From: Ronen Friedman <rfriedma@redhat.com>
Date: Mon, 1 Jul 2024 14:06:54 +0000 (-0500)
Subject: osd/scrub: improve update_scrub_job()
X-Git-Tag: v20.0.0~1493^2~5
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=75b8ecbc2a2a30dea78f8e6fd009e9d18d305a81;p=ceph.git

osd/scrub: improve update_scrub_job()

...and on_scrub_schedule_input_change()
by accepting a parameter to control whether the scheduled time for
periodic scrubs that are already due should be updated.

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

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index a8b9d41e1dab..6392a184bd32 100644
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -1361,14 +1361,18 @@ double PG::next_deepscrub_interval() const
   return info.history.last_deep_scrub_stamp + deep_scrub_interval;
 }
 
-void PG::on_scrub_schedule_input_change()
+void PG::on_scrub_schedule_input_change(Scrub::delay_ready_t delay_ready)
 {
   if (is_active() && is_primary()) {
-    dout(20) << __func__ << ": active/primary" << dendl;
+    dout(10) << fmt::format(
+		    "{}: active/primary. delay_ready={:c}", __func__,
+		    (delay_ready == Scrub::delay_ready_t::delay_ready) ? 't'
+								       : 'f')
+	     << dendl;
     ceph_assert(m_scrubber);
-    m_scrubber->update_scrub_job(m_planned_scrub);
+    m_scrubber->update_scrub_job(delay_ready);
   } else {
-    dout(20) << __func__ << ": inactive or non-primary" << dendl;
+    dout(10) << fmt::format("{}: inactive or non-primary", __func__) << dendl;
   }
 }
 
@@ -2261,7 +2265,7 @@ void PG::handle_activate_map(PeeringCtx &rctx, epoch_t range_starts_at)
   // on_scrub_schedule_input_change() as pool.info contains scrub scheduling
   // parameters.
   if (pool.info.last_change >= range_starts_at) {
-    on_scrub_schedule_input_change();
+    on_scrub_schedule_input_change(Scrub::delay_ready_t::delay_ready);
   }
 }
 
diff --git a/src/osd/PG.h b/src/osd/PG.h
index 444f59254866..b18e3e163896 100644
--- a/src/osd/PG.h
+++ b/src/osd/PG.h
@@ -269,7 +269,7 @@ public:
 	set_last_scrub_stamp(t, history, stats);
 	return true;
       });
-    on_scrub_schedule_input_change();
+    on_scrub_schedule_input_change(Scrub::delay_ready_t::delay_ready);
   }
 
   static void set_last_deep_scrub_stamp(
@@ -285,7 +285,7 @@ public:
 	set_last_scrub_stamp(t, history, stats);
 	return true;
       });
-    on_scrub_schedule_input_change();
+    on_scrub_schedule_input_change(Scrub::delay_ready_t::delay_ready);
   }
 
   static void add_objects_scrubbed_count(
@@ -531,7 +531,7 @@ public:
    * - pg stat scrub timestamps
    * - etc
    */
-  void on_scrub_schedule_input_change();
+  void on_scrub_schedule_input_change(Scrub::delay_ready_t delay_ready);
 
   void scrub_requested(scrub_level_t scrub_level, scrub_type_t scrub_type) override;
 
diff --git a/src/osd/scrubber/osd_scrub.cc b/src/osd/scrubber/osd_scrub.cc
index a11777a040a3..76e590a0930d 100644
--- a/src/osd/scrubber/osd_scrub.cc
+++ b/src/osd/scrubber/osd_scrub.cc
@@ -236,7 +236,8 @@ void OsdScrub::on_config_change()
 		    "updating scrub schedule on {}",
 		    (locked_pg->pg())->get_pgid())
 	     << dendl;
-    locked_pg->pg()->on_scrub_schedule_input_change();
+    locked_pg->pg()->on_scrub_schedule_input_change(
+	Scrub::delay_ready_t::no_delay);
   }
 }
 
diff --git a/src/osd/scrubber/pg_scrubber.cc b/src/osd/scrubber/pg_scrubber.cc
index e0bc082f7af6..8984d00f45da 100644
--- a/src/osd/scrubber/pg_scrubber.cc
+++ b/src/osd/scrubber/pg_scrubber.cc
@@ -579,12 +579,12 @@ void PgScrubber::on_primary_active_clean()
  *   guarantees that the PG is locked and the interval is still the same.
  * - in the 2nd case - we know the PG state and we know we are only called
  *   for a Primary.
-*/
-void PgScrubber::update_scrub_job(const requested_scrub_t& request_flags)
+ */
+void PgScrubber::update_scrub_job(Scrub::delay_ready_t delay_ready)
 {
   if (!is_primary() || !m_scrub_job) {
     dout(10) << fmt::format(
-		    "{}: pg[{}]: not Primary or no scrub-job", __func__,
+		    "{}: PG[{}]: not Primary or no scrub-job", __func__,
 		    m_pg_id)
 	     << dendl;
     return;
@@ -599,7 +599,7 @@ void PgScrubber::update_scrub_job(const requested_scrub_t& request_flags)
   }
 
   dout(15) << fmt::format(
-		  "{}: flags:<{}> job on entry:{}", __func__, request_flags,
+		  "{}: flags:<{}> job on entry:{}", __func__, m_planned_scrub,
 		  *m_scrub_job)
 	   << dendl;
   if (m_scrub_job->target_queued) {
@@ -610,7 +610,6 @@ void PgScrubber::update_scrub_job(const requested_scrub_t& request_flags)
 	     << dendl;
   }
 
-
   ceph_assert(m_pg->is_locked());
   const auto applicable_conf = populate_config_params();
   const auto scrub_clock_now = ceph_clock_now();
@@ -620,10 +619,10 @@ void PgScrubber::update_scrub_job(const requested_scrub_t& request_flags)
   m_scrub_job->target_queued = true;
   m_pg->publish_stats_to_osd();
 
-  dout(15) << fmt::format(
-                  "{}: flags:<{}> job on exit:{}", __func__, request_flags,
-                  *m_scrub_job)
-           << dendl;
+  dout(10) << fmt::format(
+		  "{}: flags:<{}> job on exit:{}", __func__, m_planned_scrub,
+		  *m_scrub_job)
+	   << dendl;
 }
 
 scrub_level_t PgScrubber::scrub_requested(
@@ -651,7 +650,7 @@ scrub_level_t PgScrubber::scrub_requested(
   req_flags.req_scrub = true;
   dout(20) << fmt::format("{}: planned scrub:{}", __func__, req_flags) << dendl;
 
-  update_scrub_job(req_flags);
+  update_scrub_job(delay_ready_t::no_delay);
   return deep_requested ? scrub_level_t::deep : scrub_level_t::shallow;
 }
 
@@ -661,7 +660,7 @@ void PgScrubber::request_rescrubbing(requested_scrub_t& request_flags)
   dout(10) << __func__ << " flags: " << request_flags << dendl;
 
   request_flags.need_auto = true;
-  update_scrub_job(request_flags);
+  update_scrub_job(delay_ready_t::no_delay);
 }
 
 bool PgScrubber::reserve_local()
@@ -726,7 +725,7 @@ Scrub::sched_conf_t PgScrubber::populate_config_params() const
   configs.deep_randomize_ratio = conf->osd_deep_scrub_randomize_ratio;
   configs.mandatory_on_invalid = conf->osd_scrub_invalid_stats;
 
-  dout(15) << fmt::format("updated config:{}", configs) << dendl;
+  dout(15) << fmt::format("{}: updated config:{}", __func__, configs) << dendl;
   return configs;
 }
 
@@ -2035,7 +2034,9 @@ void PgScrubber::scrub_finish()
     int tr = m_osds->store->queue_transaction(m_pg->ch, std::move(t), nullptr);
     ceph_assert(tr == 0);
   }
-  update_scrub_job(m_planned_scrub);
+
+  // determine the next scrub time
+  update_scrub_job(delay_ready_t::delay_ready);
 
   if (has_error) {
     m_pg->queue_peering_event(PGPeeringEventRef(
diff --git a/src/osd/scrubber/pg_scrubber.h b/src/osd/scrubber/pg_scrubber.h
index 2255ad8ce0ec..0106edae54ae 100644
--- a/src/osd/scrubber/pg_scrubber.h
+++ b/src/osd/scrubber/pg_scrubber.h
@@ -256,7 +256,7 @@ class PgScrubber : public ScrubPgIF,
 
   // managing scrub op registration
 
-  void update_scrub_job(const requested_scrub_t& request_flags) final;
+  void update_scrub_job(Scrub::delay_ready_t delay_ready) final;
 
   void rm_from_osd_scrubbing() final;
 
diff --git a/src/osd/scrubber_common.h b/src/osd/scrubber_common.h
index f7ac9a867fe7..832811fbad76 100644
--- a/src/osd/scrubber_common.h
+++ b/src/osd/scrubber_common.h
@@ -107,6 +107,10 @@ enum class schedule_result_t {
   target_specific_failure,  // failed to scrub this specific target
   osd_wide_failure	    // failed to scrub any target
 };
+
+/// rescheduling param: should we delay jobs already ready to execute?
+enum class delay_ready_t : bool { delay_ready = true, no_delay = false };
+
 }  // namespace Scrub
 
 namespace fmt {
@@ -547,7 +551,7 @@ struct ScrubPgIF {
    * This function assumes that the queue registration status is up-to-date,
    * i.e. the OSD "knows our name" if-f we are the Primary.
    */
-  virtual void update_scrub_job(const requested_scrub_t& request_flags) = 0;
+  virtual void update_scrub_job(Scrub::delay_ready_t delay_ready) = 0;
 
   /**
    * route incoming replica-reservations requests/responses to the