SREcon24 Europe/Middle East/Africa

Lasse Hels, Maersk


How do you run a time series database exclusively on spot nodes? With great difficulty!

Grafana Mimir is the centrepiece of our observability platform at Maersk. For a long time, rollouts of Mimir's most crucial component would consistently trigger significant performance degradations in the platform. Getting to the root cause of the issue proved laborious and took us deep into the internals of Mimir.

Join us as we go through the issue postmortem and reflect on how to create consistency in a chaotic environment. The talk touches on topics such as CPU throttling, hash rings, compute utilisation analysis and metric series cardinality.


https://www.usenix.org/conference/srecon24emea/presentation/hels