Improve systemd services recognition

AppController/djinn.rb

@@ -3431,8 +3431,8 @@ def start_search2_role
   def stop_search2_role
     # Stop Solr
     Djinn.log_debug('Stopping SOLR on this node.')
-    Djinn.log_run('systemctl stop solr')
-    Djinn.log_run('systemctl disable solr')
+    Djinn.log_run('systemctl stop appscale-solr')
+    Djinn.log_run('systemctl disable appscale-solr')
     Djinn.log_debug('Done stopping SOLR.')
   end
 

Hermes/appscale/hermes/resources/process.py

@@ -76,12 +76,6 @@ class Process(object):
   sample_time_diff = attr.ib(default=None)
 
 
-MONIT_PROCESS_PATTERN = re.compile(
-  r"^Process \'(?P<name>[^']+)\' *\n"
-  r"(^  .*\n)*?"
-  r"^  pid +(?P<pid>\d+)\n",
-  re.MULTILINE
-)
 PROCESS_ATTRS = (
   'pid', 'ppid', 'name', 'cwd', 'exe', 'cmdline', 'status', 'username',
   'cpu_times', 'cpu_percent', 'memory_info', 'io_counters', 'num_threads',
@@ -193,42 +187,50 @@ async def get_known_processes():
   """
   known_processes = {}
 
+  dependencies_cmd = ('cat /lib/systemd/system/appscale-*.target '
+                      '| grep -E "^After=.*\.service$" | cut -d "=" -f 2')
   try:
-    # Detect processes supervised by monit
-    output, error = await helper.subprocess('monit status', timeout=5)
+    # Detect appscale dependency services
+    output, error = await helper.subprocess(dependencies_cmd, timeout=5)
+    services = output.strip().split('\n')
   except SubprocessError as err:
-    logger.warning('Failed to run `monit status` ({})'.format(err))
-    output = ''
-  for match in MONIT_PROCESS_PATTERN.finditer(output):
-    monit_name = match.group('name')
-    pid = int(match.group('pid'))
-    service = find_service_by_monit_name(monit_name)
-    application_id = service.get_application_id_by_monit_name(monit_name)
-    tags = [APPSCALE_PROCESS_TAG, service.name, monit_name]
-    if application_id:
-      tags.append('app___{}'.format(application_id))
-    known_processes[pid] = tags
-
-  # Detect processes supervised by AppScale ServiceManager
-  for server in ServiceManager.get_state():
-    known_processes[server.process.pid] = [APPSCALE_PROCESS_TAG, server.type]
+    logger.warning('Failed to detect appscale dependency services '
+                   'by running `{}` ({})'.format(dependencies_cmd, err))
+    services = []
+
 
+  services_cmd = ('systemctl --no-legend list-units "appscale-*.service" '
+                  '| cut -d " " -f 1')
   try:
-    # Detect processes supervised by systemd
-    output, error = await helper.subprocess(
-      'systemctl status solr.service '
-      '| grep \'Main PID\' | awk \'{ print $3 }\'',
-      timeout=5
-    )
+    # Detect appscale own services
+    output, error = await helper.subprocess(services_cmd, timeout=5)
+    services += output.strip().split('\n')
   except SubprocessError as err:
-    logger.warning('Failed to run `systemctl status solr.service` ({})'
+    logger.warning('Failed to detect appscale own services '
+                   'by running `{}` ({})'.format(services_cmd, err))
+
+
+  for service in services:
+    try:
+      status_cmd = (
+        'systemctl status {} | grep "Main PID" | awk "{{ print $3 }}"'
+        .format(service)
+      )
+      output, error = await helper.subprocess(status_cmd, timeout=5)
+      if output.isdigit() and output != '0':
+        pid = int(output)
+        process_tags = [APPSCALE_PROCESS_TAG]
+        # Sample service names are:
+        # appscale-instance-run@testapp_default_v1_1570022208920-20000.service
+        # appscale-memcached.service
+        before_at, after_at = service[len('appscale-'):-len('.service')].split('@')
+        process_tags.append(before_at)
+        for part in after_at.split('_'):
+          process_tags.append('_{}'.format(part))
+        known_processes[pid] = process_tags
+    except SubprocessError as err:
+      logger.warning('Failed to run `systemctl status solr.service` ({})'
                    .format(err))
-    output = ''
-  if output.isdigit() and output != '0':
-    solr_pid = int(output)
-    known_processes[solr_pid] = [
-      APPSCALE_PROCESS_TAG, 'solr'
-    ]
   return known_processes
 
 

SearchService2/solr-management/ensure_solr_running.sh

@@ -63,24 +63,24 @@ export PRIVATE_IP
 envsubst '$SOLR_HEAP $ZK_HOST $PRIVATE_IP' \
  < "${SOLR_MANAGEMENT_DIR}/solr.in.sh" > "/tmp/solr.in.sh"
 envsubst '$MEMORY_LOW $MEMORY_HIGH $MEMORY_MAX'\
- < "${SOLR_MANAGEMENT_DIR}/solr.service" > "/tmp/solr.service"
+ < "${SOLR_MANAGEMENT_DIR}/appscale-solr.service" > "/tmp/appscale-solr.service"
 if cmp -s "/tmp/solr.in.sh" "/etc/default/solr.in.sh" \
-&& cmp -s "/tmp/solr.service" "/etc/systemd/system/solr.service"
+&& cmp -s "/tmp/appscale-solr.service" "/etc/systemd/system/appscale-solr.service"
 then
     echo "/etc/default/solr.in.sh has no changes."
-    echo "/etc/systemd/system/solr.service has no changes."
+    echo "/etc/systemd/system/appscale-solr.service has no changes."
     echo "Making sure Solr is running."
-    sudo systemctl enable solr
-    sudo systemctl start solr
+    sudo systemctl enable appscale-solr
+    sudo systemctl start appscale-solr
 else
     echo "Copying new solr.in.sh to /etc/default/solr.in.sh"
     sudo cp "/tmp/solr.in.sh" "/etc/default/solr.in.sh"
-    echo "Copying new solr.service to /etc/systemd/system/solr.service"
-    sudo cp "/tmp/solr.service" "/etc/systemd/system/solr.service"
+    echo "Copying new solr.service to /etc/systemd/system/appscale-solr.service"
+    sudo cp "/tmp/appscale-solr.service" "/etc/systemd/system/appscale-solr.service"
     echo "Making sure Solr is restarted."
     sudo systemctl daemon-reload
-    sudo systemctl enable solr
-    sudo systemctl restart solr
+    sudo systemctl enable appscale-solr
+    sudo systemctl restart appscale-solr
 fi
 
 echo "Making sure appscale-specific config set is uploaded to zookeeper."

SearchService2/solr-management/solr.service

@@ -1,4 +1,4 @@
-# /etc/systemd/system/solr.service
+# /etc/systemd/system/appscale-solr.service
 [Unit]
 Description=Apache SOLR
 After=syslog.target network.target remote-fs.target nss-lookup.target