mirror of https://github.com/asterisk/asterisk
This allows us to process AO2 statistics for total objects, memory usage, memory overhead and lock usage. * Install refstats.py and reflocks.py into the Asterisk scripts folder. * Enable support for reflocks.py without DEBUG_THREADS. Steal a bit from the ao2 magic to flag when an object lock is used. Remove 'lockobj' from reflocks.py since we can now record 'used' or 'unused' for those objects. Add comments to explain thread safety of the 'struct __priv_data' bitfields. Change-Id: I84e9d679cc86d772cc97c888d9d856a17e0d3a4apull/11/head
parent
5ff42a2bb5
commit
79677ead28
@ -0,0 +1,180 @@
|
||||
#!/usr/bin/env python
|
||||
"""Process a ref debug log for memory usage
|
||||
|
||||
This will provide information about total and peak
|
||||
allocations.
|
||||
|
||||
See http://www.asterisk.org for more information about
|
||||
the Asterisk project. Please do not directly contact
|
||||
any of the maintainers of this project for assistance;
|
||||
the project provides a web site, mailing lists and IRC
|
||||
channels for your use.
|
||||
|
||||
This program is free software, distributed under the terms of
|
||||
the GNU General Public License Version 2. See the LICENSE file
|
||||
at the top of the source tree.
|
||||
|
||||
Copyright (C) 2018, CFWare, LLC
|
||||
Corey Farrell <git@cfware.com>
|
||||
"""
|
||||
|
||||
from __future__ import print_function
|
||||
import sys
|
||||
import os
|
||||
|
||||
from optparse import OptionParser
|
||||
|
||||
|
||||
def create_stats():
|
||||
"""Create statistics object"""
|
||||
return {
|
||||
'count': 0,
|
||||
'overhead': 0,
|
||||
'user_data': 0,
|
||||
'totalmem': 0
|
||||
}
|
||||
|
||||
|
||||
def update_stats(current, peak, total, key, direction, delta):
|
||||
"""Update statistics objects"""
|
||||
|
||||
if direction == 1:
|
||||
total[key] += delta
|
||||
|
||||
delta *= direction
|
||||
current[key] += delta
|
||||
if current[key] > peak[key]:
|
||||
peak[key] = current[key]
|
||||
|
||||
|
||||
def process_file(options):
|
||||
"""The routine that kicks off processing a ref file"""
|
||||
|
||||
current = create_stats()
|
||||
total = create_stats()
|
||||
peak = create_stats()
|
||||
object_types = {}
|
||||
objects = {}
|
||||
filename = options.filepath
|
||||
|
||||
with open(filename, 'r') as ref_file:
|
||||
for line in ref_file:
|
||||
if 'constructor' not in line and 'destructor' not in line:
|
||||
continue
|
||||
# The line format is:
|
||||
# addr,delta,thread_id,file,line,function,state,tag
|
||||
# Only addr, file, line, function, state are used by reflocks.py
|
||||
tokens = line.strip().split(',', 7)
|
||||
addr = tokens[0]
|
||||
state = tokens[6]
|
||||
if 'constructor' in state:
|
||||
split_state = state.split("**")
|
||||
if len(split_state) < 4:
|
||||
print("File does not contain object size information", file=sys.stderr)
|
||||
sys.exit(1)
|
||||
|
||||
obj_type = '%s:%s:%s' % (tokens[3], tokens[4], tokens[5])
|
||||
if obj_type not in object_types:
|
||||
object_types[obj_type] = {
|
||||
'used': 0,
|
||||
'unused': 0,
|
||||
'none': 0
|
||||
}
|
||||
overhead = int(split_state[2])
|
||||
user_data = int(split_state[3])
|
||||
obj = objects[addr] = {
|
||||
'overhead': overhead,
|
||||
'user_data': user_data,
|
||||
'obj_type': obj_type
|
||||
}
|
||||
|
||||
direction = 1
|
||||
else:
|
||||
if addr not in objects:
|
||||
# This error would be reported by refcounter.py.
|
||||
continue
|
||||
obj = objects[addr]
|
||||
del objects[addr]
|
||||
direction = -1
|
||||
obj_type = obj['obj_type']
|
||||
if '**lock-state:unused**' in state:
|
||||
object_types[obj_type]['unused'] += 1
|
||||
elif '**lock-state:used**' in state:
|
||||
object_types[obj_type]['used'] += 1
|
||||
|
||||
# Increment current and peak usage
|
||||
update_stats(current, peak, total, 'count', direction, 1)
|
||||
update_stats(current, peak, total, 'overhead', direction, obj['overhead'])
|
||||
update_stats(current, peak, total, 'user_data', direction, obj['user_data'])
|
||||
update_stats(current, peak, total, 'totalmem', direction, obj['overhead'] + obj['user_data'])
|
||||
|
||||
print("Total usage statistics:")
|
||||
print("%20s: %d" % ("Count", total['count']))
|
||||
print("%20s: %d" % ("Total Memory (k)", total['totalmem'] / 1024))
|
||||
print("%20s: %d (%.2f%%)" % ("Overhead (k)", total['overhead'] / 1024, total['overhead'] * 100.0 / total['totalmem']))
|
||||
print("%20s: %d" % ("User Data (k)", total['user_data'] / 1024))
|
||||
print("")
|
||||
print("Peak usage statistics:")
|
||||
print("%20s: %d" % ("Count", peak['count']))
|
||||
print("%20s: %d" % ("Total Memory (k)", peak['totalmem'] / 1024))
|
||||
print("%20s: %d (%.2f%%)" % ("Overhead (k)", peak['overhead'] / 1024, peak['overhead'] * 100.0 / peak['totalmem']))
|
||||
print("%20s: %d" % ("User Data (k)", peak['user_data'] / 1024))
|
||||
print("")
|
||||
|
||||
lockbyobj = {'used': 0, 'total': 0}
|
||||
lockbytype = {'used': 0, 'total': 0}
|
||||
for (allocator, info) in object_types.items():
|
||||
lockbyobj['used'] += info['used']
|
||||
lockbyobj['total'] += info['used'] + info['unused']
|
||||
|
||||
if info['used'] != 0:
|
||||
lockbytype['used'] += 1
|
||||
elif info['unused'] == 0:
|
||||
# This object type doesn't have locking.
|
||||
continue
|
||||
lockbytype['total'] += 1
|
||||
|
||||
print("Lock usage statistics:")
|
||||
print("%20s: %d of %d used (%.2f%%)" % (
|
||||
"By object",
|
||||
lockbyobj['used'],
|
||||
lockbyobj['total'],
|
||||
lockbyobj['used'] * 100.0 / lockbyobj['total']))
|
||||
print("%20s: %d of %d used (%.2f%%)" % (
|
||||
"By type",
|
||||
lockbytype['used'],
|
||||
lockbytype['total'],
|
||||
lockbytype['used'] * 100.0 / lockbytype['total']))
|
||||
|
||||
|
||||
def main(argv=None):
|
||||
"""Main entry point for the script"""
|
||||
|
||||
ret_code = 0
|
||||
|
||||
if argv is None:
|
||||
argv = sys.argv
|
||||
|
||||
parser = OptionParser()
|
||||
|
||||
parser.add_option("-f", "--file", action="store", type="string",
|
||||
dest="filepath", default="/var/log/asterisk/refs",
|
||||
help="The full path to the refs file to process")
|
||||
|
||||
(options, args) = parser.parse_args(argv)
|
||||
|
||||
if not os.path.isfile(options.filepath):
|
||||
print("File not found: %s" % options.filepath, file=sys.stderr)
|
||||
return -1
|
||||
|
||||
try:
|
||||
process_file(options)
|
||||
except (KeyboardInterrupt, SystemExit, IOError):
|
||||
print("File processing cancelled", file=sys.stderr)
|
||||
return -1
|
||||
|
||||
return ret_code
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
sys.exit(main(sys.argv))
|
Loading…
Reference in new issue