osd: Change scrub cost in case of mClock scheduler

With osd_op_queue as WPQ, high costs were assigned to scrub in order to throttle it effectively.
In the case of mClock scheduler, mClock parameters are used to do the throttling and the cost should represent a realistic value.

Fixes: https://tracker.ceph.com/issues/61313
Signed-off-by: Aishwarya Mathuria <[email protected]>

Author: amathuria

src/common/options/global.yaml.in

@@ -3645,6 +3645,12 @@ options:
   desc: Cost for scrub operations in work queue
   default: 50_M
   with_legacy: true
+- name: osd_scrub_event_cost
+  type: size
+  level: advanced
+  desc: Cost for each scrub operation, used when osd_op_queue=mclock_scheduler
+  default: 4_K
+  with_legacy: true
 # set requested scrub priority higher than scrub priority to make the
 # requested scrubs jump the queue of scheduled scrubs
 - name: osd_requested_scrub_priority

src/osd/OSD.cc

@@ -1756,9 +1756,8 @@ void OSDService::queue_scrub_event_msg(PG* pg,
   auto msg = new MSG_TYPE(pg->get_pgid(), epoch, act_token);
   dout(15) << "queue a scrub event (" << *msg << ") for " << *pg
            << ". Epoch: " << epoch << " token: " << act_token << dendl;
-
   enqueue_back(OpSchedulerItem(
-    unique_ptr<OpSchedulerItem::OpQueueable>(msg), cct->_conf->osd_scrub_cost,
+    unique_ptr<OpSchedulerItem::OpQueueable>(msg), get_scrub_cost(),
     pg->scrub_requeue_priority(with_priority, qu_priority), ceph_clock_now(), 0, epoch));
 }
 
@@ -1769,12 +1768,22 @@ void OSDService::queue_scrub_event_msg(PG* pg,
   const auto epoch = pg->get_osdmap_epoch();
   auto msg = new MSG_TYPE(pg->get_pgid(), epoch);
   dout(15) << "queue a scrub event (" << *msg << ") for " << *pg << ". Epoch: " << epoch << dendl;
-
   enqueue_back(OpSchedulerItem(
-    unique_ptr<OpSchedulerItem::OpQueueable>(msg), cct->_conf->osd_scrub_cost,
+    unique_ptr<OpSchedulerItem::OpQueueable>(msg), get_scrub_cost(),
     pg->scrub_requeue_priority(with_priority), ceph_clock_now(), 0, epoch));
 }
 
+int64_t OSDService::get_scrub_cost()
+{
+
+  int64_t cost_for_queue = cct->_conf->osd_scrub_cost;
+  if (cct->_conf->osd_op_queue == "mclock_scheduler") {
+    cost_for_queue = cct->_conf->osd_scrub_event_cost *
+                     cct->_conf->osd_shallow_scrub_chunk_max;
+  }
+  return cost_for_queue;
+}
+
 void OSDService::queue_for_scrub(PG* pg, Scrub::scrub_prio_t with_priority)
 {
   queue_scrub_event_msg<PGScrub>(pg, with_priority);

src/osd/OSD.h

@@ -605,6 +605,7 @@ class OSDService : public Scrub::ScrubSchedListener {
   /// provided by the executing scrub (i.e. taken from PgScrubber::m_flags)
   template <class MSG_TYPE>
   void queue_scrub_event_msg(PG* pg, Scrub::scrub_prio_t with_priority);
+  int64_t get_scrub_cost();
 
   utime_t defer_recovery_until;
   uint64_t recovery_ops_active;