Why Review PG Scaling Recommendations?

The Ceph PG autoscaler continuously monitors pool sizes and suggests PG count adjustments. Reviewing these recommendations helps you understand whether pools are over or under-provisioned with PGs, catch configuration issues early, and validate autoscaler behavior before enabling automatic changes.

Running autoscale-status

From the Rook toolbox:

kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- \
  ceph osd pool autoscale-status

For formatted output:

kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- \
  ceph osd pool autoscale-status --format json | python3 -m json.tool

Reading the Output

A typical output table looks like:

POOL                     SIZE  TARGET SIZE  RATE  RAW CAPACITY  RATIO  TARGET RATIO  EFFECTIVE RATIO  BIAS  PG_NUM  NEW PG_NUM  AUTOSCALE  BULK
device_health_metrics      0               3.0      11.3T         0.0                          0.0   1.0       1              warn
replicapool               50G               3.0      11.3T         0.013                        0.0   1.0     128      32       warn
.mgr                     1.5M               1.0      11.3T         0.0                          0.0   4.0       1              warn

Key columns explained:

• SIZE: Actual logical data stored in the pool
• TARGET SIZE: Manually configured target size hint (from target_size_bytes)
• RATE: Replication multiplier (3x for 3-way replication, variable for erasure coding)
• RATIO: Current fraction of raw cluster capacity used by this pool
• PG_NUM: Current number of PGs
• NEW PG_NUM: Recommended PG count (blank means current count is optimal)
• AUTOSCALE: Current mode (off/warn/on)

Interpreting Recommendations

When NEW PG_NUM differs significantly from PG_NUM, the autoscaler is recommending a change. Ceph only acts (in on mode) when the ratio is off by a factor of 3x or more.

A recommendation to decrease PGs means:

• The pool has fewer objects than expected
• Current PG count wastes memory

A recommendation to increase PGs means:

• The pool has grown larger than the initial PG estimate accounted for
• PGs are unbalanced, with some OSDs handling too much data

Checking Why a Recommendation Was Made

Get detailed reasons for a specific pool:

kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- \
  ceph osd pool autoscale-status --format json | \
  python3 -c "import sys,json; \
  [print(p['pool_name'], p['pg_num'], '->', p.get('pg_num_final','same')) \
  for p in json.load(sys.stdin)]"

Acting on Recommendations Manually

If autoscaling is in warn mode, apply recommendations manually:

kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- \
  ceph osd pool set replicapool pg_num 32

kubectl -n rook-ceph exec -it deploy/rook-ceph-tools -- \
  ceph osd pool set replicapool pgp_num 32

Always reduce pg_num and pgp_num together and set pgp_num last.

Summary

ceph osd pool autoscale-status is the primary tool for understanding PG scaling recommendations. Review it regularly to validate pool sizing, identify misconfigurations, and understand what the autoscaler would change in on mode. Use the NEW PG_NUM column as guidance for manual adjustments when running in warn mode.