From: Ronen Friedman <rfriedma@redhat.com>
Date: Wed, 8 Oct 2025 13:26:41 +0000 (-0500)
Subject: osd/scrub: fix blocked scrub accounting
X-Git-Tag: v21.0.0~233^2~1^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=88866336759caa769cbc5220a41f4e1c8c28da3f;p=ceph.git

osd/scrub: fix blocked scrub accounting

The labeled perf counter <label>_chunk_busy is incremented whenever
the selected scrub chunk range is not available, and the scrubber
must wait for ongoing I/O to complete. This is as it should be.

<label>_locked_object should not be incremented in parallel,
serving no purpose. Instead - it should count the times the
'wait for chunk to be free' takes longer than the configured
grace period.

Fixes: https://tracker.ceph.com/issues/73415
Signed-off-by: Ronen Friedman <rfriedma@redhat.com>
---

diff --git a/src/osd/scrubber/scrub_machine.cc b/src/osd/scrubber/scrub_machine.cc
index 2d21c8d25c84..76be5a9f5a02 100644
--- a/src/osd/scrubber/scrub_machine.cc
+++ b/src/osd/scrubber/scrub_machine.cc
@@ -383,7 +383,6 @@ RangeBlocked::RangeBlocked(my_context ctx)
     m_timeout_token = machine.schedule_timer_event_after<RangeBlockedAlarm>(
       grace);
   }
-  context<Session>().m_perf_set->inc(scrbcnt_blocked);
 }
 
 sc::result RangeBlocked::react(const RangeBlockedAlarm&)
@@ -402,6 +401,7 @@ sc::result RangeBlocked::react(const RangeBlockedAlarm&)
     << ")";
 
   scrbr->set_scrub_blocked(utime_t{now_c, 0});
+  context<Session>().m_perf_set->inc(scrbcnt_blocked);
   return discard_event();
 }