mds: optimize state check of peer mds

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

diff --git a/src/mds/Locker.cc b/src/mds/Locker.cc
index e9cd6434dc4d0ddcd6a639c17dfbbbfb4eb60dd7..899cd0082c5bc138742025a68d26abeacb3147e4 100644 (file)
--- a/src/mds/Locker.cc
+++ b/src/mds/Locker.cc
@@ -135,7 +135,8 @@ void Locker::send_lock_message(SimpleLock *lock, int msg)
   for (compact_map<mds_rank_t,unsigned>::iterator it = lock->get_parent()->replicas_begin();
        it != lock->get_parent()->replicas_end();
        ++it) {
-    if (mds->mdsmap->get_state(it->first) < MDSMap::STATE_REJOIN) 
+    if (mds->is_cluster_degraded() &&
+       mds->mdsmap->get_state(it->first) < MDSMap::STATE_REJOIN) 
       continue;
     MLock *m = new MLock(lock, msg, mds->get_nodeid());
     mds->send_message_mds(m, it->first);
@@ -147,7 +148,8 @@ void Locker::send_lock_message(SimpleLock *lock, int msg, const bufferlist &data
   for (compact_map<mds_rank_t,unsigned>::iterator it = lock->get_parent()->replicas_begin();
        it != lock->get_parent()->replicas_end();
        ++it) {
-    if (mds->mdsmap->get_state(it->first) < MDSMap::STATE_REJOIN) 
+    if (mds->is_cluster_degraded() &&
+       mds->mdsmap->get_state(it->first) < MDSMap::STATE_REJOIN) 
       continue;
     MLock *m = new MLock(lock, msg, mds->get_nodeid());
     m->set_data(data);
@@ -401,7 +403,8 @@ bool Locker::acquire_locks(MDRequestRef& mdr,
       dout(10) << "requesting remote auth_pins from mds." << p->first << dendl;
 
       // wait for active auth
-      if (!mds->mdsmap->is_clientreplay_or_active_or_stopping(p->first)) {
+      if (mds->is_cluster_degraded() &&
+         !mds->mdsmap->is_clientreplay_or_active_or_stopping(p->first)) {
        dout(10) << " mds." << p->first << " is not active" << dendl;
        if (mdr->more()->waiting_on_slave.empty())
          mds->wait_for_active_peer(p->first, new C_MDS_RetryRequest(mdcache, mdr));
@@ -648,7 +651,8 @@ void Locker::_drop_non_rdlocks(MutationImpl *mut, set<CInode*> *pneed_issue)
   }
 
   for (set<mds_rank_t>::iterator p = slaves.begin(); p != slaves.end(); ++p) {
-    if (mds->mdsmap->get_state(*p) >= MDSMap::STATE_REJOIN) {
+    if (!mds->is_cluster_degraded() ||
+       mds->mdsmap->get_state(*p) >= MDSMap::STATE_REJOIN) {
       dout(10) << "_drop_non_rdlocks dropping remote locks on mds." << *p << dendl;
       MMDSSlaveRequest *slavereq = new MMDSSlaveRequest(mut->reqid, mut->attempt,
                                                        MMDSSlaveRequest::OP_DROPLOCKS);
@@ -797,7 +801,8 @@ void Locker::eval_gather(SimpleLock *lock, bool first, bool *pneed_issue, list<M
        return;
       }
 
-      if (mds->mdsmap->get_state(auth) >= MDSMap::STATE_REJOIN) {
+      if (!mds->is_cluster_degraded() ||
+         mds->mdsmap->get_state(auth) >= MDSMap::STATE_REJOIN) {
        switch (lock->get_state()) {
        case LOCK_SYNC_LOCK:
          mds->send_message_mds(new MLock(lock, LOCK_AC_LOCKACK, mds->get_nodeid()),
@@ -1195,7 +1200,8 @@ bool Locker::_rdlock_kick(SimpleLock *lock, bool as_anon)
     } else {
       // request rdlock state change from auth
       mds_rank_t auth = lock->get_parent()->authority().first;
-      if (mds->mdsmap->is_clientreplay_or_active_or_stopping(auth)) {
+      if (!mds->is_cluster_degraded() ||
+         mds->mdsmap->is_clientreplay_or_active_or_stopping(auth)) {
        dout(10) << "requesting rdlock from auth on "
                 << *lock << " on " << *lock->get_parent() << dendl;
        mds->send_message_mds(new MLock(lock, LOCK_AC_REQRDLOCK, mds->get_nodeid()), auth);
@@ -1423,7 +1429,8 @@ bool Locker::wrlock_start(SimpleLock *lock, MDRequestRef& mut, bool nowait)
       // replica.
       // auth should be auth_pinned (see acquire_locks wrlock weird mustpin case).
       mds_rank_t auth = lock->get_parent()->authority().first;
-      if (mds->mdsmap->is_clientreplay_or_active_or_stopping(auth)) {
+      if (!mds->is_cluster_degraded() ||
+         mds->mdsmap->is_clientreplay_or_active_or_stopping(auth)) {
        dout(10) << "requesting scatter from auth on "
                 << *lock << " on " << *lock->get_parent() << dendl;
        mds->send_message_mds(new MLock(lock, LOCK_AC_REQSCATTER, mds->get_nodeid()), auth);
@@ -1471,7 +1478,8 @@ void Locker::remote_wrlock_start(SimpleLock *lock, mds_rank_t target, MDRequestR
   dout(7) << "remote_wrlock_start mds." << target << " on " << *lock << " on " << *lock->get_parent() << dendl;
 
   // wait for active target
-  if (!mds->mdsmap->is_clientreplay_or_active_or_stopping(target)) {
+  if (mds->is_cluster_degraded() &&
+      !mds->mdsmap->is_clientreplay_or_active_or_stopping(target)) {
     dout(7) << " mds." << target << " is not active" << dendl;
     if (mut->more()->waiting_on_slave.empty())
       mds->wait_for_active_peer(target, new C_MDS_RetryRequest(mdcache, mut));
@@ -1501,7 +1509,8 @@ void Locker::remote_wrlock_finish(SimpleLock *lock, mds_rank_t target,
   
   dout(7) << "remote_wrlock_finish releasing remote wrlock on mds." << target
          << " " << *lock->get_parent()  << dendl;
-  if (mds->mdsmap->get_state(target) >= MDSMap::STATE_REJOIN) {
+  if (!mds->is_cluster_degraded() ||
+      mds->mdsmap->get_state(target) >= MDSMap::STATE_REJOIN) {
     MMDSSlaveRequest *slavereq = new MMDSSlaveRequest(mut->reqid, mut->attempt,
                                                      MMDSSlaveRequest::OP_UNWRLOCK);
     slavereq->set_lock_type(lock->get_type());
@@ -1573,7 +1582,8 @@ bool Locker::xlock_start(SimpleLock *lock, MDRequestRef& mut)
     
     // wait for active auth
     mds_rank_t auth = lock->get_parent()->authority().first;
-    if (!mds->mdsmap->is_clientreplay_or_active_or_stopping(auth)) {
+    if (mds->is_cluster_degraded() &&
+       !mds->mdsmap->is_clientreplay_or_active_or_stopping(auth)) {
       dout(7) << " mds." << auth << " is not active" << dendl;
       if (mut->more()->waiting_on_slave.empty())
        mds->wait_for_active_peer(auth, new C_MDS_RetryRequest(mdcache, mut));
@@ -1647,7 +1657,8 @@ void Locker::xlock_finish(SimpleLock *lock, MutationImpl *mut, bool *pneed_issue
     // tell auth
     dout(7) << "xlock_finish releasing remote xlock on " << *lock->get_parent()  << dendl;
     mds_rank_t auth = lock->get_parent()->authority().first;
-    if (mds->mdsmap->get_state(auth) >= MDSMap::STATE_REJOIN) {
+    if (!mds->is_cluster_degraded() ||
+       mds->mdsmap->get_state(auth) >= MDSMap::STATE_REJOIN) {
       MMDSSlaveRequest *slavereq = new MMDSSlaveRequest(mut->reqid, mut->attempt,
                                                        MMDSSlaveRequest::OP_UNXLOCK);
       slavereq->set_lock_type(lock->get_type());
@@ -2124,7 +2135,8 @@ void Locker::request_inode_file_caps(CInode *in)
     }
 
     mds_rank_t auth = in->authority().first;
-    if (mds->mdsmap->get_state(auth) == MDSMap::STATE_REJOIN) {
+    if (mds->is_cluster_degraded() &&
+       mds->mdsmap->get_state(auth) == MDSMap::STATE_REJOIN) {
       mds->wait_for_active_peer(auth, new C_MDL_RequestInodeFileCaps(this, in));
       return;
     }
@@ -2135,7 +2147,8 @@ void Locker::request_inode_file_caps(CInode *in)
 
     in->replica_caps_wanted = wanted;
 
-    if (mds->mdsmap->is_clientreplay_or_active_or_stopping(auth))
+    if (!mds->is_cluster_degraded() ||
+       mds->mdsmap->is_clientreplay_or_active_or_stopping(auth))
       mds->send_message_mds(new MInodeFileCaps(in->ino(), in->replica_caps_wanted),
                            auth);
   }
@@ -4501,7 +4514,8 @@ void Locker::scatter_nudge(ScatterLock *lock, MDSInternalContextBase *c, bool fo
             << *lock << " on " << *p << dendl;
     // request unscatter?
     mds_rank_t auth = lock->get_parent()->authority().first;
-    if (mds->mdsmap->is_clientreplay_or_active_or_stopping(auth))
+    if (!mds->is_cluster_degraded() ||
+       mds->mdsmap->is_clientreplay_or_active_or_stopping(auth))
       mds->send_message_mds(new MLock(lock, LOCK_AC_NUDGE, mds->get_nodeid()), auth);
 
     // wait...

diff --git a/src/mds/MDBalancer.cc b/src/mds/MDBalancer.cc
index afe4296f179f691557301972550bc6d7db26724d..4914e76551713774deb2ae926465ed4383ba3852 100644 (file)
--- a/src/mds/MDBalancer.cc
+++ b/src/mds/MDBalancer.cc
@@ -211,7 +211,7 @@ void MDBalancer::send_heartbeat()
 {
   utime_t now = ceph_clock_now();
   
-  if (mds->mdsmap->is_degraded()) {
+  if (mds->is_cluster_degraded()) {
     dout(10) << "send_heartbeat degraded" << dendl;
     return;
   }
@@ -284,7 +284,7 @@ void MDBalancer::handle_heartbeat(MHeartbeat *m)
     return;
   }
 
-  if (mds->mdsmap->is_degraded()) {
+  if (mds->is_cluster_degraded()) {
     dout(10) << " degraded, ignoring" << dendl;
     goto out;
   }

diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index c8f21e21f21b7f793a7584e23b51f322a9b76fb0..029d9d0dc49eba8f96fd8606060ace7501416610 100644 (file)
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -6528,9 +6528,10 @@ void MDCache::send_expire_messages(map<mds_rank_t, MCacheExpire*>& expiremap)
   for (map<mds_rank_t, MCacheExpire*>::iterator it = expiremap.begin();
        it != expiremap.end();
        ++it) {
-    if (mds->mdsmap->get_state(it->first) < MDSMap::STATE_REJOIN ||
-       (mds->mdsmap->get_state(it->first) == MDSMap::STATE_REJOIN &&
-        rejoin_sent.count(it->first) == 0)) {
+    if (mds->is_cluster_degraded() &&
+       (mds->mdsmap->get_state(it->first) < MDSMap::STATE_REJOIN ||
+        (mds->mdsmap->get_state(it->first) == MDSMap::STATE_REJOIN &&
+         rejoin_sent.count(it->first) == 0))) {
       it->second->put();
       continue;
     }
@@ -8159,7 +8160,8 @@ void MDCache::open_remote_dirfrag(CInode *diri, frag_t approxfg, MDSInternalCont
 
   mds_rank_t auth = diri->authority().first;
 
-  if (mds->mdsmap->get_state(auth) >= MDSMap::STATE_REJOIN) {
+  if (!mds->is_cluster_degraded() ||
+      mds->mdsmap->get_state(auth) >= MDSMap::STATE_REJOIN) {
     discover_dir_frag(diri, approxfg, fin);
   } else {
     // mds is down or recovering.  forge a replica!
@@ -10755,7 +10757,7 @@ bool MDCache::can_fragment(CInode *diri, list<CDir*>& dirs)
     dout(7) << "can_fragment: read-only FS, no fragmenting for now" << dendl;
     return false;
   }
-  if (mds->mdsmap->is_degraded()) {
+  if (mds->is_cluster_degraded()) {
     dout(7) << "can_fragment: cluster degraded, no fragmenting for now" << dendl;
     return false;
   }

diff --git a/src/mds/MDSRank.cc b/src/mds/MDSRank.cc
index fd5e233cf47adba9c099a15e53f883406eb3391d..b98102f48e136ceb29cbf2ab15994226fb7fb561 100644 (file)
--- a/src/mds/MDSRank.cc
+++ b/src/mds/MDSRank.cc
@@ -66,7 +66,7 @@ MDSRank::MDSRank(
                g_conf->osd_num_op_tracker_shard),
     last_state(MDSMap::STATE_BOOT),
     state(MDSMap::STATE_BOOT),
-    stopping(false),
+    cluster_degraded(false), stopping(false),
     purge_queue(g_ceph_context, whoami_,
       mdsmap_->get_metadata_pool(), objecter,
       new FunctionContext(
@@ -1573,8 +1573,10 @@ void MDSRankDispatcher::handle_mds_map(
     }
   }
 
-  if (oldmap->is_degraded() && !mdsmap->is_degraded() && state >= MDSMap::STATE_ACTIVE)
+  cluster_degraded = mdsmap->is_degraded();
+  if (oldmap->is_degraded() && !cluster_degraded && state >= MDSMap::STATE_ACTIVE) {
     dout(1) << "cluster recovered." << dendl;
+  }
 
   // did someone go active?
   if (oldstate >= MDSMap::STATE_CLIENTREPLAY &&

diff --git a/src/mds/MDSRank.h b/src/mds/MDSRank.h
index e8d7af2fc5ce1454f6f39f3f1f247885cbffcf3a..717dbd268b183c09bda4a3e643574cba6a04b428 100644 (file)
--- a/src/mds/MDSRank.h
+++ b/src/mds/MDSRank.h
@@ -181,6 +181,8 @@ class MDSRank {
     // The state assigned to me by the MDSMap
     MDSMap::DaemonState state;
 
+    bool cluster_degraded;
+
     MDSMap::DaemonState get_state() const { return state; } 
     MDSMap::DaemonState get_want_state() const { return beacon.get_want_state(); } 
 
@@ -197,6 +199,7 @@ class MDSRank {
     bool is_stopping() const { return state == MDSMap::STATE_STOPPING; }
     bool is_any_replay() const { return (is_replay() || is_standby_replay()); }
     bool is_stopped() const { return mdsmap->is_stopped(whoami); }
+    bool is_cluster_degraded() const { return cluster_degraded; }
 
     void handle_write_error(int err);
 

diff --git a/src/mds/Migrator.cc b/src/mds/Migrator.cc
index 52df1588205402dc75a767423458f61af244d1ce..dedd3bea46b0ec234c17ad25aaaa143a095691c9 100644 (file)
--- a/src/mds/Migrator.cc
+++ b/src/mds/Migrator.cc
@@ -280,7 +280,9 @@ void Migrator::export_try_cancel(CDir *dir, bool notify_peer)
     it->second.state = EXPORT_CANCELLED;
     dir->unfreeze_tree();  // cancel the freeze
     dir->auth_unpin(this);
-    if (notify_peer && mds->mdsmap->is_clientreplay_or_active_or_stopping(it->second.peer)) // tell them.
+    if (notify_peer &&
+       (!mds->is_cluster_degraded() ||
+        mds->mdsmap->is_clientreplay_or_active_or_stopping(it->second.peer))) // tell them.
       mds->send_message_mds(new MExportDirCancel(dir->dirfrag(), it->second.tid), it->second.peer);
     break;
 
@@ -288,7 +290,9 @@ void Migrator::export_try_cancel(CDir *dir, bool notify_peer)
     dout(10) << "export state=freezing : canceling freeze" << dendl;
     it->second.state = EXPORT_CANCELLED;
     dir->unfreeze_tree();  // cancel the freeze
-    if (notify_peer && mds->mdsmap->is_clientreplay_or_active_or_stopping(it->second.peer)) // tell them.
+    if (notify_peer &&
+       (!mds->is_cluster_degraded() ||
+        mds->mdsmap->is_clientreplay_or_active_or_stopping(it->second.peer))) // tell them.
       mds->send_message_mds(new MExportDirCancel(dir->dirfrag(), it->second.tid), it->second.peer);
     break;
 
@@ -325,7 +329,9 @@ void Migrator::export_try_cancel(CDir *dir, bool notify_peer)
     dir->unfreeze_tree();
     cache->adjust_subtree_auth(dir, mds->get_nodeid());
     cache->try_subtree_merge(dir);
-    if (notify_peer && mds->mdsmap->is_clientreplay_or_active_or_stopping(it->second.peer)) // tell them.
+    if (notify_peer &&
+       (!mds->is_cluster_degraded() ||
+        mds->mdsmap->is_clientreplay_or_active_or_stopping(it->second.peer))) // tell them.
       mds->send_message_mds(new MExportDirCancel(dir->dirfrag(), it->second.tid), it->second.peer);
     break;
 
@@ -765,7 +771,7 @@ void Migrator::export_dir(CDir *dir, mds_rank_t dest)
     dout(7) << "read-only FS, no exports for now" << dendl;
     return;
   }
-  if (mds->mdsmap->is_degraded()) {
+  if (mds->is_cluster_degraded()) {
     dout(7) << "cluster degraded, no exports for now" << dendl;
     return;
   }
@@ -1180,7 +1186,8 @@ void Migrator::handle_export_prep_ack(MExportDirPrepAck *m)
        p != dir->replicas_end();
        ++p) {
     if (p->first == it->second.peer) continue;
-    if (!mds->mdsmap->is_clientreplay_or_active_or_stopping(p->first))
+    if (mds->is_cluster_degraded() &&
+       !mds->mdsmap->is_clientreplay_or_active_or_stopping(p->first))
       continue;  // only if active
     it->second.warning_ack_waiting.insert(p->first);
     it->second.notify_ack_waiting.insert(p->first);  // we'll eventually get a notifyack, too!
@@ -1771,7 +1778,8 @@ void Migrator::export_logged_finish(CDir *dir)
     export_finish(dir);  // skip notify/notify_ack stage.
   } else {
     // notify peer to send cap import messages to clients
-    if (mds->mdsmap->is_clientreplay_or_active_or_stopping(stat.peer)) {
+    if (!mds->is_cluster_degraded() ||
+       mds->mdsmap->is_clientreplay_or_active_or_stopping(stat.peer)) {
       mds->send_message_mds(new MExportDirFinish(dir->dirfrag(), false, stat.tid), stat.peer);
     } else {
       dout(7) << "not sending MExportDirFinish, dest has failed" << dendl;
@@ -1854,7 +1862,8 @@ void Migrator::export_finish(CDir *dir)
   }
 
   // send finish/commit to new auth
-  if (mds->mdsmap->is_clientreplay_or_active_or_stopping(it->second.peer)) {
+  if (!mds->is_cluster_degraded() ||
+      mds->mdsmap->is_clientreplay_or_active_or_stopping(it->second.peer)) {
     mds->send_message_mds(new MExportDirFinish(dir->dirfrag(), true, it->second.tid), it->second.peer);
   } else {
     dout(7) << "not sending MExportDirFinish last, dest has failed" << dendl;
@@ -2561,7 +2570,8 @@ void Migrator::import_notify_abort(CDir *dir, set<CDir*>& bounds)
   import_state_t& stat = import_state[dir->dirfrag()];
   for (set<mds_rank_t>::iterator p = stat.bystanders.begin();
        p != stat.bystanders.end(); ) {
-    if (!mds->mdsmap->is_clientreplay_or_active_or_stopping(*p)) {
+    if (mds->is_cluster_degraded() &&
+       !mds->mdsmap->is_clientreplay_or_active_or_stopping(*p)) {
       // this can happen if both exporter and bystander fail in the same mdsmap epoch
       stat.bystanders.erase(p++);
       continue;

diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index 02acd6ebee143e16b5f5efc430714a04f38d4aa6..4ecc7085bd701afc37f37e4cb8b2bdcc1f3f4e2c 100644 (file)
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -5086,7 +5086,8 @@ void Server::_link_remote(MDRequestRef& mdr, bool inc, CDentry *dn, CInode *targ
   // 1. send LinkPrepare to dest (journal nlink++ prepare)
   mds_rank_t linkauth = targeti->authority().first;
   if (mdr->more()->witnessed.count(linkauth) == 0) {
-    if (!mds->mdsmap->is_clientreplay_or_active_or_stopping(linkauth)) {
+    if (mds->is_cluster_degraded() &&
+       !mds->mdsmap->is_clientreplay_or_active_or_stopping(linkauth)) {
       dout(10) << " targeti auth mds." << linkauth << " is not active" << dendl;
       if (mdr->more()->waiting_on_slave.empty())
        mds->wait_for_active_peer(linkauth, new C_MDS_RetryRequest(mdcache, mdr));
@@ -5768,7 +5769,8 @@ void Server::_unlink_local_finish(MDRequestRef& mdr,
 
 bool Server::_rmdir_prepare_witness(MDRequestRef& mdr, mds_rank_t who, vector<CDentry*>& trace, CDentry *straydn)
 {
-  if (!mds->mdsmap->is_clientreplay_or_active_or_stopping(who)) {
+  if (mds->is_cluster_degraded() &&
+      !mds->mdsmap->is_clientreplay_or_active_or_stopping(who)) {
     dout(10) << "_rmdir_prepare_witness mds." << who << " is not active" << dendl;
     if (mdr->more()->waiting_on_slave.empty())
       mds->wait_for_active_peer(who, new C_MDS_RetryRequest(mdcache, mdr));
@@ -6609,7 +6611,8 @@ void Server::_rename_finish(MDRequestRef& mdr, CDentry *srcdn, CDentry *destdn,
 bool Server::_rename_prepare_witness(MDRequestRef& mdr, mds_rank_t who, set<mds_rank_t> &witnesse,
                                     vector<CDentry*>& srctrace, vector<CDentry*>& dsttrace, CDentry *straydn)
 {
-  if (!mds->mdsmap->is_clientreplay_or_active_or_stopping(who)) {
+  if (mds->is_cluster_degraded() &&
+      !mds->mdsmap->is_clientreplay_or_active_or_stopping(who)) {
     dout(10) << "_rename_prepare_witness mds." << who << " is not active" << dendl;
     if (mdr->more()->waiting_on_slave.empty())
       mds->wait_for_active_peer(who, new C_MDS_RetryRequest(mdcache, mdr));
@@ -7286,7 +7289,8 @@ void Server::handle_slave_rename_prep(MDRequestRef& mdr)
       // make sure bystanders have received all lock related messages
       for (set<mds_rank_t>::iterator p = srcdnrep.begin(); p != srcdnrep.end(); ++p) {
        if (*p == mdr->slave_to_mds ||
-           !mds->mdsmap->is_clientreplay_or_active_or_stopping(*p))
+           (mds->is_cluster_degraded() &&
+            !mds->mdsmap->is_clientreplay_or_active_or_stopping(*p)))
          continue;
        MMDSSlaveRequest *notify = new MMDSSlaveRequest(mdr->reqid, mdr->attempt,
            MMDSSlaveRequest::OP_RENAMENOTIFY);