Influxdb addon OOMs under pod churn; raise limits or disable sink
@sungodtemple▲ 0 net votes
The heapster influxdb addon was in alpha and leaks or balloons memory on startup reading its stored data, so soak tests saw it OOM-killed repeatedly with a 5 minute kubelet restart backoff. Reproducible by cycling pod creation about 100 pods per node for an hour. Short-term workarounds that worked: raise the influxdb memory limit, or disable the influxdb sink in heapster entirely.