tahoe-lafs/misc/operations_helpers/munin/tahoe_overhead

68 lines
2.9 KiB
Python

#!/usr/bin/env python
# This is a munin plugin which pulls total-used data from the server in
# misc/operations_helpers/spacetime/diskwatcher.tac, and a total-deep-size number from custom
# PHP database-querying scripts on a different server. It produces a graph of
# how much garbage/overhead is present in the grid: the ratio of total-used
# over (total-deep-size*N/k), expressed as a percentage. No overhead would be
# 0, using twice as much space as we'd prefer would be 100. This is the
# percentage which could be saved if we made GC work perfectly and reduced
# other forms of overhead to zero. This script assumes 3-of-10.
# A second graph is produced with how much of the total-deep-size number
# would be saved if we removed data from inactive accounts. This is also on a
# percentage scale.
# A separate number (without a graph) is produced with the "effective
# expansion factor". If there were no overhead, with 3-of-10, this would be
# 3.33 .
# Overhead is caused by the following problems (in order of size):
# uncollected garbage: files that are no longer referenced but not yet deleted
# inactive accounts: files that are referenced by cancelled accounts
# share storage overhead: bucket directories
# filesystem overhead: 4kB minimum block sizes
# share overhead: hashes, pubkeys, lease information
# This plugin should be configured with env_diskwatcher_url= pointing at the
# diskwatcher.tac webport, and env_deepsize_url= pointing at the PHP script.
from __future__ import print_function
import os, sys, urllib, json
if len(sys.argv) > 1 and sys.argv[1] == "config":
print("""\
graph_title Tahoe Overhead Calculator
graph_vlabel Percentage
graph_category tahoe
graph_info This graph shows the estimated amount of storage overhead (ratio of actual disk usage to ideal disk usage). The 'overhead' number is how much space we could save if we implemented GC, and the 'inactive' number is how much additional space we could save if we could delete data for cancelled accounts.
overhead.label disk usage overhead
overhead.draw LINE2
inactive.label inactive account usage
inactive.draw LINE1
effective_expansion.label Effective Expansion Factor
effective_expansion.graph no""")
sys.exit(0)
diskwatcher_url = os.environ["diskwatcher_url"]
total = json.load(urllib.urlopen(diskwatcher_url))["used"]
deepsize_url = os.environ["deepsize_url"]
deepsize = json.load(urllib.urlopen(deepsize_url))
k = 3; N = 10
expansion = float(N) / k
ideal = expansion * deepsize["all"]
overhead = (total - ideal) / ideal
if overhead > 0:
# until all the storage-servers come online, this number will be nonsense
print("overhead.value %f" % (100.0 * overhead))
# same for this one
effective_expansion = total / deepsize["all"]
print("effective_expansion.value %f" % effective_expansion)
# this value remains valid, though
inactive_savings = (deepsize["all"] - deepsize["active"]) / deepsize["active"]
print("inactive.value %f" % (100.0 * inactive_savings))