Modified path to qemu package source to /sf/adapt/qemu
This is to conform with the convention that adaptation-level packages
should be located in /sf/adapt rather than /sf/adaptation.
This is an interim workaround for bug 3744.
#! /usr/bin/env python
# Extract statistics from ftp daemon log.
# Usage:
# ftpstats [-m maxitems] [-s search] [file]
# -m maxitems: restrict number of items in "top-N" lists, default 25.
# -s string: restrict statistics to lines containing this string.
# Default file is /usr/adm/ftpd; a "-" means read standard input.
# The script must be run on the host where the ftp daemon runs.
# (At CWI this is currently buizerd.)
import os
import sys
import re
import string
import getopt
pat = '^([a-zA-Z0-9 :]*)!(.*)!(.*)!([<>].*)!([0-9]+)!([0-9]+)$'
prog = re.compile(pat)
def main():
maxitems = 25
search = None
try:
opts, args = getopt.getopt(sys.argv[1:], 'm:s:')
except getopt.error, msg:
print msg
print 'usage: ftpstats [-m maxitems] [file]'
sys.exit(2)
for o, a in opts:
if o == '-m':
maxitems = string.atoi(a)
if o == '-s':
search = a
file = '/usr/adm/ftpd'
if args: file = args[0]
if file == '-':
f = sys.stdin
else:
try:
f = open(file, 'r')
except IOError, msg:
print file, ':', msg
sys.exit(1)
bydate = {}
bytime = {}
byfile = {}
bydir = {}
byhost = {}
byuser = {}
bytype = {}
lineno = 0
try:
while 1:
line = f.readline()
if not line: break
lineno = lineno + 1
if search and string.find(line, search) < 0:
continue
if prog.match(line) < 0:
print 'Bad line', lineno, ':', repr(line)
continue
items = prog.group(1, 2, 3, 4, 5, 6)
(logtime, loguser, loghost, logfile, logbytes,
logxxx2) = items
## print logtime
## print '-->', loguser
## print '--> -->', loghost
## print '--> --> -->', logfile
## print '--> --> --> -->', logbytes
## print '--> --> --> --> -->', logxxx2
## for i in logtime, loghost, logbytes, logxxx2:
## if '!' in i: print '???', i
add(bydate, logtime[-4:] + ' ' + logtime[:6], items)
add(bytime, logtime[7:9] + ':00-59', items)
direction, logfile = logfile[0], logfile[1:]
# The real path probably starts at the last //...
while 1:
i = string.find(logfile, '//')
if i < 0: break
logfile = logfile[i+1:]
add(byfile, logfile + ' ' + direction, items)
logdir = os.path.dirname(logfile)
## logdir = os.path.normpath(logdir) + '/.'
while 1:
add(bydir, logdir + ' ' + direction, items)
dirhead = os.path.dirname(logdir)
if dirhead == logdir: break
logdir = dirhead
add(byhost, loghost, items)
add(byuser, loguser, items)
add(bytype, direction, items)
except KeyboardInterrupt:
print 'Interrupted at line', lineno
show(bytype, 'by transfer direction', maxitems)
show(bydir, 'by directory', maxitems)
show(byfile, 'by file', maxitems)
show(byhost, 'by host', maxitems)
show(byuser, 'by user', maxitems)
showbar(bydate, 'by date')
showbar(bytime, 'by time of day')
def showbar(dict, title):
n = len(title)
print '='*((70-n)//2), title, '='*((71-n)//2)
list = []
keys = dict.keys()
keys.sort()
for key in keys:
n = len(str(key))
list.append((len(dict[key]), key))
maxkeylength = 0
maxcount = 0
for count, key in list:
maxkeylength = max(maxkeylength, len(key))
maxcount = max(maxcount, count)
maxbarlength = 72 - maxkeylength - 7
for count, key in list:
barlength = int(round(maxbarlength*float(count)/maxcount))
bar = '*'*barlength
print '%5d %-*s %s' % (count, maxkeylength, key, bar)
def show(dict, title, maxitems):
if len(dict) > maxitems:
title = title + ' (first %d)'%maxitems
n = len(title)
print '='*((70-n)//2), title, '='*((71-n)//2)
list = []
keys = dict.keys()
for key in keys:
list.append((-len(dict[key]), key))
list.sort()
for count, key in list[:maxitems]:
print '%5d %s' % (-count, key)
def add(dict, key, item):
if dict.has_key(key):
dict[key].append(item)
else:
dict[key] = [item]
if __name__ == "__main__":
main()