Merge branch 'telegraf' into 'master'

enable ethtool stats as well

See merge request hpc-team/ansible_cluster_in_a_box!264

roles/telegraf/files/telegraf_mountstats.py

+#!/usr/bin/python
+xprtudpcounters="proto,port,bind_count,rpcsends,rpcreceives,badxids,inflightsends,backlogutil".split(',')
+xprttcpcounters="proto,port,bind_count,connect_count,connect_time,idle_time,rpcsends,rpcreceives,badxids,inflightsends,backlogutil".split(',')
+xprtrdmacounters="proto,port,bind_count,connect_count,idle_time,rpcsends,rpcreceives,badxids,backlogutil,read_chunks,write_chunks,reply_chunks,total_rdma_req',total_rdma_rep,pullup,fixup,hardway,failed_marshal,bad_reply".split(',')
+NfsOpCounters="operations,transmissions,major_timeouts,bytes_sent,bytes_recv,queue_time,response_time,request_time".split(',')
+OPS="WRITE,OPEN,CLOSE,ACCESS,RENAME,SYMLINK,CREATE".split(',')
+
+NfsEventCounters = [
+    'inoderevalidates',
+    'dentryrevalidates',
+    'datainvalidates',
+    'attrinvalidates',
+    'vfsopen',
+    'vfslookup',
+    'vfspermission',
+    'vfsupdatepage',
+    'vfsreadpage',
+    'vfsreadpages',
+    'vfswritepage',
+    'vfswritepages',
+    'vfsreaddir',
+    'vfssetattr',
+    'vfsflush',
+    'vfsfsync',
+    'vfslock',
+    'vfsrelease',
+    'congestionwait',
+    'setattrtrunc',
+    'extendwrite',
+    'sillyrenames',
+    'shortreads',
+    'shortwrites',
+    'delay'
+]
+
+NfsByteCounters = [
+    'normalreadbytes',
+    'normalwritebytes',
+    'directreadbytes',
+    'directwritebytes',
+    'serverreadbytes',
+    'serverwritebytes',
+    'readpages',
+    'writepages'
+]
+
+class DeviceData:
+    """DeviceData objects provide methods for parsing and displaying
+    data for a single mount grabbed from /proc/self/mountstats
+    """
+    def __init__(self):
+        self.__nfs = dict()
+        self.__nfs_device = dict()
+
+    def fstype(self):
+        return self.__nfs_device['fstype']
+
+    def tags(self):
+        return ",".join(["{}={}".format(key,value) for key,value in self.__nfs_device.items()])
+
+    def values(self):
+        try:
+            values = ",".join(["{}={}".format(key,value) for key,value in self.__nfs['bytes']])
+            values +=","
+            values += ",".join(["{}={}".format(key,value) for key,value in self.__nfs['events']])
+        except KeyError as e:
+            # key error occurs if we haven't filtered the lustre mount points from the NFS mount points yet
+            return None
+        return values
+
+    def opvalues(self,op):
+        return ",".join(["{}={}".format(key,value) for key,value in self.__nfs[op]])
+
+
+    def __parse_device_line(self, words):
+        if words[0] == 'device':
+            self.__nfs_device['export'] = words[1]
+            self.__nfs_device['mountpoint'] = words[4]
+            self.__nfs_device['fstype'] = words[7]
+
+    def __parse_bytes_line(self, words):
+        if words[0] == 'bytes:':
+            self.__nfs['bytes'] = zip(NfsByteCounters,[ int(x) for x in words[1:]])
+
+    def __parse_events_line(self,words):
+        if words[0] == 'events:':
+            self.__nfs['events'] = zip(NfsEventCounters,[int(x) for x in words[1:]])
+
+    def __parse_ops_line(self,words):
+        if words[0][:-1] in OPS:
+            self.__nfs[words[0][:-1]] = zip(NfsOpCounters, [ int(x) for x in words[1:]])
+
+    def __parse_xprt_line(self, words):
+        if words[0] == 'xprt:':
+            if words[1] == 'udp':
+                self._rpc = zip(xprtudpcounters, words[1:11])
+            if words[1] == 'tcp':
+                self._rpc = zip(xprttcpcounters, words[1:11])
+            if words[1] == 'rdma':
+                self._rpc = zip(xprtrdmacounters, words[1:11])
+
+
+    def parse_stats(self, lines):
+        """Turn a list of lines from a mount stat file into a 
+        dictionary full of stats, keyed by name
+        """
+        foundnfs = False
+        foundrpc = False
+        for line in lines:
+            words = line.split()
+            if len(words) == 0:
+                continue
+            self.__parse_device_line(words)
+            self.__parse_bytes_line(words)
+            self.__parse_events_line(words)
+            self.__parse_ops_line(words)
+            self.__parse_xprt_line(words)
+
+
+def parse_stats_file(filename):
+    """pop the contents of a mountstats file into a dictionary,
+    keyed by mount point.  each value object is a list of the
+    lines in the mountstats file corresponding to the mount
+    point named in the key.
+    """
+    ms_dict = dict()
+    key = ''
+
+    f = open(filename)
+    for line in f.readlines():
+        words = line.split()
+        if len(words) == 0:
+            continue
+        if line.startswith("no device mounted") :
+            continue
+        if words[0] == 'device':
+            key = words[4]
+            new = [ line.strip() ]
+        elif 'nfs' in words or 'nfs4' in words:
+            key = words[3]
+            new = [ line.strip() ]
+        else:
+            new += [ line.strip() ]
+        ms_dict[key] = new
+    f.close
+
+    return ms_dict
+
+
+def iostats(mountstats):
+    stats = {}
+    for device in mountstats:
+        stats[device] = DeviceData()
+        stats[device].parse_stats(mountstats[device])
+    return stats
+
+def print_influx_line_proto(device,stats):
+
+    try:
+        if not 'nfs' in stats.fstype():
+            return
+        print("mountstats,{} {}".format(stats.tags(), stats.values()))
+        for op in OPS:
+            print("nfsops,{},op={} {}".format(stats.tags(),op,stats.opvalues(op)))
+    except:
+        return
+
+mountstats = parse_stats_file("/proc/self/mountstats")
+stats = iostats(mountstats)
+for device in stats:
+    print_influx_line_proto(device,stats[device])
+

roles/telegraf/tasks/main.yml

@@ -22,6 +22,25 @@
   become: true
   become_user: root
 
+- name: Make a directory for extra files
+  file: 
+    state: directory
+    mode: 'u=rwx,g=rx,o=rx'
+    owner: 'telegraf'
+    group: 'telegraf'
+    path: '/opt/telegraf/bin'
+  become: true
+  become_user: root
+
+- name: copy mountstats plugin
+  copy:
+    mode: 'u=rwx,g=rx,o=rx'
+    src: telegraf_mountstats.py
+    dest: '/opt/telegraf/bin/telegraf_mountstats.py'
+  become: true
+  become_user: root
+
+
 - name: Install Telegraf config
   template:
     src: telegraf.conf.j2

roles/telegraf/templates/telegraf.conf.j2

@@ -115,6 +115,15 @@
 [[inputs.netstat]]
   # no configuration
 
+[[inputs.exec]]
+  commands = [
+    "/opt/telegraf/bin/telegraf_mountstats.py"
+  ]
+  data_format = "influx"
+  timeout="4s"
+  interval="300s"
+
+
 ###############################################################################
 #                              SERVICE INPUTS                                 #
 ###############################################################################