 		Host string
 		Port uint16
+		User string
+		Pass string
 	}
 	Targets []targetConfig
 }
@@ -76,6 +78,38 @@ type PDUCommand struct {
 	State  bool
 }
 
+func getPDUState(snmp *gosnmp.GoSNMP) (*PDUState, error) {
+	res, err := snmp.Get([]string{sPDUMasterConfigPDUName, sPDUIdentSerialNumber, sPDUIdentModelNumber})
+	if err != nil {
+		return nil, fmt.Errorf("get PDU info: %w", err)
+	}
+	state := PDUState{
+		Name:   string(res.Variables[0].Value.([]byte)),
+		Serial: string(res.Variables[1].Value.([]byte)),
+		Model:  string(res.Variables[2].Value.([]byte)),
+	}
+
+	outletNames, err := snmp.WalkAll(sPDUOutletName)
+	if err != nil {
+		return nil, fmt.Errorf("walk PDU outlet names: %w", err)
+	}
+
+	for _, val := range outletNames {
+		state.Outlets = append(state.Outlets, Outlet{
+			Name: string(val.Value.([]byte)),
+		})
+	}
+
+	outletStates, err := snmp.WalkAll(sPDUOutletCtl)
+	if err != nil {
+		return nil, fmt.Errorf("walk PDU outlet states: %w", err)
+	}
+	for i, val := range outletStates {
+		state.Outlets[i].State = val.Value.(int) == 1
+	}
+	return &state, nil
+}
+
 func runSNMP(host string, port uint16, stateCh chan PDUState, commandCh chan PDUCommand) {
 	snmp := &gosnmp.GoSNMP{
 		Target:    host,
@@ -86,36 +120,27 @@ func runSNMP(host string, port uint16, stateCh chan PDUState, commandCh chan PDU
 		Version:   gosnmp.Version1,
 		Retries:   3,
 	}
-	err := snmp.Connect()
+	// Connect loop
-	check(err)
+connect_loop:
+	for {
+		log.Info("Opening SNMP connection")
+		if err := snmp.Connect(); err != nil {
+			log.Warnf("Error connecting SNMP target %s: `%s`. Sleeping...", host, err.Error())
+			time.Sleep(time.Second * 5)
+			continue // Try again
+		}
 		defer snmp.Conn.Close()
 		poll := time.NewTicker(1 * time.Second)
 		for {
 			select {
 			case <-poll.C:
-			res, err := snmp.Get([]string{sPDUMasterConfigPDUName, sPDUIdentSerialNumber, sPDUIdentModelNumber})
+				state, err := getPDUState(snmp)
-			check(err)
+				if err != nil {
-			state := PDUState{
+					log.Errorf("getting pdu state: %s", err.Error())
-				Name:   string(res.Variables[0].Value.([]byte)),
+					continue connect_loop // Reconnect
-				Serial: string(res.Variables[1].Value.([]byte)),
-				Model:  string(res.Variables[2].Value.([]byte)),
 				}
-
+				log.Debugf("Got state: %s", state)
-			outletNames, err := snmp.WalkAll(sPDUOutletName)
+				stateCh <- *state
-			check(err)
-
-			for _, val := range outletNames {
-				state.Outlets = append(state.Outlets, Outlet{
-					Name: string(val.Value.([]byte)),
-				})
-			}
-
-			outletStates, err := snmp.WalkAll(sPDUOutletCtl)
-			check(err)
-			for i, val := range outletStates {
-				state.Outlets[i].State = val.Value.(int) == 1
-			}
-			stateCh <- state
 			case cmd := <-commandCh:
 				var value int
 				if cmd.State {
@@ -128,12 +153,15 @@ func runSNMP(host string, port uint16, stateCh chan PDUState, commandCh chan PDU
 					Name:  fmt.Sprintf("%s.%d", sPDUOutletCtl, cmd.Outlet),
 					Type:  gosnmp.Integer,
 				}})
-			check(err)
+				if err != nil {
+					log.Errorf("setting pdu state: %s", err.Error())
+				}
 				if res.Error != gosnmp.NoError {
 					log.Errorf("error in snmp set: %s", res.Error)
 				}
 			}
 		}
+	}
 }
 
 type HassDeviceConfig struct {
@@ -202,21 +230,33 @@ func spawnTarget(target targetConfig, mqttClient mqtt.Client) {
 }
 func main() {
 	var configpath = flag.String("conf", "config.toml", "Path to toml config file")
+	var verbose = flag.Bool("v", false, "Enable debug logging")
 	flag.Parse()
+	if *verbose {
+		log.SetLevel(log.DebugLevel)
+	}
 	log.Infof("Starting version %s", version)
 	conf, err := parseConfig(*configpath)
 	check(err)
 
 	mqttOpts := mqtt.NewClientOptions().AddBroker(
 		fmt.Sprintf("tcp://%s:%d", conf.MQTT.Host, conf.MQTT.Port),
-	).SetClientID("apc2mqtt")
+	).SetClientID("apc2mqtt").SetConnectTimeout(time.Second * 5)
+	if conf.MQTT.User != "" {
+		mqttOpts.SetUsername(conf.MQTT.User)
+	}
+	if conf.MQTT.Pass != "" {
+		mqttOpts.SetPassword(conf.MQTT.Pass)
+	}
 	mqttClient := mqtt.NewClient(mqttOpts)
+	for {
 		if token := mqttClient.Connect(); token.Wait() && token.Error() != nil {
-		panic(token.Error())
+			log.Fatalf("connect mqtt: %s", token.Error())
 		}
 
 		for _, target := range conf.Targets {
 			go spawnTarget(target, mqttClient)
 		}
 		select {}
+	}
 }