Merge pull request #30 from cloudflare/watchdog-fixes

Watchdog fixes

README.md

@@ -97,6 +97,11 @@ This option can also be set using `-alertmanager.ttl` flag. Example:
 
 Default is `1m`.
 
+Note that the maximum value for this option is `15m`.
+The UI has a watchdog that tracks the timestamp of the last pull. If the UI
+does not receive updates for more than 15 minutes it will print an error and
+reload the page.
+
 #### ALERTMANAGER_URI
 
 URI of the Alertmanager instance, unsee will use it to pull alert groups and

assets/static/unsee.js

@@ -26,7 +26,7 @@ var Unsee = (function(params) {
         Grid.Init();
         Autocomplete.Init();
         Filters.Init();
-        Watchdog.Init(10, 300);
+        Watchdog.Init(30, 60*15); // set watchdog to 15 minutes
 
         $(selectors.refreshButton).click(function() {
             if (!$(selectors.refreshButton).prop('disabled')) {
@@ -53,7 +53,6 @@ var Unsee = (function(params) {
             }
         }
         refreshInterval = rate;
-        Watchdog.UpdateTolerance(rate);
         Progress.Reset();
     }
 
@@ -131,10 +130,12 @@ var Unsee = (function(params) {
                             Colors.Update(resp['colors']);
                             Alerts.Update(resp);
                             updateCompleted();
-                            Watchdog.Pong(moment(resp['timestamp']).unix());
+                            Watchdog.Pong(moment(resp['timestamp']);
                             Unsee.WaitForNextReload();
-                            $(selectors.errors).html('');
-                            $(selectors.errors).hide('');
+                            if (!Watchdog.IsFatal()) {
+                              $(selectors.errors).html('');
+                              $(selectors.errors).hide('');
+                            }
                         } catch (err) {
                             Counter.Unknown();
                             handleError(err);

assets/static/watchdog.js

@@ -21,15 +21,13 @@ var Watchdog = (function() {
         // don't raise an error if autorefresh is disabled
         if (!Config.GetOption('autorefresh').Get()) return;
 
-        var now = moment().unix();
+        var now = moment().utc().unix();
         if (now - lastTs > maxLag) {
-            Grid.Clear();
             $('#errors').html(haml.compileHaml('fatal-error')({
                 last_ts: lastTs,
                 seconds_left: fatalCountdown
-            }));
+            })).show();
             Counter.Unknown();
-            Summary.Reset();
             if (!inCountdown) {
                 fatalCountdown = 60;
                 fatalReloadTimer = setTimeout(function() {
@@ -54,13 +52,8 @@ var Watchdog = (function() {
     }
 
 
-    updateMaxLag = function(interval) {
-        maxLag = Math.max(interval + 50, 300);
-    }
-
-
     updateTs = function(ts) {
-        lastTs = ts;
+        lastTs = ts.utc().unix();
     }
 
 
@@ -75,7 +68,6 @@ var Watchdog = (function() {
 
     return {
         Init: init,
-        UpdateTolerance: updateMaxLag,
         Pong: updateTs,
         GetLastUpdate: getTs,
         IsFatal: getFatal