mon,mds: make standby-replay state more capable

src/mds/FSMap.cc

@@ -856,8 +856,7 @@ const MDSMap::mds_info_t* FSMap::find_replacement_for(mds_role_t role) const
   for (const auto& [gid, info] : fs.mds_map.mds_info) {
     if (info.rank == role.rank && info.state == MDSMap::STATE_STANDBY_REPLAY) {
       if (info.is_frozen()) {
-        /* the standby-replay is frozen, do nothing! */
-        return nullptr;
+        continue;
       } else {
         ceph_assert(info.compat.writeable(fs.mds_map.compat));
         return &info;

src/mds/MDSMap.cc

@@ -1263,16 +1263,25 @@ mds_gid_t MDSMap::get_standby_replay(mds_rank_t r) const {
   return MDS_GID_NONE;
 }
 
-bool MDSMap::is_followable(mds_rank_t r) const {
+int MDSMap::is_followable(mds_rank_t r) const {
+  bool found = true;
+  bool degraded = false;
+  int standby_replay_count = 0;
   if (auto it1 = up.find(r); it1 != up.end()) {
-    if (auto it2 = mds_info.find(it1->second); it2 != mds_info.end()) {
-      auto& info = it2->second;
-      if (!info.is_degraded() && !has_standby_replay(r)) {
-        return true;
+    for (auto& [gid,info] : mds_info) {
+      if (gid == it1->second) {
+        found = true;
+        if (info.is_degraded()) {
+          degraded = true;
+          break;
+        }
+      }
+      if (info.rank == r && info.state == STATE_STANDBY_REPLAY) {
+        ++standby_replay_count;
       }
     }
   }
-  return false;
+  return found && !degraded ? standby_replay_count : 0;
 }
 
 bool MDSMap::is_laggy_gid(mds_gid_t gid) const {

src/mds/MDSMap.h

@@ -522,7 +522,7 @@ class MDSMap {
     return get_standby_replay(r) != MDS_GID_NONE;
   }
 
-  bool is_followable(mds_rank_t r) const;
+  int is_followable(mds_rank_t r) const;
   bool is_laggy_gid(mds_gid_t gid) const;
 
   // degraded = some recovery in process.  fixes active membership and

src/mds/MDSRank.cc

@@ -750,7 +750,8 @@ void MDSRankDispatcher::tick()
 	set_mdsmap_multimds_snaps_allowed();
       }
     }
-
+  }
+  if (is_active() || is_standby_replay()) {
     if (whoami == 0) {
       scrubstack->advance_scrub_status();
       scrubstack->purge_old_scrub_counters();
@@ -2874,7 +2875,7 @@ void MDSRankDispatcher::handle_asok_command(
     r = config_client(client_id, !got_value, option, value, *css);
   } else if (command == "scrub start" ||
 	     command == "scrub_start") {
-    if (!is_active()) {
+    if (!is_active() && ! is_standby_replay()) {
       *css << "MDS is not active";
       r = -EINVAL;
       goto out;
@@ -2905,7 +2906,7 @@ void MDSRankDispatcher::handle_asok_command(
 	}));
     return;
   } else if (command == "scrub abort") {
-    if (!is_active()) {
+    if (!is_active() && !is_standby_replay()) {
       *css << "MDS is not active";
       r = -EINVAL;
       goto out;
@@ -2925,7 +2926,7 @@ void MDSRankDispatcher::handle_asok_command(
         }));
     return;
   } else if (command == "scrub pause") {
-    if (!is_active()) {
+    if (!is_active() && !is_standby_replay()) {
       *css << "MDS is not active";
       r = -EINVAL;
       goto out;
@@ -2945,7 +2946,7 @@ void MDSRankDispatcher::handle_asok_command(
         }));
     return;
   } else if (command == "scrub resume") {
-    if (!is_active()) {
+    if (!is_active() && !is_standby_replay()) {
       *css << "MDS is not active";
       r = -EINVAL;
       goto out;

src/mon/MDSMonitor.cc

@@ -2445,13 +2445,14 @@ bool MDSMonitor::maybe_promote_standby(FSMap &fsmap, const Filesystem& fs)
       bool changed = false;
       for (const auto& rank : mds_map.in) {
         dout(20) << "examining " << rank << dendl;
-        if (mds_map.is_followable(rank)) {
+        int followables = mds_map.is_followable(rank);
+        if (followables < 2 ) {
           dout(1) << "  setting mds." << info->global_id
                   << " to follow mds rank " << rank << dendl;
           fsmap.assign_standby_replay(info->global_id, fs.get_fscid(), rank);
           do_propose = true;
           changed = true;
-          break;
+          //break;
         }
       }
       if (!changed) break;