Event listeners are setup on the modal div, but that div is not removed when the modal is hidden, only its children. This means that event listeners are stacked every time you show and hide a form. To reproduce this issue:
1. open silence form, click up button above hours, it will increment from 1 to 2
2. hide the form
3. re-open silence form, click up button above hours, it will increment from 1 to 3
4. repeat and every time increment will get a new listener that will cause higher value bump
This PR moves all listeners to the silence form element, which is removed when modal is closed, so all even listeners are destroyed properly and this bug no longer gets triggered.
Same fix is applied to the quick filter modal, it's present there but the effect there isn't as visible
There's a bogus duplicated mouse event for a.silence-duration-btn, which I added to fix css issues, get rid of it and replace with proper css selectors. And with that drop the event.stopPropagation call as it shouldn't be needed anymore
This moves raven init into a dedicated file where it's configured on load, we no longer load it on /help, but there's only css there so there wasn't anything that could fail anyway. We will also now fail in the most visible way if the value passed to raven is invalid (as in invalid syntax)
Instead of showing start & end inputs show duration input element, explicit start & end are still present under tabs, where datatime picker is inlined (so it doesn't require any click to show controls). Tabs will also show desription of relative start and end, so it should be more readable what is the real duration of a silence
This commit turns label checkboxes into multiselect dropdowns that are populated from all labels of all instances of current alert (matched by alertname label). There is also a cocollapsible preview box that allows user to inspect raw silence JSON object we will send to the Alertmanager API. Fixes#69
Drop haml templates in favor of underscore. Haml templates are harder to maintain and require extra dependencies, we already have underscore.js included and it provides a fast templating engine. Rewrite all client side templates using it.
Performance with underscore is pretty much the same as with haml (with 10k alerts).
JS code have a watchdog timer that will check if last data from unsee isn't too old, if that happens it will print an error in the UI and reload the page. This is a fail safe for weird bugs and cases where unsee is running unattended. Currently timeout for this watchdog is configured dynamically based on refresh rate, but if the backend is configured to pull data from Alertmanager less often than watchdog tolerance, then watchdog will trigger false positive. This commit changes watchdog tolerance to 15 minutes (it seems reasonable to assume people won't use Alertmanager pull interval longer than 1-5 minutes) and adds a note to the readme.
