Some useful Grafana plots

Below are some links to various plots from grafana, an explanation of what they show and why they are useful to check.

Many of the plots on grafana have a drop-down menu called “cluster” with three options referring to the different clusters operates on:

  • “prometheus” refers to the GKE cluster (US)

  • “OVH prometheus” refers to the OVH cluster (EU)

  • “default” is the sum across the clusters

  • Launch/Build Success Rate - This chart is the main indicator that is healthy. If there is a problem, a dropping success rate indicates that it’s impacting users. The caveat is that it’s an indicator of current status: if builds or launches are failing due to timeouts that it can take a long time to show up as the failure is only reported once it’s completed failing. Including retries, this can take several minutes. The Launch Success Rate graph shows similar statistics but also includes the number of launch attempts. This can tell us if no-one tried to launch a repo, or if a lot of people tried but failed. Ocassionally, large spikes in launch attempts happen which can be interesting to investigate.

  • Number of Pods per Node - This graph shows which nodes are serving the users. This can indicate when auto-scaling may need a helping hand, for example, if one node has had a low load for a long time, it may need to be manually cordoned and drained because a pod is stuck on it.

  • Number of User Pods over time - This graph shows how many user pods have been running on If this seems to be larger than usual, it could be indicative of something interesting, such as posting a blog post or a link on HackerNews that generates a lot of traffic.

  • Popular repositories - This graph shows the most popular repos by number of launches in the last hour and, for example, we can see classes that use