[ceph_mon] Replace an obsolete command

mon_status command is no longer available since Octopus
release. Replaced it with its ceph tell equivalent that
works in both older as well as newer Ceph releases.

[ceph_osd/mgr] Improve the method to locate the daemon's IDs.

Signed-off-by: Ponnuvel Palaniyappan <pponnuvel@gmail.com>

3 files changed, 34 insertions, 9 deletions

diff --git a/sos/report/plugins/ceph_mgr.py b/sos/report/plugins/ceph_mgr.py
index 8ee60687..ff0c50cc 100644
--- a/sos/report/plugins/ceph_mgr.py
+++ b/sos/report/plugins/ceph_mgr.py
@@ -79,10 +79,13 @@ class CephMGR(Plugin, RedHatPlugin, UbuntuPlugin):
             # Extract the OSD ids from valid output lines
             for procs in out['output'].splitlines():
                 proc = procs.split()
-                if len(proc) < 6:
-                    continue
-                if proc[4] == '--id' and "ceph-mgr" in proc[0]:
-                    mgr_ids.append("mgr.%s" % proc[5])
+                # Locate the '--id' value
+                if proc and proc[0].endswith("ceph-mgr"):
+                    try:
+                        id_index = proc.index("--id")
+                        mgr_ids.append("mgr.%s" % proc[id_index+1])
+                    except (IndexError, ValueError):
+                        self.log_warn("could not find ceph-mgr id: %s", procs)
 
         # If containerized, run commands in containers
         try:
diff --git a/sos/report/plugins/ceph_mon.py b/sos/report/plugins/ceph_mon.py
index 142c0822..4ab5c6ba 100644
--- a/sos/report/plugins/ceph_mon.py
+++ b/sos/report/plugins/ceph_mon.py
@@ -49,7 +49,6 @@ class CephMON(Plugin, RedHatPlugin, UbuntuPlugin):
             "ceph config log",
             "ceph config generate-minimal-conf",
             "ceph config-key dump",
-            "ceph mon_status",
             "ceph osd metadata",
             "ceph osd erasure-code-profile ls",
             "ceph osd crush show-tunables",
@@ -90,6 +89,26 @@ class CephMON(Plugin, RedHatPlugin, UbuntuPlugin):
             "ceph %s --format json-pretty" % s for s in ceph_cmds
         ], subdir="json_output", tags="insights_ceph_health_detail")
 
+        mon_ids = []
+        # Get the ceph user processes
+        out = self.exec_cmd('ps -u ceph -o args')
+
+        if out['status'] == 0:
+            # Extract the mon ids
+            for procs in out['output'].splitlines():
+                proc = procs.split()
+                # Locate the '--id' value of ceph-mon
+                if proc and proc[0].endswith("ceph-mon"):
+                    try:
+                        id_index = proc.index("--id")
+                        mon_ids.append(proc[id_index+1])
+                    except (IndexError, ValueError):
+                        self.log_warn("could not find ceph-mon id: %s", procs)
+
+        self.add_cmd_output([
+            "ceph tell mon.%s mon_status" % mon_id for mon_id in mon_ids
+        ], subdir="json_output", tags="insights_ceph_health_detail")
+
         # these can be cleaned up too but leaving them for safety for now
         self.add_forbidden_path([
             "/etc/ceph/*keyring*",
diff --git a/sos/report/plugins/ceph_osd.py b/sos/report/plugins/ceph_osd.py
index 52cf52df..dc3d3af1 100644
--- a/sos/report/plugins/ceph_osd.py
+++ b/sos/report/plugins/ceph_osd.py
@@ -73,10 +73,13 @@ class CephOSD(Plugin, RedHatPlugin, UbuntuPlugin):
             # Extract the OSD ids from valid output lines
             for procs in out['output'].splitlines():
                 proc = procs.split()
-                if len(proc) < 6:
-                    continue
-                if proc[4] == '--id' and proc[5].isdigit():
-                    osd_ids.append("osd.%s" % proc[5])
+                # Locate the '--id' value
+                if proc and proc[0].endswith("ceph-osd"):
+                    try:
+                        id_index = proc.index("--id")
+                        osd_ids.append("osd.%s" % proc[id_index+1])
+                    except (IndexError, ValueError):
+                        self.log_warn("could not find ceph-osd id: %s", procs)
 
         try:
             cname = self.get_all_containers_by_regex("ceph-osd*")[0][1]