Vulcand Dashboard about vulcand HOT 7 OPEN

Some suggestions that have come up in discussion:

• Focus on real-time data that can be used to quickly determine the status of deployments and identifying problems.
• Build higher-level statistics from the requests/second and failures/second data provided by the API. For example, we can determine error rate for an endpoint or roll up endpoint statistics to determine the overall health of a host.
• Provide two views: one view for hosts and locations, and another view for upstream and endpoints.
• Production Vulcand deployments will contain hundreds of locations, upstreams, and endpoints. Provide searching and filtering capabilities to reduce data displayed on a single dashboard to a comprehensible level. Allow custom dashboards to be persisted.
• When compared to nginx or HAProxy, Vulcand is distinguished by its dynamic nature - all aspects of the configuration may change as services announce themselves to etcd. The ability to visualize changes in topology will be helpful when debugging deployments, pushing canary builds, and performing rolling restarts.

from vulcand.

Check out vulcanctl status -w 1 for some ideas:

• It displays top locations sorted by: error rates first, requests second

So for the top panel you get a top-like list of locations that are highlighted if there's error rate what is the first thing I do when deploying something new - to watch for errors and anomalies.

from vulcand.

@klizhentas Put this together last night.

Here's the basic idea:

• Choose between host-centric view or upstream-centric view using the navigation. This also needs a way to search/filter by host, location, upstream, or endpoint.
• Overview section show real-time info on requests/second and errors/second.
• Table shows realtime stats broken down by host and location. We could highlight based on error rate. You could sort by any header, and error rate would be the default sort.

from vulcand.

@bradgignac that's a great start. Merge it into master once ready, we'll see how useful it is on production.

from vulcand.

@bradgignac I've just realized that you don't have enough datapoints to draw the real chart. Should I add API endpoints returning time series for it? What format would you prefer?

from vulcand.

Right now, I show the last X seconds of data, and it takes X seconds to populate the entire graph. Good enough for now.

from vulcand.

ok, even better for a start!

from vulcand.