I have been having this (or similar issue) and it seems to be that it runs out of memory to ingest the metrics coming in to prometheus.
And usually when it dies it really dies and I have to restart the docker.
I have been bumping up the memory with this command:
METRICS_MEMORY=2097152
In the docker run command line. It has some fairly small default - that was working fine until I added one too many servers or has a peak in events.
You might take a look at the Prometheus graph in grafana.
My running server looks like this when I run docker stats pmm-server
CONTAINER CPU % MEM USAGE / LIMIT MEM % NET I/O BLOCK I/O
pmm-server 71.24% 6.77 GB / 7.934 GB 85.33% 6.932 GB / 379.2 MB 3.766 GB / 18.15 GB