osd/PG: pass scrub priority to replica

If we are scrubbing with high priority on the primary, pass that along
to the replica so that it can schedule its scrub work accordingly.

Signed-off-by: Sage Weil <sage@redhat.com>
(cherry picked from commit d9fd07696058cf79a62c327ecf08a5f8fb5b6a28)

- fixed encode vs ::encode conflict

diff --git a/src/messages/MOSDRepScrub.h b/src/messages/MOSDRepScrub.h
index 90b0746d6d64c75afd64f4be700a638fc17a6f13..f4f26ee1f2d45860ebd97a9d5211e145add12f90 100644 (file)
--- a/src/messages/MOSDRepScrub.h
+++ b/src/messages/MOSDRepScrub.h
@@ -24,7 +24,7 @@
 
 struct MOSDRepScrub : public MOSDFastDispatchOp {
 
-  static const int HEAD_VERSION = 8;
+  static const int HEAD_VERSION = 9;
   static const int COMPAT_VERSION = 6;
 
   spg_t pgid;             // PG to scrub
@@ -36,6 +36,8 @@ struct MOSDRepScrub : public MOSDFastDispatchOp {
   hobject_t end;         // upper bound of scrub, exclusive
   bool deep;             // true if scrub should be deep
   bool allow_preemption = false;
+  int32_t priority = 0;
+  bool high_priority = false;
 
   epoch_t get_map_epoch() const override {
     return map_epoch;
@@ -54,7 +56,7 @@ struct MOSDRepScrub : public MOSDFastDispatchOp {
 
   MOSDRepScrub(spg_t pgid, eversion_t scrub_to, epoch_t map_epoch, epoch_t min_epoch,
                hobject_t start, hobject_t end, bool deep,
-              bool preemption)
+              bool preemption, int prio, bool highprio)
     : MOSDFastDispatchOp(MSG_OSD_REP_SCRUB, HEAD_VERSION, COMPAT_VERSION),
       pgid(pgid),
       scrub_to(scrub_to),
@@ -64,7 +66,9 @@ struct MOSDRepScrub : public MOSDFastDispatchOp {
       start(start),
       end(end),
       deep(deep),
-      allow_preemption(preemption) { }
+      allow_preemption(preemption),
+      priority(prio),
+      high_priority(highprio) { }
 
 
 private:
@@ -82,6 +86,8 @@ public:
         << ",deep:" << deep
         << ",version:" << header.version
        << ",allow_preemption:" << (int)allow_preemption
+       << ",priority=" << priority
+       << (high_priority ? " (high)":"")
        << ")";
   }
 
@@ -98,6 +104,8 @@ public:
     ::encode((uint32_t)-1, payload); // seed
     ::encode(min_epoch, payload);
     ::encode(allow_preemption, payload);
+    ::encode(priority, payload);
+    ::encode(high_priority, payload);
   }
   void decode_payload() override {
     bufferlist::iterator p = payload.begin();
@@ -122,6 +130,10 @@ public:
     if (header.version >= 8) {
       ::decode(allow_preemption, p);
     }
+    if (header.version >= 9) {
+      ::decode(priority, p);
+      ::decode(high_priority, p);
+    }
   }
 };
 

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index 382eb2bf959dd11fce583ca806335a6e3728ae97..0641e59913e829c19c6ef0c34ad3e445da48d177 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -3858,7 +3858,9 @@ void PG::_request_scrub_map(
     get_osdmap()->get_epoch(),
     get_last_peering_reset(),
     start, end, deep,
-    allow_preemption);
+    allow_preemption,
+    scrubber.priority,
+    ops_blocked_by_scrub());
   // default priority, we want the rep scrub processed prior to any recovery
   // or client io messages (we are holding a lock!)
   osd->send_message_osd_cluster(
@@ -4394,12 +4396,17 @@ void PG::replica_scrub(
   scrubber.end = msg->end;
   scrubber.deep = msg->deep;
   scrubber.epoch_start = info.history.same_interval_since;
+  if (msg->priority) {
+    scrubber.priority = msg->priority;
+  } else {
+    scrubber.priority = get_scrub_priority();
+  }
 
   scrub_can_preempt = msg->allow_preemption;
   scrub_preempted = false;
   scrubber.replica_scrubmap_pos.reset();
 
-  requeue_scrub(false);
+  requeue_scrub(msg->high_priority);
 }
 
 /* Scrub: