mon/OSDMonitor: trim no-longer-exist failure reporters

remove the report from failure_info in OSDMonitor::check_failure(), if
the reporter does not exist in the osdmap any longer. otherwise, we will
run into assert() failure when trying to lookup the osd in osdmap using
`osdmap.get_xinfo()`.

Fixes: https://tracker.ceph.com/issues/41680
Signed-off-by: NancySu05 <su_nan@inspur.com>

diff --git a/src/mon/OSDMonitor.cc b/src/mon/OSDMonitor.cc
index 9b08b0bba7b43dfb470dc04c1acf92d4f1443901..e21753e4d86d881a93e87d505d211f3945315434 100644 (file)
--- a/src/mon/OSDMonitor.cc
+++ b/src/mon/OSDMonitor.cc
@@ -876,7 +876,6 @@ void OSDMonitor::update_from_paxos(bool *need_bootstrap)
 
   share_map_with_random_osd();
   update_logger();
-
   process_failures();
 
   // make sure our feature bits reflect the latest map
@@ -3133,23 +3132,27 @@ bool OSDMonitor::check_failure(utime_t now, int target_osd, failure_info_t& fi)
   // help us localize the grace correction to a subset of the system
   // (say, a rack with a bad switch) that is unhappy.
   ceph_assert(fi.reporters.size());
-  for (map<int,failure_reporter_t>::iterator p = fi.reporters.begin();
-       p != fi.reporters.end();
-       ++p) {
+  for (auto p = fi.reporters.begin(); p != fi.reporters.end();) {
     // get the parent bucket whose type matches with "reporter_subtree_level".
     // fall back to OSD if the level doesn't exist.
-    map<string, string> reporter_loc = osdmap.crush->get_full_location(p->first);
-    map<string, string>::iterator iter = reporter_loc.find(reporter_subtree_level);
-    if (iter == reporter_loc.end()) {
-      reporters_by_subtree.insert("osd." + to_string(p->first));
+    if (osdmap.exists(p->first)) {
+      auto reporter_loc = osdmap.crush->get_full_location(p->first);
+      if (auto iter = reporter_loc.find(reporter_subtree_level);
+          iter == reporter_loc.end()) {
+        reporters_by_subtree.insert("osd." + to_string(p->first));
+      } else {
+        reporters_by_subtree.insert(iter->second);
+      }
+      if (g_conf()->mon_osd_adjust_heartbeat_grace) {
+        const osd_xinfo_t& xi = osdmap.get_xinfo(p->first);
+        utime_t elapsed = now - xi.down_stamp;
+        double decay = exp((double)elapsed * decay_k);
+        peer_grace += decay * (double)xi.laggy_interval * xi.laggy_probability;
+      }
+      ++p;
     } else {
-      reporters_by_subtree.insert(iter->second);
-    }
-    if (g_conf()->mon_osd_adjust_heartbeat_grace) {
-      const osd_xinfo_t& xi = osdmap.get_xinfo(p->first);
-      utime_t elapsed = now - xi.down_stamp;
-      double decay = exp((double)elapsed * decay_k);
-      peer_grace += decay * (double)xi.laggy_interval * xi.laggy_probability;
+      fi.cancel_report(p->first);;
+      p = fi.reporters.erase(p);
     }
   }