From: Ronen Friedman <rfriedma@redhat.com>
Date: Mon, 2 Sep 2024 18:09:13 +0000 (-0500)
Subject: osd/scrub: add 'snap-trimming' as a distinct delay reason
X-Git-Tag: v20.0.0~1113^2~2
X-Git-Url: http://git.apps.os.sepia.ceph.com/?a=commitdiff_plain;h=a84c3c457547498deec6998651158b5d5107f77c;p=ceph.git

osd/scrub: add 'snap-trimming' as a distinct delay reason

allowing the configuration of lower delay times (compared
to 'pg_state', now denoting PGs that are not active or
not clean) for PGs that failed to be scrubbed due to performing
snap-trimming.

Signed-off-by: Ronen Friedman <rfriedma@redhat.com>
---

diff --git a/src/common/options/osd.yaml.in b/src/common/options/osd.yaml.in
index 946d1bfa1e942..226e22e18ff33 100644
--- a/src/common/options/osd.yaml.in
+++ b/src/common/options/osd.yaml.in
@@ -560,6 +560,17 @@ options:
   see_also:
   - osd_scrub_retry_delay
   with_legacy: false
+- name: osd_scrub_retry_trimming
+  type: int
+  level: advanced
+  desc: Period (in seconds) before retrying to scrub a previously snap-trimming PG
+  long_desc: Minimum delay after a failed attempt to scrub a PG that was performing
+    snap trimming and not available for scrubbing.
+  default: 10
+  min: 1
+  see_also:
+  - osd_scrub_retry_delay
+  with_legacy: false
 - name: osd_scrub_disable_reservation_queuing
   type: bool
   level: advanced
diff --git a/src/osd/scrubber/pg_scrubber.cc b/src/osd/scrubber/pg_scrubber.cc
index 0e044810fcd67..b0bdc9fcb9594 100644
--- a/src/osd/scrubber/pg_scrubber.cc
+++ b/src/osd/scrubber/pg_scrubber.cc
@@ -2332,7 +2332,8 @@ Scrub::schedule_result_t PgScrubber::start_scrub_session(
     // i.e. some time before setting 'snaptrim'.
     dout(10) << __func__ << ": cannot scrub while snap-trimming" << dendl;
     requeue_penalized(
-	s_or_d, delay_both_targets_t::yes, delay_cause_t::pg_state, clock_now);
+	s_or_d, delay_both_targets_t::yes, delay_cause_t::snap_trimming,
+	clock_now);
     return schedule_result_t::target_specific_failure;
   }
 
diff --git a/src/osd/scrubber/scrub_job.cc b/src/osd/scrubber/scrub_job.cc
index dd9f8b56de038..c74648bae2578 100644
--- a/src/osd/scrubber/scrub_job.cc
+++ b/src/osd/scrubber/scrub_job.cc
@@ -313,6 +313,9 @@ SchedTarget& ScrubJob::delay_on_failure(
     case delay_cause_t::pg_state:
       delay = seconds(cct->_conf.get_val<int64_t>("osd_scrub_retry_pg_state"));
       break;
+    case delay_cause_t::snap_trimming:
+      delay = seconds(cct->_conf.get_val<int64_t>("osd_scrub_retry_trimming"));
+      break;
     case delay_cause_t::local_resources:
     default:
       // for all other possible delay causes: use the default delay
diff --git a/src/osd/scrubber_common.h b/src/osd/scrubber_common.h
index 60c1a68d2bea4..5e510a03a82ef 100644
--- a/src/osd/scrubber_common.h
+++ b/src/osd/scrubber_common.h
@@ -229,7 +229,8 @@ enum class delay_cause_t {
   none,		    ///< scrub attempt was successful
   replicas,	    ///< failed to reserve replicas
   flags,	    ///< noscrub or nodeep-scrub
-  pg_state,	    ///< e.g. snap-trimming
+  pg_state,	    ///< not active+clean
+  snap_trimming,    ///< snap-trimming is in progress
   restricted_time,  ///< time restrictions or busy CPU
   local_resources,  ///< too many scrubbing PGs
   aborted,	    ///< scrub was aborted w/ unspecified reason
@@ -252,6 +253,7 @@ struct formatter<Scrub::delay_cause_t> : ::fmt::formatter<std::string_view> {
       case replicas:            desc = "replicas"; break;
       case flags:               desc = "noscrub"; break;
       case pg_state:            desc = "pg-state"; break;
+      case snap_trimming:       desc = "snap-trim"; break;
       case restricted_time:     desc = "time/load"; break;
       case local_resources:     desc = "local-cnt"; break;
       case aborted:             desc = "aborted"; break;