The repair PG state indicates that Ceph is actively fixing object inconsistencies found during a scrub. It is a short-lived transitional state that follows the detection of an inconsistent PG.

What repair Means

When ceph pg repair <pg-id> is called, or when Ceph automatically repairs during a scrub (if osd_scrub_auto_repair is enabled), the PG enters the repair state. During repair:

1. The primary OSD identifies the authoritative copy of each inconsistent object
2. It overwrites the corrupt or incorrect copies on replica OSDs
3. After all objects are repaired, the PG exits the repair state

Triggering Repair

Repair is triggered manually or automatically:

# Manual repair
ceph pg repair <pg-id>

# Repair all inconsistent PGs in a pool
for pg in $(ceph pg dump | grep inconsistent | awk '{print $1}'); do
  ceph pg repair $pg
done

Enable automatic repair:

ceph config set osd osd_scrub_auto_repair true
ceph config set osd osd_scrub_auto_repair_num_errors 5  # max errors to auto-repair

Monitoring Repair Progress

ceph pg stat | grep repair

# Watch the repair state clear
watch 'ceph health detail | grep -E "inconsistent|repair"'

For a specific PG:

ceph pg <pg-id> query | jq '.state'

The repair state should clear within minutes for small PGs.

How Ceph Chooses the Authoritative Copy

Ceph uses the primary OSD's copy as authoritative in most cases. The decision is based on:

1. The acting primary's object version
2. Which replica has the highest version number
3. If versions are equal, the primary wins

You can inspect the decision:

ceph pg <pg-id> query | jq '.peer_info'

When Repair Succeeds

After successful repair:

ceph health detail
# No more inconsistent PG messages

ceph pg dump | grep <pg-id>
# State should show active+clean

Confirm the fix with another scrub:

ceph pg deep-scrub <pg-id>

When Repair Fails

If repair does not clear the inconsistency:

ceph health detail | grep "inconsistent"

Possible causes:

• All copies are equally corrupt (no authoritative version)
• The OSD containing the good copy is offline

In this case:

# Identify which objects are still inconsistent
rados list-inconsistent-obj <pg-id>

# If all copies are bad, the object may be unrecoverable
# Remove the corrupt object as a last resort
rados -p <pool-name> rm <object-name>

Repair During Scrub

When osd_scrub_auto_repair is enabled, repair happens inline during the scrub cycle. The PG will show scrubbing+repair or scrubbing+deep+repair:

ceph pg dump | grep "scrubbing"

Summary

The repair PG state is a positive indicator: Ceph has found inconsistencies and is actively correcting them. Repair is triggered manually via ceph pg repair or automatically via osd_scrub_auto_repair. Monitor its completion with ceph health detail and follow up with a deep scrub to verify the repair was successful. Persistent repair failures indicate hardware issues requiring physical inspection.