mgr: fix crash due to multiple sessions from daemons with same name

Don't assume perf counter instances are stable. If there are
multiple sessions from daemons reporting the same name (e.g. rgws),
the instances are cleared in DaemonServer::handle_open.
If the daemons have different counter names (for counters like
"objecter-0x55b6f4f1a630.op_active") it will throw out_of_range in
DaemonPerfCounters::update when accessing the instance removed by
another session.

The regression was introduced when adding avgcount support.

Fixes: https://tracker.ceph.com/issues/36244
Signed-off-by: Mykola Golub <mgolub@suse.com>
(cherry picked from commit 5b6037aeb43aff90111fdae08807ddda290bc570)

diff --git a/src/mgr/DaemonState.cc b/src/mgr/DaemonState.cc
index aeb80060c34c625c734348cca190dc02afc34259..1f41f2ad49457d5c0376773935b96144e1196349 100644 (file)
--- a/src/mgr/DaemonState.cc
+++ b/src/mgr/DaemonState.cc
@@ -141,8 +141,6 @@ void DaemonPerfCounters::update(MMgrReport *report)
   for (const auto &t : report->declare_types) {
     types.insert(std::make_pair(t.path, t));
     session->declared_types.insert(t.path);
-    instances.insert(std::pair<std::string, PerfCounterInstance>(
-                     t.path, PerfCounterInstance(t.type)));
   }
   // Remove any old types
   for (const auto &t : report->undeclare_types) {
@@ -156,6 +154,13 @@ void DaemonPerfCounters::update(MMgrReport *report)
   DECODE_START(1, p);
   for (const auto &t_path : session->declared_types) {
     const auto &t = types.at(t_path);
+    auto instances_it = instances.find(t_path);
+    // Always check the instance exists, as we don't prevent yet
+    // multiple sessions from daemons with the same name, and one
+    // session clearing stats created by another on open.
+    if (instances_it == instances.end()) {
+      instances_it = instances.insert({t_path, t.type}).first;
+    }
     uint64_t val = 0;
     uint64_t avgcount = 0;
     uint64_t avgcount2 = 0;
@@ -164,9 +169,9 @@ void DaemonPerfCounters::update(MMgrReport *report)
     if (t.type & PERFCOUNTER_LONGRUNAVG) {
       decode(avgcount, p);
       decode(avgcount2, p);
-      instances.at(t_path).push_avg(now, val, avgcount);
+      instances_it->second.push_avg(now, val, avgcount);
     } else {
-      instances.at(t_path).push(now, val);
+      instances_it->second.push(now, val);
     }
   }
   DECODE_FINISH(p);