osd/pg: use next when calling collection_list for pg removal

Signed-off-by: Igor Fedotov <ifedotov@suse.com>
(cherry picked from commit 7f047005fc72e1f37a45cde2d742bb2eb1e62881)

diff --git a/src/crimson/osd/pg.cc b/src/crimson/osd/pg.cc
index 4a60f385cbb27922b5f453f1683cdefdd42cb658..c65ad15ae433f3ebb00ad16b78769f47ffb3f74f 100644 (file)
--- a/src/crimson/osd/pg.cc
+++ b/src/crimson/osd/pg.cc
@@ -297,7 +297,8 @@ void PG::prepare_write(pg_info_t &info,
   }
 }
 
-void PG::do_delete_work(ceph::os::Transaction &t)
+ghobject_t PG::do_delete_work(ceph::os::Transaction &t,
+  ghobject_t _next)
 {
   // TODO
   shard_services.dec_pg_num();

diff --git a/src/crimson/osd/pg.h b/src/crimson/osd/pg.h
index 7ec64d319c7c5c9a13ca2513120b3651fda11826..72f43c0eaaa082ba72ab5236a0fd9aa0ee3f1875 100644 (file)
--- a/src/crimson/osd/pg.h
+++ b/src/crimson/osd/pg.h
@@ -273,7 +273,8 @@ public:
   void on_removal(ceph::os::Transaction &t) final {
     // TODO
   }
-  void do_delete_work(ceph::os::Transaction &t) final;
+  ghobject_t do_delete_work(ceph::os::Transaction &t,
+    ghobject_t _next) final;
 
   // merge/split not ready
   void clear_ready_to_merge() final {}

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index e5a921499438693557fc41fde73d848a7028fd63..e50a22a47b89385a674f661cf1e967e29c388b89 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -3812,7 +3812,8 @@ void PG::C_DeleteMore::complete(int r) {
   delete this;
 }
 
-void PG::do_delete_work(ObjectStore::Transaction &t)
+ghobject_t PG::do_delete_work(ObjectStore::Transaction &t,
+                        ghobject_t _next)
 {
   dout(10) << __func__ << dendl;
 
@@ -3838,25 +3839,45 @@ void PG::do_delete_work(ObjectStore::Transaction &t)
       osd->sleep_timer.add_event_at(delete_schedule_time,
                                    delete_requeue_callback);
       dout(20) << __func__ << " Delete scheduled at " << delete_schedule_time << dendl;
-      return;
+      return _next;
     }
   }
 
   delete_needs_sleep = true;
 
+  ghobject_t next;
+
   vector<ghobject_t> olist;
   int max = std::min(osd->store->get_ideal_list_max(),
                     (int)cct->_conf->osd_target_transaction_size);
-  ghobject_t next;
+
   osd->store->collection_list(
     ch,
-    next,
+    _next,
     ghobject_t::get_max(),
     max,
     &olist,
     &next);
   dout(20) << __func__ << " " << olist << dendl;
 
+  // make sure we've removed everything
+  // by one more listing from the beginning
+  if (_next != ghobject_t() && olist.empty()) {
+    next = ghobject_t();
+    osd->store->collection_list(
+      ch,
+      next,
+      ghobject_t::get_max(),
+      max,
+      &olist,
+      &next);
+    if (!olist.empty()) {
+      dout(0) << __func__ << " additional unexpected onode list"
+              <<" (new onodes has appeared since PG removal started"
+              << olist << dendl;
+    }
+  }
+
   OSDriver::OSTransaction _t(osdriver.get_transaction(&t));
   int64_t num = 0;
   for (auto& oid : olist) {
@@ -3879,7 +3900,6 @@ void PG::do_delete_work(ObjectStore::Transaction &t)
     Context *fin = new C_DeleteMore(this, get_osdmap_epoch());
     t.register_on_commit(fin);
   } else {
-    dout(20) << __func__ << " finished" << dendl;
     if (cct->_conf->osd_inject_failure_on_pg_removal) {
       _exit(1);
     }
@@ -3914,6 +3934,7 @@ void PG::do_delete_work(ObjectStore::Transaction &t)
       osd->logger->dec(l_osd_pg_removing);
     }
   }
+  return next;
 }
 
 int PG::pg_stat_adjust(osd_stat_t *ns)

diff --git a/src/osd/PG.h b/src/osd/PG.h
index 13f115b8b9bb8b2e26b70cebdd20bbf1e2860825..30ec9a515879987aecb7dc8ba045f8ac2ea6b31b 100644 (file)
--- a/src/osd/PG.h
+++ b/src/osd/PG.h
@@ -469,7 +469,8 @@ public:
     return std::make_unique<PG::PGLogEntryHandler>(this, &t);
   }
 
-  void do_delete_work(ObjectStore::Transaction &t) override;
+  ghobject_t do_delete_work(ObjectStore::Transaction &t,
+    ghobject_t _next) override;
 
   void clear_ready_to_merge() override;
   void set_not_ready_to_merge_target(pg_t pgid, pg_t src) override;

diff --git a/src/osd/PeeringState.cc b/src/osd/PeeringState.cc
index 47ba7d3614cd71f3b61755b17c2993a3bab9db76..d7495123475b1bd79f065b633856aa6d95233315 100644 (file)
--- a/src/osd/PeeringState.cc
+++ b/src/osd/PeeringState.cc
@@ -6304,7 +6304,10 @@ PeeringState::Deleting::Deleting(my_context ctx)
   : my_base(ctx),
     NamedState(context< PeeringMachine >().state_history, "Started/ToDelete/Deleting")
 {
+  start = ceph::mono_clock::now();
+
   context< PeeringMachine >().log_enter(state_name);
+
   DECLARE_LOCALS;
   ps->deleting = true;
   ObjectStore::Transaction &t = context<PeeringMachine>().get_cur_transaction();
@@ -6325,7 +6328,8 @@ boost::statechart::result PeeringState::Deleting::react(
   const DeleteSome& evt)
 {
   DECLARE_LOCALS;
-  pl->do_delete_work(context<PeeringMachine>().get_cur_transaction());
+  next = pl->do_delete_work(context<PeeringMachine>().get_cur_transaction(),
+    next);
   return discard_event();
 }
 
@@ -6335,6 +6339,9 @@ void PeeringState::Deleting::exit()
   DECLARE_LOCALS;
   ps->deleting = false;
   pl->cancel_local_background_io_reservation();
+  psdout(20) << "Deleting::" << __func__ << this <<" finished in "
+           << ceph::mono_clock::now() - start
+           << dendl;
 }
 
 /*--------GetInfo---------*/

diff --git a/src/osd/PeeringState.h b/src/osd/PeeringState.h
index ede2258065b85b4abf587da70632ce65a84b7ea1..68f842104b03ab5a562d5f288eb90e141144189c 100644 (file)
--- a/src/osd/PeeringState.h
+++ b/src/osd/PeeringState.h
@@ -379,7 +379,8 @@ public:
     /// Notification of removal complete, t must be populated to complete removal
     virtual void on_removal(ObjectStore::Transaction &t) = 0;
     /// Perform incremental removal work
-    virtual void do_delete_work(ObjectStore::Transaction &t) = 0;
+    virtual ghobject_t do_delete_work(ObjectStore::Transaction &t,
+      ghobject_t _next) = 0;
 
     // ======================= PG Merge =========================
     virtual void clear_ready_to_merge() = 0;
@@ -1220,6 +1221,8 @@ public:
       boost::statechart::custom_reaction< DeleteSome >,
       boost::statechart::transition<DeleteInterrupted, WaitDeleteReserved>
       > reactions;
+    ghobject_t next;
+    ceph::mono_clock::time_point start;
     explicit Deleting(my_context ctx);
     boost::statechart::result react(const DeleteSome &evt);
     void exit();