Merge branch 'PGPRO-4333' into dev

Author: Alexander Popov

mamonsu/lib/plugin.py

@@ -135,6 +135,7 @@ def is_enabled(self):
         return self._enabled
 
     def disable(self):
+        self._plugin_config['enabled'] = 'False'
         self._enabled = False
 
     def set_sender(self, sender):

mamonsu/plugins/pgsql/__init__.py

@@ -6,5 +6,6 @@
 __all__ += ['archive_command']
 __all__ += ['prepared_transaction']
 __all__ += ['relations_size']
+__all__ += ['memory_leak_diagnostic']
 
 from . import *

mamonsu/plugins/pgsql/memory_leak_diagnostic.py

@@ -0,0 +1,170 @@
+from mamonsu.plugins.pgsql.plugin import PgsqlPlugin as Plugin
+import os
+from .pool import Pooler
+import re
+from distutils.version import LooseVersion
+import mamonsu.lib.platform as platform
+
+
+class MemoryLeakDiagnostic(Plugin):
+    DEFAULT_CONFIG = {'enabled': 'False',
+                      'private_anon_mem_threshold': '1GB'}
+    Interval = 60
+
+    query = 'select pid from pg_stat_activity'
+    key_count_diff = 'pgsql.memory_leak_diagnostic.count_diff[]'
+    key_count_diff_error = 'pgsql.memory_leak_diagnostic.msg_text[]'
+    name_count_diff = 'PostgreSQL: number of pids which private anonymous memory exceeds ' \
+                      'private_anon_mem_threshold'
+    name_count_diff_error = 'PostgreSQL: number of pids which private anonymous memory ' \
+                            'exceeds private_anon_mem_threshold, text of message'
+
+    def __init__(self, config):
+        super(Plugin, self).__init__(config)
+        if not platform.LINUX:
+            self.disable()
+            self.log.error('Plugin {name} work only on Linux. '.format(name=self.__class__.__name__))
+
+        if self.is_enabled():
+            self.page_size = os.sysconf('SC_PAGE_SIZE')
+
+            private_anon_mem_threshold_row = self.plugin_config('private_anon_mem_threshold').upper()
+            private_anon_mem_threshold, prefix = re.match(r'([0-9]*)([A-Z]*)',
+                                                          private_anon_mem_threshold_row, re.I).groups()
+            ratio = 0
+
+            if prefix == 'MB':
+                ratio = 1024 * 1024
+            elif prefix == 'GB':
+                ratio = 1024 * 1024 * 1024
+            elif prefix == 'TB':
+                ratio = 1024 * 1024 * 1024 * 1024
+            else:
+                self.disable()
+                self.log.error('Error in config, section [{section}], parameter private_anon_mem_threshold. '
+                               'Possible values MB, GB, TB. For example 1GB.'
+                               .format(section=self.__class__.__name__.lower()))
+
+            self.diff = ratio * int(private_anon_mem_threshold)
+
+            self.os_release = os.uname().release
+            os_release_file = '/etc/os-release'
+            try:
+                release_file = open(os_release_file, 'r').readlines()
+            except Exception as e:
+                self.log.info(f'Cannot read file {os_release_file} : {e}')
+                release_file = None
+
+            if release_file:
+                for line in release_file:
+                    if line.strip('"\n') != '':
+                        k, v = line.split('=', 1)
+                        if k == 'ID':
+                            self.os_name = v.strip('"\n')
+                        elif k == 'VERSION_ID':
+                            self.os_version = v.strip('"\n')
+            else:
+                self.os_name = None
+                self.os_version = None
+
+    def run(self, zbx):
+        pids = []
+        count_diff = 0
+        diffs = []
+        msg_text = ''
+
+        for row in Pooler.query(query=self.query):
+            pids.append(row[0])
+
+        if (LooseVersion(self.os_release) < LooseVersion("4.5")
+            and not (self.os_name == 'centos' and self.os_version == '7'))\
+                or (not self.os_name and not self.os_version):
+            for pid in pids:
+                try:
+                    statm = open(f'/proc/{pid}/statm', 'r').read().split(' ')
+                except FileNotFoundError:
+                    continue
+
+                RES = int(statm[1]) * self.page_size
+                SHR = int(statm[2]) * self.page_size
+                if RES - SHR > self.diff:
+                    count_diff += 1
+                    diffs.append({'pid': pid, 'RES': RES, 'SHR': SHR, 'diff': self.diff})
+            if diffs:
+                for diff in diffs:
+                    msg_text += 'pid: {pid},  RES {RES} - SHR {SHR} more then {diff}\n'.format_map(diff)
+        else:
+            for pid in pids:
+                try:
+                    statm = open(f'/proc/{pid}/status', 'r').readlines()
+                except FileNotFoundError:
+                    continue
+
+                for line in statm:
+                    VmRSS = 0
+                    RssAnon = 0
+                    RssFile = 0
+                    RssShmem = 0
+                    k, v = line.split(':\t', 1)
+
+                    if k == 'VmRSS':
+                        VmRSS = int(v.strip('"\n\t ').split(' ')[0]) * 1024
+                    elif k == 'RssAnon':
+                        RssAnon = int(v.strip('"\n\t ').split(' ')[0]) * 1024
+                    elif k == 'RssFile':
+                        RssFile = int(v.strip('"\n\t ').split(' ')[0]) * 1024
+                    elif k == 'RssShmem':
+                        RssShmem = int(v.strip('"\n\t ').split(' ')[0]) * 1024
+                    if RssAnon > self.diff:
+                        count_diff += 1
+                        diffs.append(
+                            {'pid': pid, 'VmRSS': VmRSS, 'RssAnon': RssAnon, 'RssFile': RssFile, 'RssShmem': RssShmem,
+                             'diff': self.diff})
+            if diffs:
+                for diff in diffs:
+                    msg_text += 'pid: {pid},  RssAnon {RssAnon} more then {diff}, VmRSS {VmRSS}, ' \
+                                       'RssFile {RssFile}, RssShmem {RssShmem} \n'.format_map(diff)
+
+        zbx.send(self.key_count_diff, int(count_diff))
+        zbx.send(self.key_count_diff_error, msg_text)
+
+    def items(self, template):
+        result = template.item(
+            {
+                'name': self.name_count_diff,
+                'key': self.key_count_diff,
+                'delay': self.plugin_config('interval')
+            }
+        )
+        result += template.item(
+            {
+                'name': self.name_count_diff_error,
+                'key': self.key_count_diff_error,
+                'delay': self.plugin_config('interval'),
+                'value_type': Plugin.VALUE_TYPE.text
+            }
+        )
+        return result
+
+    def graphs(self, template):
+        result = template.graph(
+            {
+                'name': self.name_count_diff,
+                'items': [
+                    {
+                        'key': self.key_count_diff,
+                        'color': 'FF0000'
+                    }
+                ]
+            }
+        )
+        return result
+
+    def triggers(self, template):
+        result = template.trigger(
+            {
+                'name': self.name_count_diff + ' on {HOSTNAME}. {ITEM.LASTVALUE}',
+                'expression': '{{#TEMPLATE:{name}.strlen()'
+                              '}}&gt;1'.format(name=self.key_count_diff_error)
+            })
+        return result

packaging/conf/example.conf

@@ -196,3 +196,18 @@ enabled = False
 relations=postgres.pg_catalog.pg_class,postgres.pg_catalog.pg_user
 interval = 300
 
+# This plugin allows detects possible memory leaks while working with PostgreSQL using /proc/pid/status and /proc/pid/statm
+# We use RES and SHR difference to calculate approximate volume of private anonymous backend memory.
+# If it exceeds private_anon_mem_threshold then that pid will be added to a message. An example is presented below
+# statm -  'pid: {pid},  RES {RES} - SHR {SHR} more then {private_anon_mem_threshold}\n'
+# Since Linux 4.5 RssAnon, RssFile and RssShmem have been added.
+# They allows to distinguish types of memory such as private anonymous, file-backed, and shared anonymous memory.
+# We are interested in RssAnon. If its value exceeds private_anon_mem_threshold then that pid will also be added to a message.
+# By default this plugin disabled. To enable this plugin - set bellow "enabled = False"
+# #interval - (onitoring frequency in seconds. 60 seconds by default
+# private_anon_mem_threshold - memory volume threshold after which we need an investigation about memory leak. 1GB by default.
+# Possible values MB, GB, TB. For example 1GB
+[memoryleakdiagnostic]
+enabled = False
+interval = 60
+private_anon_mem_threshold = 1GB