From: Brad Hubbard <bhubbard@redhat.com>
Date: Tue, 3 Apr 2018 07:49:08 +0000 (+1000)
Subject: mon/PGMap: Summarise OSDs in blocked/stuck requests
X-Git-Tag: v12.2.6~168^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=81fa8a7499a7b688aeaa28fba181033dcf692f54;p=ceph.git

mon/PGMap: Summarise OSDs in blocked/stuck requests

This fix is luminous specific since this code was changed in Mimic to
use the new SLOW_OPS errors.

Fixes: https://tracker.ceph.com/issues/23205

Signed-off-by: Brad Hubbard <bhubbard@redhat.com>
---

diff --git a/src/mon/PGMap.cc b/src/mon/PGMap.cc
index 504adb53cd94..db19af042359 100644
--- a/src/mon/PGMap.cc
+++ b/src/mon/PGMap.cc
@@ -3185,14 +3185,11 @@ void PGMap::get_health_checks(
     }
 
     if (!warn_detail.empty()) {
-      ostringstream ss;
-      ss << warn << " slow requests are blocked > "
-	 << cct->_conf->mon_osd_warn_op_age << " sec";
-      auto& d = checks->add("REQUEST_SLOW", HEALTH_WARN, ss.str());
-      d.detail.swap(warn_detail);
       int left = max;
+      set<int> implicated_osds;
       for (auto& p : warn_osd_by_max) {
 	ostringstream ss;
+        implicated_osds.insert(p.second.begin(), p.second.end());
 	if (p.second.size() > 1) {
 	  ss << "osds " << p.second
              << " have blocked requests > " << p.first << " sec";
@@ -3200,21 +3197,24 @@ void PGMap::get_health_checks(
 	  ss << "osd." << *p.second.begin()
              << " has blocked requests > " << p.first << " sec";
 	}
-	d.detail.push_back(ss.str());
+	warn_detail.push_back(ss.str());
 	if (--left == 0) {
 	  break;
 	}
       }
+      ostringstream ss;
+      ss << warn << " slow requests are blocked > "
+	 << cct->_conf->mon_osd_warn_op_age << " sec. Implicated osds "
+         << implicated_osds;
+      auto& d = checks->add("REQUEST_SLOW", HEALTH_WARN, ss.str());
+      d.detail.swap(warn_detail);
     }
     if (!error_detail.empty()) {
-      ostringstream ss;
-      ss << error << " stuck requests are blocked > "
-	 << err_age << " sec";
-      auto& d = checks->add("REQUEST_STUCK", HEALTH_ERR, ss.str());
-      d.detail.swap(error_detail);
       int left = max;
+      set<int> implicated_osds;
       for (auto& p : error_osd_by_max) {
 	ostringstream ss;
+        implicated_osds.insert(p.second.begin(), p.second.end());
 	if (p.second.size() > 1) {
 	  ss << "osds " << p.second
              << " have stuck requests > " << p.first << " sec";
@@ -3222,11 +3222,16 @@ void PGMap::get_health_checks(
 	  ss << "osd." << *p.second.begin()
              << " has stuck requests > " << p.first << " sec";
 	}
-	d.detail.push_back(ss.str());
+	error_detail.push_back(ss.str());
 	if (--left == 0) {
 	  break;
 	}
       }
+      ostringstream ss;
+      ss << error << " stuck requests are blocked > "
+	 << err_age << " sec. Implicated osds " << implicated_osds;
+      auto& d = checks->add("REQUEST_STUCK", HEALTH_ERR, ss.str());
+      d.detail.swap(error_detail);
     }
   }