update sources to v12.2.5

[ceph.git] / ceph / src / osd / OSD.h

diff --git a/ceph/src/osd/OSD.h b/ceph/src/osd/OSD.h
index bfa511e8b077df2d9864dae429eba84946a5b954..ea611cbae1cb7010d09dc8c431bf6e94c07dcbc0 100644 (file)
--- a/ceph/src/osd/OSD.h
+++ b/ceph/src/osd/OSD.h
@@ -127,6 +127,7 @@ enum {
   l_osd_pg_primary,
   l_osd_pg_replica,
   l_osd_pg_stray,
+  l_osd_pg_removing,
   l_osd_hb_to,
   l_osd_map,
   l_osd_mape,
@@ -837,11 +838,22 @@ public:
   // -- pg_temp --
 private:
   Mutex pg_temp_lock;
-  map<pg_t, vector<int> > pg_temp_wanted;
-  map<pg_t, vector<int> > pg_temp_pending;
+  struct pg_temp_t {
+    pg_temp_t()
+    {}
+    pg_temp_t(vector<int> v, bool f)
+      : acting{v}, forced{f}
+    {}
+    vector<int> acting;
+    bool forced = false;
+  };
+  map<pg_t, pg_temp_t> pg_temp_wanted;
+  map<pg_t, pg_temp_t> pg_temp_pending;
   void _sent_pg_temp();
+  friend std::ostream& operator<<(std::ostream&, const pg_temp_t&);
 public:
-  void queue_want_pg_temp(pg_t pgid, vector<int>& want);
+  void queue_want_pg_temp(pg_t pgid, const vector<int>& want,
+                         bool forced = false);
   void remove_want_pg_temp(pg_t pgid);
   void requeue_pg_temp();
   void send_pg_temp();
@@ -947,9 +959,10 @@ public:
     }
   }
   // delayed pg activation
-  void queue_for_recovery(PG *pg, bool front = false) {
+  void queue_for_recovery(PG *pg) {
     Mutex::Locker l(recovery_lock);
-    if (front) {
+
+    if (pg->get_state() & (PG_STATE_FORCED_RECOVERY | PG_STATE_FORCED_BACKFILL)) {
       awaiting_throttle.push_front(make_pair(pg->get_osdmap()->get_epoch(), pg));
     } else {
       awaiting_throttle.push_back(make_pair(pg->get_osdmap()->get_epoch(), pg));
@@ -961,6 +974,7 @@ public:
     _queue_for_recovery(make_pair(queued, pg), reserved_pushes);
   }
 
+  void adjust_pg_priorities(const vector<PGRef>& pgs, int newflags);
 
   // osd map cache (past osd maps)
   Mutex map_cache_lock;
@@ -1045,7 +1059,8 @@ public:
 
   void update_osd_stat(vector<int>& hb_peers);
   osd_stat_t set_osd_stat(const struct store_statfs_t &stbuf,
-                          vector<int>& hb_peers);
+                          vector<int>& hb_peers,
+                         int num_pgs);
   osd_stat_t get_osd_stat() {
     Mutex::Locker l(stat_lock);
     ++seq;
@@ -1141,6 +1156,8 @@ public:
     return ret;
   }
 
+  void request_osdmap_update(epoch_t e);
+
   // -- stopping --
   Mutex is_stopping_lock;
   Cond is_stopping_cond;
@@ -1220,6 +1237,7 @@ protected:
   std::string dev_path, journal_path;
 
   bool store_is_rotational = true;
+  bool journal_is_rotational = true;
 
   ZTracer::Endpoint trace_endpoint;
   void create_logger();
@@ -1451,6 +1469,9 @@ private:
   void osdmap_subscribe(version_t epoch, bool force_request);
   /** @} monc helpers */
 
+  Mutex osdmap_subscribe_lock;
+  epoch_t latest_subscribed_epoch{0};
+
   // -- heartbeat --
   /// information about a heartbeat peer
   struct HeartbeatInfo {
@@ -1937,6 +1958,11 @@ protected:
   RWLock pg_map_lock; // this lock orders *above* individual PG _locks
   ceph::unordered_map<spg_t, PG*> pg_map; // protected by pg_map lock
 
+  std::mutex pending_creates_lock;
+  using create_from_osd_t = std::pair<pg_t, bool /* is primary*/>;
+  std::set<create_from_osd_t> pending_creates_from_osd;
+  unsigned pending_creates_from_mon = 0;
+
   map<spg_t, list<PG::CephPeeringEvtRef> > peering_wait_for_split;
   PGRecoveryStats pg_recovery_stats;
 
@@ -1948,6 +1974,11 @@ protected:
 public:
   PG   *lookup_lock_pg(spg_t pgid);
 
+  int get_num_pgs() {
+    RWLock::RLocker l(pg_map_lock);
+    return pg_map.size();
+  }
+
 protected:
   PG   *_open_lock_pg(OSDMapRef createmap,
                      spg_t pg, bool no_lockdep_check=false);
@@ -1981,7 +2012,9 @@ protected:
     const PastIntervals& pi,
     epoch_t epoch,
     PG::CephPeeringEvtRef evt);
-  
+  bool maybe_wait_for_max_pg(spg_t pgid, bool is_mon_create);
+  void resume_creating_pg();
+
   void load_pgs();
   void build_past_intervals_parallel();
 
@@ -2169,6 +2202,8 @@ protected:
   void handle_pg_backfill_reserve(OpRequestRef op);
   void handle_pg_recovery_reserve(OpRequestRef op);
 
+  void handle_force_recovery(Message *m);
+
   void handle_pg_remove(OpRequestRef op);
   void _remove_pg(PG *pg);
 
@@ -2274,8 +2309,18 @@ protected:
     void _clear() override {
       remove_queue.clear();
     }
+    int get_remove_queue_len() {
+      lock();
+      int r = remove_queue.size();
+      unlock();
+      return r;
+    }
   } remove_wq;
 
+  // -- status reporting --
+  MPGStats *collect_pg_stats();
+  std::vector<OSDHealthMetric> get_health_metrics();
+
 private:
   bool ms_can_fast_dispatch_any() const override { return true; }
   bool ms_can_fast_dispatch(const Message *m) const override {
@@ -2301,6 +2346,8 @@ private:
     case MSG_OSD_REP_SCRUBMAP:
     case MSG_OSD_PG_UPDATE_LOG_MISSING:
     case MSG_OSD_PG_UPDATE_LOG_MISSING_REPLY:
+    case MSG_OSD_PG_RECOVERY_DELETE:
+    case MSG_OSD_PG_RECOVERY_DELETE_REPLY:
       return true;
     default:
       return false;
@@ -2329,14 +2376,15 @@ private:
       srand(time(NULL));
       unsigned which = rand() % (sizeof(index_lookup) / sizeof(index_lookup[0]));
       return index_lookup[which];
-    } else if (cct->_conf->osd_op_queue == "wpq") {
-      return io_queue::weightedpriority;
+    } else if (cct->_conf->osd_op_queue == "prioritized") {
+      return io_queue::prioritized;
     } else if (cct->_conf->osd_op_queue == "mclock_opclass") {
       return io_queue::mclock_opclass;
     } else if (cct->_conf->osd_op_queue == "mclock_client") {
       return io_queue::mclock_client;
     } else {
-      return io_queue::prioritized;
+      // default / catch-all is 'wpq'
+      return io_queue::weightedpriority;
     }
   }
 
@@ -2344,10 +2392,11 @@ private:
     if (cct->_conf->osd_op_queue_cut_off == "debug_random") {
       srand(time(NULL));
       return (rand() % 2 < 1) ? CEPH_MSG_PRIO_HIGH : CEPH_MSG_PRIO_LOW;
-    } else if (cct->_conf->osd_op_queue_cut_off == "low") {
-      return CEPH_MSG_PRIO_LOW;
-    } else {
+    } else if (cct->_conf->osd_op_queue_cut_off == "high") {
       return CEPH_MSG_PRIO_HIGH;
+    } else {
+      // default / catch-all is 'low'
+      return CEPH_MSG_PRIO_LOW;
     }
   }
 
@@ -2387,7 +2436,8 @@ private:
   int update_crush_device_class();
   int update_crush_location();
 
-  static int write_meta(ObjectStore *store,
+  static int write_meta(CephContext *cct,
+                       ObjectStore *store,
                        uuid_d& cluster_fsid, uuid_d& osd_fsid, int whoami);
 
   void handle_pg_scrub(struct MOSDScrub *m, PG* pg);
@@ -2399,6 +2449,8 @@ private:
   int get_num_op_shards();
   int get_num_op_threads();
 
+  float get_osd_recovery_sleep();
+
 public:
   static int peek_meta(ObjectStore *store, string& magic,
                       uuid_d& cluster_fsid, uuid_d& osd_fsid, int& whoami);