update sources to v12.1.3

[ceph.git] / ceph / src / mds / FSMap.cc

diff --git a/ceph/src/mds/FSMap.cc b/ceph/src/mds/FSMap.cc
index eb08d02b7dcb007362210a1d6cb9298b79204f00..c605dfcd15716a1dc1585750f110ff58c60f51e6 100644 (file)
--- a/ceph/src/mds/FSMap.cc
+++ b/ceph/src/mds/FSMap.cc
@@ -18,6 +18,8 @@
 #include <sstream>
 using std::stringstream;
 
+#include "mon/health_check.h"
+
 
 void Filesystem::dump(Formatter *f) const
 {
@@ -124,16 +126,10 @@ void FSMap::print_summary(Formatter *f, ostream *out) const
       f->dump_unsigned("max", fs->mds_map.max_mds);
     }
   } else {
-    if (filesystems.size() == 1) {
-      auto fs = filesystems.begin()->second;
-      *out << fs->mds_map.up.size() << "/" << fs->mds_map.in.size() << "/"
-           << fs->mds_map.max_mds << " up";
-    } else {
-      for (auto i : filesystems) {
-        auto fs = i.second;
-        *out << fs->mds_map.fs_name << "-" << fs->mds_map.up.size() << "/"
-             << fs->mds_map.in.size() << "/" << fs->mds_map.max_mds << " up ";
-      }
+    for (auto i : filesystems) {
+      auto fs = i.second;
+      *out << fs->mds_map.fs_name << "-" << fs->mds_map.up.size() << "/"
+          << fs->mds_map.in.size() << "/" << fs->mds_map.max_mds << " up ";
     }
   }
 
@@ -293,6 +289,12 @@ void FSMap::reset_filesystem(fs_cluster_id_t fscid)
   new_fs->mds_map.standby_count_wanted = fs->mds_map.standby_count_wanted;
   new_fs->mds_map.enabled = true;
 
+  // Remember mds ranks that have ever started. (They should load old inotable
+  // instead of creating new one if they start again.)
+  new_fs->mds_map.stopped.insert(fs->mds_map.in.begin(), fs->mds_map.in.end());
+  new_fs->mds_map.stopped.insert(fs->mds_map.stopped.begin(), fs->mds_map.stopped.end());
+  new_fs->mds_map.stopped.erase(mds_rank_t(0));
+
   // Persist the new FSMap
   filesystems[new_fs->fscid] = new_fs;
 }
@@ -327,6 +329,55 @@ bool FSMap::check_health(void)
   return changed;
 }
 
+void FSMap::get_health_checks(health_check_map_t *checks) const
+{
+  mds_rank_t standby_count_wanted = 0;
+  for (const auto &i : filesystems) {
+    const auto &fs = i.second;
+    health_check_map_t fschecks;
+
+    fs->mds_map.get_health_checks(&fschecks);
+
+    // Some of the failed ranks might be transient (i.e. there are standbys
+    // ready to replace them).  We will report only on "stuck" failed, i.e.
+    // ranks which are failed and have no standby replacement available.
+    std::set<mds_rank_t> stuck_failed;
+
+    for (const auto &rank : fs->mds_map.failed) {
+      const mds_gid_t replacement = find_replacement_for(
+          {fs->fscid, rank}, {}, g_conf->mon_force_standby_active);
+      if (replacement == MDS_GID_NONE) {
+        stuck_failed.insert(rank);
+      }
+    }
+
+    // FS_WITH_FAILED_MDS
+    if (!stuck_failed.empty()) {
+      health_check_t& fscheck = checks->get_or_add(
+        "FS_WITH_FAILED_MDS", HEALTH_WARN,
+        "%num% filesystem%plurals% %isorare% have a failed mds daemon");
+      ostringstream ss;
+      ss << "fs " << fs->mds_map.fs_name << " has " << stuck_failed.size()
+         << " failed mds" << (stuck_failed.size() > 1 ? "s" : "");
+      fscheck.detail.push_back(ss.str()); }
+
+    checks->merge(fschecks);
+    standby_count_wanted = std::max(
+      standby_count_wanted,
+      fs->mds_map.get_standby_count_wanted((mds_rank_t)standby_daemons.size()));
+  }
+
+  // MDS_INSUFFICIENT_STANDBY
+  if (standby_count_wanted) {
+    std::ostringstream oss, dss;
+    oss << "insufficient standby MDS daemons available";
+    auto& d = checks->get_or_add("MDS_INSUFFICIENT_STANDBY", HEALTH_WARN, oss.str());
+    dss << "have " << standby_daemons.size() << "; want " << standby_count_wanted
+       << " more";
+    d.detail.push_back(dss.str());
+  }
+}
+
 void FSMap::encode(bufferlist& bl, uint64_t features) const
 {
   if (features & CEPH_FEATURE_SERVER_JEWEL) {