ok. passes tests now for #793

sarracenia/config.py

@@ -360,6 +360,21 @@ def get_log_filename(hostdir, component, configuration, no):
 
     return logdir + os.sep + component + configuration + '_%02d' % no + '.log'
 
+def get_metrics_filename(hostdir, component, configuration, no):
+    """
+      return the name of a single logfile for a single instance.
+   """
+    metricsdir = get_user_cache_dir(hostdir) + os.sep + 'metrics'
+
+    if configuration is None:
+        configuration = ''
+    else:
+        configuration = '_' + configuration
+
+    if configuration[-5:] == '.conf':
+        configuration = configuration[:-5]
+
+    return metricsdir + os.sep + component + configuration + '_%02d' % no + '.json'
 
 def wget_config(urlstr, path, remote_config_url=False):
     logger.debug("wget_config %s %s" % (urlstr, path))
@@ -1829,10 +1844,9 @@ def finalize(self, component=None, config=None):
                 hostdir = self.hostdir
             else:
                 hostdir = None
-            self.pid_filename = get_pid_filename(hostdir, component, cfg,
-                                                 self.no)
+            self.metricsFilename = get_metrics_filename(hostdir, component, cfg, self.no)
+            self.pid_filename = get_pid_filename(hostdir, component, cfg, self.no)
             self.retry_path = self.pid_filename.replace('.pid', '.retry')
-            self.metricsFilename = self.pid_filename.replace('.pid', '.metrics')
             self.novipFilename = self.pid_filename.replace('.pid', '.noVip')
 
 
@@ -2581,10 +2595,24 @@ def cfglogs(cfg_preparse, component, config, logLevel, child_inst):
         else:
             hostdir = None
 
-        logfilename = get_log_filename(
-            hostdir, component, config, child_inst)
+        metricsfilename = get_metrics_filename( hostdir, component, config, child_inst)
+
+        dir_not_there = not os.path.exists(os.path.dirname(metricsfilename))
+        while dir_not_there:
+            try:
+                os.makedirs(os.path.dirname(metricsfilename), exist_ok=True)
+                dir_not_there = False
+            except FileExistsError:
+                dir_not_there = False
+            except Exception as ex:
+                logging.error( "makedirs {} failed err={}".format(os.path.dirname(metricsfilename),ex))
+                logging.debug("Exception details:", exc_info=True)
+                os.sleep(1)
+
+        cfg_preparse.metricsFilename = metricsfilename
+
+        logfilename = get_log_filename( hostdir, component, config, child_inst)
 
-        #print('logfilename= %s' % logfilename )
         dir_not_there = not os.path.exists(os.path.dirname(logfilename))
         while dir_not_there:
             try:

sarracenia/instance.py

@@ -147,10 +147,24 @@ def start(self):
             else:
                 hostdir = None
 
-            logfilename = sarracenia.config.get_log_filename(
-                hostdir, component, config, cfg_preparse.no)
+            metricsfilename = sarracenia.config.get_metrics_filename( hostdir, component, config, cfg_preparse.no)
+
+            dir_not_there = not os.path.exists(os.path.dirname(metricsfilename))
+            while dir_not_there:
+                try:
+                    os.makedirs(os.path.dirname(metricsfilename), exist_ok=True)
+                    dir_not_there = False
+                except FileExistsError:
+                    dir_not_there = False
+                except Exception as ex:
+                    logging.error( "makedirs {} failed err={}".format(os.path.dirname(metricsfilename),ex))
+                    logging.debug("Exception details:", exc_info=True)
+
+            cfg_preparse.metricsFilename = metricsfilename
+
+
+            logfilename = sarracenia.config.get_log_filename( hostdir, component, config, cfg_preparse.no)
 
-            #print('logfilename= %s' % logfilename )
             dir_not_there = not os.path.exists(os.path.dirname(logfilename))
             while dir_not_there:
                 try:
@@ -209,7 +223,6 @@ def start(self):
 
         cfg = sarracenia.config.one_config(component, config)
 
-        cfg.metricsFilename = pidfilename.replace(".pid", ".metrics")
         cfg.novipFilename = pidfilename.replace(".pid", ".noVip")
 
         if not hasattr(cfg, 'env_declared'):

sarracenia/sr.py

@@ -469,7 +469,7 @@ def _read_state_dir(self, dir1):
 
                         for pathname in os.listdir():
                             p = pathlib.Path(pathname)
-                            if p.suffix in ['.pid', '.qname', '.state', '.metrics', '.noVip']:
+                            if p.suffix in ['.pid', '.qname', '.state', '.noVip']:
                                 if sys.version_info[0] > 3 or sys.version_info[
                                         1] > 4:
                                     t = p.read_text().strip()
@@ -501,13 +501,54 @@ def _read_state_dir(self, dir1):
                         os.chdir('..')
                 os.chdir('..')
 
+    def _read_metrics_dir(self,metrics_parent_dir):
+        # read in metrics files
+
+        dir1 = metrics_parent_dir + os.sep + 'metrics'
+
+        if not os.path.isdir(dir1):
+            return
+        os.chdir(dir1)
+
+        for l in os.listdir(dir1):
+
+            if not l.endswith('.json'):
+                continue
+
+            ll = os.path.basename(l.replace('.json','')).split('_')
+            if len(ll) < 3:
+                continue
+
+            c= ll[0]
+            cfg = '_'.join(ll[1:-1])
+            i = int(ll[-1].replace('i',''))
+
+            if (not c in self.components) or (not cfg in self.states[c]):
+                continue
+
+            if not 'instance_metrics' in self.states[c][cfg]:
+                self.states[c][cfg]['instance_metrics'] = {}
+
+            try:
+                p = pathlib.Path(l)
+                with p.open() as f:
+                    t = f.read().strip()
+
+                self.states[c][cfg]['instance_metrics'][i] = json.loads(t)
+                self.states[c][cfg]['instance_metrics'][i]['status'] = { 'mtime':os.stat(p).st_mtime }
+            except:
+                logger.error( f"corrupt metrics file {pathname}: {t}" )
+
+
     def _read_states(self):
         self.states = {}
         for c in self.components:
             self.states[c] = {}
 
         self._read_state_dir(self.user_cache_dir)
         self._read_state_dir(self.user_cache_dir + os.sep + self.hostdir)
+        self._read_metrics_dir(self.user_cache_dir)
+        self._read_metrics_dir(self.user_cache_dir + os.sep + self.hostdir)
 
     def _find_missing_instances_dir(self, dir):
         """ find processes which are no longer running, based on pidfiles in state, and procs.