mgr/cephadm: Get host facts data in host cache

Include cephadm gather-facts information in host cache

Signed-off-by: Juan Miguel Olmo Martínez <[email protected]>

src/cephadm/cephadm

@@ -5436,7 +5436,6 @@ class HostFacts():
         up_secs, _ = raw_time.split()
         return float(up_secs)
 
-    @property
     def kernel_security(self):
         # type: () -> Dict[str, str]
         """Determine the security features enabled in the kernel - SELinux, AppArmor"""
@@ -5501,6 +5500,23 @@ class HostFacts():
             "description": "Linux Security Module framework is not available"
         }
 
+    @property
+    def kernel_parameters(self):
+        # type: () -> Dict[str, str]
+        """Get kernel parameters required/used in Ceph clusters"""
+
+        k_param = {}
+        out, _, _ = call_throws(['sysctl', '-a'])
+        if out:
+            param_list = out.split('\n')
+            param_dict = { param.split(" = ")[0]:param.split(" = ")[-1] for param in param_list}
+
+            # return only desired parameters
+            if 'net.ipv4.ip_nonlocal_bind' in param_dict:
+                k_param['net.ipv4.ip_nonlocal_bind'] = param_dict['net.ipv4.ip_nonlocal_bind']
+
+        return k_param
+
     def dump(self):
         # type: () -> str
         """Return the attributes of this HostFacts object as json"""

src/pybind/mgr/cephadm/inventory.py

@@ -204,6 +204,8 @@ def __init__(self, mgr):
         self.daemons = {}   # type: Dict[str, Dict[str, orchestrator.DaemonDescription]]
         self.last_daemon_update = {}   # type: Dict[str, datetime.datetime]
         self.devices = {}              # type: Dict[str, List[inventory.Device]]
+        self.facts = {}                # type: Dict[str, Dict[str, Any]]
+        self.last_facts_update = {}    # type: Dict[str, datetime.datetime]
         self.osdspec_previews = {}     # type: Dict[str, List[Dict[str, Any]]]
         self.networks = {}             # type: Dict[str, Dict[str, List[str]]]
         self.last_device_update = {}   # type: Dict[str, datetime.datetime]
@@ -278,6 +280,11 @@ def update_host_daemons(self, host, dm):
         self.daemons[host] = dm
         self.last_daemon_update[host] = datetime.datetime.utcnow()
 
+    def update_host_facts(self, host, facts):
+        # type: (str, Dict[str, Dict[str, Any]]) -> None
+        self.facts[host] = facts
+        self.last_facts_update[host] = datetime.datetime.utcnow()
+
     def update_host_devices_networks(self, host, dls, nets):
         # type: (str, List[inventory.Device], Dict[str,List[str]]) -> None
         self.devices[host] = dls
@@ -366,6 +373,10 @@ def rm_host(self, host):
             del self.daemons[host]
         if host in self.devices:
             del self.devices[host]
+        if host in self.facts:
+            del self.facts[host]
+        if host in self.last_facts_update:
+            del self.last_facts_update[host]
         if host in self.osdspec_previews:
             del self.osdspec_previews[host]
         if host in self.loading_osdspec_preview:
@@ -463,6 +474,17 @@ def host_needs_daemon_refresh(self, host):
             return True
         return False
 
+    def host_needs_facts_refresh(self, host):
+        # type: (str) -> bool
+        if host in self.mgr.offline_hosts:
+            logger.debug(f'Host "{host}" marked as offline. Skipping gather facts refresh')
+            return False
+        cutoff = datetime.datetime.utcnow() - datetime.timedelta(
+            seconds=self.mgr.facts_cache_timeout)
+        if host not in self.last_facts_update or self.last_facts_update[host] < cutoff:
+            return True
+        return False
+
     def host_had_daemon_refresh(self, host: str) -> bool:
         """
         ... at least once.

src/pybind/mgr/cephadm/module.py

@@ -142,6 +142,12 @@ class CephadmOrchestrator(orchestrator.Orchestrator, MgrModule,
             'default': 10 * 60,
             'desc': 'seconds to cache service (daemon) inventory',
         },
+        {
+            'name': 'facts_cache_timeout',
+            'type': 'secs',
+            'default': 1 * 60,
+            'desc': 'seconds to cache host facts data',
+        },
         {
             'name': 'host_check_interval',
             'type': 'secs',
@@ -293,6 +299,7 @@ def __init__(self, *args, **kwargs):
             self.ssh_config_file = None  # type: Optional[str]
             self.device_cache_timeout = 0
             self.daemon_cache_timeout = 0
+            self.facts_cache_timeout = 0
             self.host_check_interval = 0
             self.mode = ''
             self.container_image_base = ''

src/pybind/mgr/cephadm/serve.py

@@ -110,6 +110,7 @@ def _refresh_hosts_and_daemons(self) -> None:
 
         @forall_hosts
         def refresh(host):
+
             if self.mgr.cache.host_needs_check(host):
                 r = self._check_host(host)
                 if r is not None:
@@ -133,6 +134,12 @@ def refresh(host):
                 if r:
                     failures.append(r)
 
+            if self.mgr.cache.host_needs_facts_refresh(host):
+                self.log.info(('refreshing %s facts' % host))
+                r = self._refresh_facts(host)
+                if r:
+                    failures.append(r)
+
             if self.mgr.cache.host_needs_osdspec_preview_refresh(host):
                 self.log.debug(f"refreshing OSDSpec previews for {host}")
                 r = self._refresh_host_osdspec_previews(host)
@@ -246,6 +253,20 @@ def _refresh_host_daemons(self, host) -> Optional[str]:
         self.mgr.cache.save_host(host)
         return None
 
+    def _refresh_facts(self, host):
+        try:
+            out, err, code = self.mgr._run_cephadm(
+                host, cephadmNoImage, 'gather-facts', [],
+                error_ok=True, no_fsid=True)
+
+            if code:
+                return 'host %s gather-facts returned %d: %s' % (
+                    host, code, err)
+        except Exception as e:
+            return 'host %s gather facts failed: %s' % (host, e)
+        self.log.debug('Refreshed host %s facts' % (host))
+        self.mgr.cache.update_host_facts(host, json.loads(''.join(out)))
+
     def _refresh_host_devices(self, host) -> Optional[str]:
         try:
             out, err, code = self.mgr._run_cephadm(