mds: record completed cap flushes in session map

This information is for detecting duplicated cap flushes when MDS
failovers. We only record completed cap flushes for clients whose
cap messages include TID of oldest cap flushes.

Signed-off-by: Yan, Zheng <zyan@redhat.com>

diff --git a/src/mds/Locker.cc b/src/mds/Locker.cc
index 9d4bf2dbc5e996aaa56a857c61bd0298cf243e1a..7a4956381c2d39f6d14b59bc752c12b34bf732c6 100644 (file)
--- a/src/mds/Locker.cc
+++ b/src/mds/Locker.cc
@@ -1715,8 +1715,18 @@ void Locker::file_update_finish(CInode *in, MutationRef& mut, bool share, client
 
   mut->apply();
   
-  if (ack)
-    mds->send_message_client_counted(ack, client);
+  if (ack) {
+    Session *session = mds->get_session(client);
+    if (session) {
+      // "oldest flush tid" > 0 means client uses unique TID for each flush
+      if (ack->get_oldest_flush_tid() > 0)
+       session->add_completed_flush(ack->get_client_tid());
+      mds->send_message_client_counted(ack, session);
+    } else {
+      dout(10) << " no session for client." << client << " " << *ack << dendl;
+      ack->put();
+    }
+  }
 
   set<CInode*> need_issue;
   drop_locks(mut.get(), &need_issue);
@@ -2456,6 +2466,13 @@ void Locker::handle_client_caps(MClientCaps *m)
     return;
   }
 
+  // "oldest flush tid" > 0 means client uses unique TID for each flush
+  if (m->get_oldest_flush_tid() > 0) {
+    if (session->trim_completed_flushes(m->get_oldest_flush_tid())) {
+      mds->mdlog->get_current_segment()->touched_sessions.insert(session->info.inst.name);
+    }
+  }
+
   CInode *head_in = mdcache->get_inode(m->get_ino());
   if (!head_in) {
     dout(7) << "handle_client_caps on unknown ino " << m->get_ino() << ", dropping" << dendl;
@@ -2526,6 +2543,7 @@ void Locker::handle_client_caps(MClientCaps *m)
       ack = new MClientCaps(CEPH_CAP_OP_FLUSHSNAP_ACK, in->ino(), 0, 0, 0, 0, 0, m->get_dirty(), 0, mds->get_osd_epoch_barrier());
       ack->set_snap_follows(follows);
       ack->set_client_tid(m->get_client_tid());
+      ack->set_oldest_flush_tid(m->get_oldest_flush_tid());
     }
 
     if (in == head_in ||
@@ -2601,6 +2619,7 @@ void Locker::handle_client_caps(MClientCaps *m)
       ack = new MClientCaps(CEPH_CAP_OP_FLUSH_ACK, in->ino(), 0, cap->get_cap_id(), m->get_seq(),
                            m->get_caps(), 0, m->get_dirty(), 0, mds->get_osd_epoch_barrier());
       ack->set_client_tid(m->get_client_tid());
+      ack->set_oldest_flush_tid(m->get_oldest_flush_tid());
     }
 
     // filter wanted based on what we could ever give out (given auth/replica status)
@@ -2877,6 +2896,11 @@ void Locker::_do_snap_update(CInode *in, snapid_t snap, int dirty, snapid_t foll
   mdcache->predirty_journal_parents(mut, &le->metablob, in, 0, PREDIRTY_PRIMARY, 0, follows);
   mdcache->journal_dirty_inode(mut.get(), &le->metablob, in, follows);
 
+  // "oldest flush tid" > 0 means client uses unique TID for each flush
+  if (ack && ack->get_oldest_flush_tid() > 0)
+    le->metablob.add_client_flush(metareqid_t(m->get_source(), ack->get_client_tid()),
+                                 ack->get_oldest_flush_tid());
+
   mds->mdlog->submit_entry(le, new C_Locker_FileUpdate_finish(this, in, mut,
                                                                 false,
                                                                 client, NULL,
@@ -3122,6 +3146,11 @@ bool Locker::_do_cap_update(CInode *in, Capability *cap,
   mdcache->predirty_journal_parents(mut, &le->metablob, in, 0, PREDIRTY_PRIMARY, 0, follows);
   mdcache->journal_dirty_inode(mut.get(), &le->metablob, in, follows);
 
+  // "oldest flush tid" > 0 means client uses unique TID for each flush
+  if (ack && ack->get_oldest_flush_tid() > 0)
+    le->metablob.add_client_flush(metareqid_t(m->get_source(), ack->get_client_tid()),
+                                 ack->get_oldest_flush_tid());
+
   mds->mdlog->submit_entry(le, new C_Locker_FileUpdate_finish(this, in, mut,
                                                                 change_max,
                                                                 client, cap,

diff --git a/src/mds/SessionMap.h b/src/mds/SessionMap.h
index 93eb66095bee85548a2cf36af68ff1f7c2d47d7f..b73246e5baab34aaa937bf8cc61738a0c4453b82 100644 (file)
--- a/src/mds/SessionMap.h
+++ b/src/mds/SessionMap.h
@@ -261,6 +261,25 @@ public:
     return true;
   }
 
+  void add_completed_flush(ceph_tid_t tid) {
+    info.completed_flushes.insert(tid);
+  }
+  bool trim_completed_flushes(ceph_tid_t mintid) {
+    bool erased_any = false;
+    while (!info.completed_flushes.empty() &&
+       (mintid == 0 || *info.completed_flushes.begin() < mintid)) {
+      info.completed_flushes.erase(info.completed_flushes.begin());
+      erased_any = true;
+    }
+    if (erased_any) {
+      completed_requests_dirty = true;
+    }
+    return erased_any;
+  }
+  bool have_completed_flush(ceph_tid_t tid) const {
+    return info.completed_flushes.count(tid);
+  }
+
   unsigned get_num_completed_requests() const { return info.completed_requests.size(); }
   unsigned get_num_trim_requests_warnings() { return num_trim_requests_warnings; }
   void inc_num_trim_requests_warnings() { ++num_trim_requests_warnings; }

diff --git a/src/mds/mdstypes.cc b/src/mds/mdstypes.cc
index 28843a58a2879a1a3c35b8055549703994f05c2d..d0218ec8061264f34f01e0a1b99371d269e23412 100644 (file)
--- a/src/mds/mdstypes.cc
+++ b/src/mds/mdstypes.cc
@@ -665,12 +665,13 @@ void old_rstat_t::generate_test_instances(list<old_rstat_t*>& ls)
  */
 void session_info_t::encode(bufferlist& bl) const
 {
-  ENCODE_START(4, 3, bl);
+  ENCODE_START(5, 3, bl);
   ::encode(inst, bl);
   ::encode(completed_requests, bl);
   ::encode(prealloc_inos, bl);   // hacky, see below.
   ::encode(used_inos, bl);
   ::encode(client_metadata, bl);
+  ::encode(completed_flushes, bl);
   ENCODE_FINISH(bl);
 }
 
@@ -695,6 +696,9 @@ void session_info_t::decode(bufferlist::iterator& p)
   if (struct_v >= 4) {
     ::decode(client_metadata, p);
   }
+  if (struct_v >= 5) {
+    ::decode(completed_flushes, p);
+  }
   DECODE_FINISH(p);
 }
 

diff --git a/src/mds/mdstypes.h b/src/mds/mdstypes.h
index 9ed51e2499389c055f4a5b5c802d4322d4d1b26a..6fb5f12de4de5eb196ae23769c094d5804684b2c 100644 (file)
--- a/src/mds/mdstypes.h
+++ b/src/mds/mdstypes.h
@@ -642,6 +642,7 @@ struct session_info_t {
   interval_set<inodeno_t> prealloc_inos;   // preallocated, ready to use.
   interval_set<inodeno_t> used_inos;       // journaling use
   std::map<std::string, std::string> client_metadata;
+  std::set<ceph_tid_t> completed_flushes;
 
   client_t get_client() const { return client_t(inst.name.num()); }
 
@@ -649,6 +650,7 @@ struct session_info_t {
     prealloc_inos.clear();
     used_inos.clear();
     completed_requests.clear();
+    completed_flushes.clear();
   }
 
   void encode(bufferlist& bl) const;