source: trunk/misc/operations_helpers/munin/tahoe_nodememory

#!/usr/bin/env python

# This munin plugin isolates processes by looking for the 'pid' file created
# by 'allmydata start', then extracts the amount of memory they consume (both
# VmSize and VmRSS) from /proc


import os, sys, re

# for testing
# os.environ["nodememory_warner1"] = "run/warner1"
# os.environ["nodememory_warner2"] = "run/warner2"

nodedirs = []
for k,v in os.environ.items():
    if k.startswith("nodememory_"):
        nodename = k[len("nodememory_"):]
        nodedirs.append((nodename, v))
nodedirs.sort(lambda a,b: cmp(a[0],b[0]))

pids = {}

for node,nodedir in nodedirs:
    pidfile = os.path.join(nodedir, "twistd.pid")
    if os.path.exists(pidfile):
        pid = int(open(pidfile,"r").read())
        pids[node] = pid

fields = ["VmSize", "VmRSS"]


if len(sys.argv) > 1:
    if sys.argv[1] == "config":
        configinfo = \
        """graph_title Memory Consumed by Nodes
graph_vlabel bytes
graph_category Tahoe
graph_info This graph shows the memory used by specific processes
"""
        for nodename,nodedir in nodedirs:
            for f in fields:
                configinfo += "%s_%s.label %s used by %s\n" % (nodename, f,
                                                               f, nodename)
                linetype = "LINE1"
                if f == "VmSize":
                    linetype = "LINE2"
                configinfo += "%s_%s.draw %s\n" % (nodename, f, linetype)
                if f == "VmData":
                    configinfo += "%s_%s.graph no\n" % (nodename, f)

        print(configinfo)
        sys.exit(0)

nodestats = {}
for node,pid in pids.items():
    stats = {}
    statusfile = "/proc/%s/status" % pid
    if not os.path.exists(statusfile):
        continue
    for line in open(statusfile,"r").readlines():
        for f in fields:
            if line.startswith(f + ":"):
                m = re.search(r'(\d+)', line)
                stats[f] = int(m.group(1))
    nodestats[node] = stats

for node,stats in nodestats.items():
    for f,value in stats.items():
        # TODO: not sure if /proc/%d/status means 1000 or 1024 when it says
        # 'kB'
        print("%s_%s.value %d" % (node, f, 1024*value))