osd: prioritize recovery based on pool's customized priority

Fixes: 13121
Signed-off-by: Guang Yang <yguang@yahoo-inc.com>

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index 263c88bed5f0b3ac70211826f05a2e35675db2b0..9329b5de8296806c1b70660c6dbeeaf9dc94fd23 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -2078,26 +2078,33 @@ void PG::mark_clean()
 unsigned PG::get_recovery_priority()
 {
   // a higher value -> a higher priority
-  return OSD_RECOVERY_PRIORITY_MAX;
+
+  int pool_recovery_priority = 0;
+  pool.info.opts.get(pool_opts_t::RECOVERY_PRIORITY, &pool_recovery_priority);
+
+  unsigned ret = OSD_RECOVERY_PRIORITY_BASE + pool_recovery_priority;
+  if (ret > OSD_RECOVERY_PRIORITY_MAX)
+    ret = OSD_RECOVERY_PRIORITY_MAX;
+  return ret;
 }
 
 unsigned PG::get_backfill_priority()
 {
   // a higher value -> a higher priority
 
-  // undersized: 200 + num missing replicas
+  unsigned ret = OSD_BACKFILL_PRIORITY_BASE;
   if (is_undersized()) {
+    // undersized: OSD_BACKFILL_DEGRADED_PRIORITY_BASE + num missing replicas
     assert(pool.info.size > actingset.size());
-    return 200 + (pool.info.size - actingset.size());
-  }
+    ret = OSD_BACKFILL_DEGRADED_PRIORITY_BASE + (pool.info.size - actingset.size());
 
-  // degraded: baseline degraded
-  if (is_degraded()) {
-    return 200;
+  } else if (is_degraded()) {
+    // degraded: baseline degraded
+    ret = OSD_BACKFILL_DEGRADED_PRIORITY_BASE;
   }
+  assert (ret < OSD_RECOVERY_PRIORITY_MAX);
 
-  // baseline
-  return 1;
+  return ret;
 }
 
 void PG::finish_recovery(list<Context*>& tfin)

diff --git a/src/osd/osd_types.h b/src/osd/osd_types.h
index 98ae300eb791120b66d895d3653911bf4d4a212d..832321e3da6d6dfefe317ab7546c81cc93a53584 100644 (file)
--- a/src/osd/osd_types.h
+++ b/src/osd/osd_types.h
@@ -63,6 +63,14 @@
 /// max recovery priority for MBackfillReserve
 #define OSD_RECOVERY_PRIORITY_MAX 255u
 
+/// base recovery priority for MBackfillReserve
+#define OSD_RECOVERY_PRIORITY_BASE 230u
+
+/// base backfill priority for MBackfillReserve (degraded PG)
+#define OSD_BACKFILL_DEGRADED_PRIORITY_BASE 200u
+
+/// base backfill priority for MBackfillReserve
+#define OSD_BACKFILL_PRIORITY_BASE 1u
 
 typedef hobject_t collection_list_handle_t;