Python Script to Upload Memcache Metrics to CloudWatch

upload_memcache_stats.py

import sys, time, subprocess, socket, telnetlib
from datetime import datetime
from collections import defaultdict

from boto.ec2.cloudwatch import CloudWatchConnection


MAPPINGS = {
    # Memcached name:       (AWS Name, AWS Metric Type, Calculation Method)
    'uptime':               ('Uptime', 'Count', 'gauge'),
    'curr_connections':     ('CurrConnections', 'Count', 'gauge'),
    'cmd_get':              ('CmdGet', 'Count', 'sampling'),
    'cmd_set':              ('CmdSet', 'Count', 'sampling'),
    'cmd_flush':            ('CmdFlush', 'Count', 'sampling'),
    'cmd_touch':            ('CmdTouch', 'Count', 'sampling'),
    'get_hits':             ('GetHits', 'Count', 'sampling'),
    'get_misses':           ('GetMisses', 'Count', 'sampling'),
    'delete_hits':          ('DeleteHits', 'Count', 'sampling'),
    'delete_misses':        ('DeleteMisses', 'Count', 'sampling'),
    'incr_hits':            ('IncrHits', 'Count', 'sampling'),
    'incr_misses':          ('IncrMisses', 'Count', 'sampling'),
    'decr_hits':            ('DecrHits', 'Count', 'sampling'),
    'decr_misses':          ('DecrMisses', 'Count', 'sampling'),
    'cas_hits':             ('CasHits', 'Count', 'sampling'),
    'cas_misses':           ('CasMisses', 'Count', 'sampling'),
    'touch_hits':           ('TouchHits', 'Count', 'sampling'),
    'touch_misses':         ('TouchMisses', 'Count', 'sampling'),
    'bytes_read':           ('BytesReadIntoMemcached', 'Bytes', 'sampling'),
    'bytes_written':        ('BytesWrittenOutFromMemcached', 'Bytes', 'sampling'),
    'evictions':            ('Evictions', 'Count', 'sampling'),
    'bytes':                ('BytesUsedForCacheItems', 'Bytes', 'gauge'),
    'curr_items':           ('CurrItems', 'Count', 'gauge')
}

# Stats not readable from Memcached 'stats' directly,
# we need to fetch or calculate these explicity
DERIVED_STATS = {
    'new_connections':      ('NewConnections', 'Count', 'sampling'),
    'new_items':            ('NewItems', 'Count', 'sampling'),
    'cpu_util':             ('CPUUtilization', 'Percent', 'gauge'),
    'freeable_mem':         ('FreeableMemory', 'Kilobytes', 'gauge'),
    'network_in':           ('NetworkBytesIn', 'Bytes', 'sampling'),
    'network_out':          ('NetworkBytesOut', 'Bytes', 'sampling'),
    'swap_usage':           ('SwapUsage', 'Kilobytes', 'gauge')
}

def client():
    global _client
    if _client is None:
        _client = telnetlib.Telnet('127.0.0.1', '11211')
    return _client

def command(client, cmd):
    'Write a command to telnet and return the response'
    client.write("%s\n" % cmd)
    return client.read_until('END')

def get_stats():
    c = client()
    return command(c, 'stats')

def parse_stats(new_stats):
    # Clear our parsed stats
    parsed_stats = {}
    # Parse the Memcached 'stats' output
    for stat in new_stats.split('\n'):
        if stat.startswith('STAT'):
            stat = stat.split()
            key = stat[1]
            value = stat[2]

            # Filter down to the keys we care about
            if key in MAPPINGS.keys():
                parsed_stats[key] = int(value)

    return parsed_stats

def calculate_regular_stats(current_stats, old_stats):
    calculated_stats = {}

    for key in current_stats.keys():
        # Check to see if we are a measurement where we
        # are interested in the diff between pollings
        if 'sampling' in MAPPINGS[key][2]:
            diff = current_stats[key] - old_stats[key]
            calculated_stats[key] = diff
        # Otherwise this is a gauge and we just care about
        # the value, not the difference
        else:
            calculated_stats[key] = current_stats[key]

    return calculated_stats

def _calculate_new_connections(current_stats, old_stats):
    # Not sure how amazon is calculating this
    pass

def _calculate_new_items(current_stats, old_stats):
    # Not sure how amazon is calculating this
    pass

def _get_network_in(current_stats, old_stats):
    # Not sure how amazon is calculating this
    pass

def _get_network_out(current_stats, old_stats):
    # Not sure how amazon is calculating this
    pass

def _get_cpu_usage():
    # Get the 3rd field of the ps aux command which should be CPU usage
    # of the process as a percentage
    command = "ps aux | grep '/usr/bin/memcached' | grep -v grep | awk '{print $3}'"
    # Create our new process
    proc = subprocess.Popen(command, shell=True, stderr=subprocess.PIPE, stdout=subprocess.PIPE)
    # Filter the result
    result = proc.stdout.readline().rstrip()

    if result is not None:
        return float(result)
    else:
        return 0.0

def _parse_memory():
    mem_info = defaultdict(int)
    # Get the systems memory info
    command = "cat /proc/meminfo"
    # Create our new process
    proc = subprocess.Popen(command, shell=True, stderr=subprocess.PIPE, stdout=subprocess.PIPE)

    # Parse through the lines gathering relevant information
    for line in proc.stdout.readlines():
        for word in ['SwapCached:', 'MemFree:', 'Cached:', 'Buffers:']:
            if line.startswith(word):
                # Pull out the value of the line and convert it to an int
                value = int(line.split()[1])
                if 'swap' in word.lower():
                    mem_info['swap_usage'] = value
                else:
                    mem_info['freeable_mem'] += value
    return mem_info


def calculate_derived_stats():
    calculated_stats = {}
    mem_info = _parse_memory()
    calculated_stats.update(mem_info)
    calculated_stats.update({ 'cpu_util': _get_cpu_usage() })
    return calculated_stats

def compile_stats(old_stats):
    final_stats = {}
    global previous_stats

    new_stats = get_stats()
    current_stats = parse_stats(new_stats)
    previous_stats = current_stats

    final_stats.update(calculate_regular_stats(current_stats, old_stats))
    final_stats.update(calculate_derived_stats())

    return final_stats

def send_stats():
    # Uncomfortable with the global here
    global previous_stats
    stats = compile_stats(previous_stats)
    for key, value in stats.iteritems():
        name = "{0}:{1}".format(hostname, key)
        if key in MAPPINGS.keys():
            unit = MAPPINGS[key][1]
        elif key in DERIVED_STATS.keys():
            unit = DERIVED_STATS[key][1]
        else:
            unit = "count"
        print namespace, name, value, unit
        conn.put_metric_data(namespace=namespace, name=name, value=float(value), unit=unit)

# We want to have these stats set to 0
_client = None
previous_stats = defaultdict(int)
conn = CloudWatchConnection()
namespace = "Memcached"
hostname = socket.gethostname()

while True:
    if datetime.now().second == 0:
        send_stats()
    time.sleep(1)