What is the creating PG State

When Ceph creates a new pool or increases the PG count of an existing pool, placement groups (PGs) go through a creating state. This is a transient phase where Ceph allocates PGs to OSDs and begins distributing data. In most cases it resolves in seconds. When it lingers, you have a problem worth diagnosing.

Check PG states with:

kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph pg stat

You may see output like:

128 pgs: 16 creating, 112 active+clean; 500 GiB data

Common Causes of Stuck creating PGs

Not enough OSDs: Ceph requires a minimum number of OSDs to satisfy the pool's replication rule. If a pool has size=3 but only two OSDs are available, PGs cannot be created.

Broken CRUSH map: If the CRUSH rule references a bucket type that no longer exists or has no OSDs mapped, PG creation stalls.

OSD down during creation: If an OSD was added and immediately failed, in-flight PG creation can pause.

Diagnosing the Issue

Check the overall cluster health:

kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph health detail

List all PGs in creating state:

kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph pg dump | grep creating

Check OSD count and status:

kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph osd stat
kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph osd tree

Inspect the pool's CRUSH rule:

kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph osd pool get <pool-name> all
kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph osd crush rule dump

Resolving Stuck creating PGs

Add more OSDs: If the cluster is undersized, add OSD nodes via your Rook CephCluster spec:

storage:
  storageClassDeviceSets:
    - name: set1
      count: 3
      volumeClaimTemplates:
        - metadata:
            name: data
          spec:
            resources:
              requests:
                storage: 1Ti
            storageClassName: local-storage
            volumeMode: Block
            accessModes:
              - ReadWriteOnce

Fix CRUSH rules: If the CRUSH map is broken, recompile it:

kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- bash -c '\
  ceph osd getcrushmap -o /tmp/crushmap && \
  crushtool -d /tmp/crushmap -o /tmp/crushmap.txt && \
  # Edit /tmp/crushmap.txt as needed, then recompile
  crushtool -c /tmp/crushmap.txt -o /tmp/crushmap.new && \
  ceph osd setcrushmap -i /tmp/crushmap.new'

Restart stuck OSDs: If an OSD is down and blocking creation, identify and restart it:

kubectl -n rook-ceph get pods | grep osd
kubectl -n rook-ceph delete pod <stuck-osd-pod>

Monitoring PG Recovery

After applying fixes, watch PG states converge:

watch kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- ceph pg stat

PGs should exit creating and move to active+clean within a few minutes once all OSDs are available and CRUSH rules are satisfied.

Summary

The creating PG state is normal during cluster initialization and pool creation but should be transient. Persistent creating states usually indicate insufficient OSDs, a misconfigured CRUSH rule, or a failed OSD during the creation window. Use ceph health detail and ceph osd tree to pinpoint the cause and restore cluster health quickly.