I wrote up what it took to ship logs and metrics from two small Docker Swarm
clusters into PostHog with the OpenTelemetry Collector: an agent per node as a
global service, one gateway holding the credentials.

Every broken version of the config passed otelcol validate. The collector
tailing its own logs (1.8M lines of debug in about 70 seconds), duplicate lines
from a discovery rule that matched once per port, a memory limiter “safety”
change that made the pipeline drop data, and a folded-YAML regex that could
never match anything. Plus a CapRover trap: you can’t convert a replicated
service to global, so the agent has to be recreated from its own spec.

Full write-up with all the configs:

Curious how others collect logs on Swarm — is anyone running something simpler?