High CPU usage when beacon node is overloaded
barnabasbusa opened this issue · comments
Metrics exporter keeps hammering a the beacon node when its unable to produce events fast enough, thus causing very high CPU utilization when beacon node is unable to respond.
To fix, we might want to slow down with the ethereum metrics exporter when we don't get a message back from the beacon node.
Logs:
nats: slow consumer, messages dropped on connection [4] for subscription on "raw_event"
CPU up to 600% utilization.
Some exponential backoff would probably make sense. This should apply to all interactions with EL/CL apis.