About monitoring and observability

Knowing that a cluster is healthy takes three different signals: database metrics and query performance, the state of the Kubernetes objects underneath, and the logs the components write. The Operator has a separate mechanism for each.

What to monitor with what

Signal	Mechanism	Custom Resource field
Database metrics and query analytics	Percona Monitoring and Management (PMM)	pmm
Kubernetes objects - Pods, PVCs, StatefulSets	A Kubernetes-layer monitoring stack	none; deployed alongside
Component logs	The log collector sidecar	logcollector

Database monitoring with PMM

PMM runs as a client container beside each database Pod and reports to a PMM Server you run yourself. It is enabled with pmm.enabled and pointed at a server with pmm.serverHost. You can pass extra collection flags for mongod and mongos through pmm.mongodParams and pmm.mongosParams.

See Monitor with Percona Monitoring and Management (PMM).

Kubernetes-layer monitoring

PMM tells you about the database; it does not tell you that a PVC is full or a Pod is stuck Pending. For that, monitor the Kubernetes objects themselves - see Monitor Kubernetes.

Logs

Container logs vanish when a Pod is replaced, which is precisely when you most want them. The logcollector section enables a collector that keeps logs beyond the life of the Pod, and rotation keeps them from filling the volume.

• Persistent logging
• Log rotation

See also

• Monitor with Percona Monitoring and Management (PMM)
• Persistent logging
• Initial troubleshooting

---

Last update: September 8, 2026
Created: September 3, 2026