| 1 |
#! /usr/bin/env python |
| 2 |
# $Id$ |
| 3 |
# Source: $URL$ |
| 4 |
# J. A. Templon, NIKHEF/PDP 2011 |
| 5 |
|
| 6 |
import optparse |
| 7 |
|
| 8 |
p = optparse.OptionParser(description="Program to make rrdtool plots " + \ |
| 9 |
"of job running jobs by unix group") |
| 10 |
|
| 11 |
# p.add_option("-r",action="store",dest="minsize",default='0',help="minimum size of dirs considered; can use suffixes k,M,G for multiples of 1000**{1,2,3} bytes") |
| 12 |
# p.add_option("--qdel",action="store_true",dest="deljobs",help="delete jobs for which TMPDIR is larger than MINSIZE",default=False) |
| 13 |
|
| 14 |
p.add_option("--rank-only",action="store_true",dest="rankonly", |
| 15 |
help="don't plot, just print ranking of groups",default=False) |
| 16 |
|
| 17 |
debug = 0 |
| 18 |
|
| 19 |
opts, args = p.parse_args() |
| 20 |
|
| 21 |
import os |
| 22 |
|
| 23 |
NUMGROUPS=8 |
| 24 |
DATADIR=os.environ['HOME'] + '/ndpfdata/' |
| 25 |
PLOTDIR=os.environ['HOME'] + '/public_html/' |
| 26 |
# 8 class qualitative paired color scheme |
| 27 |
|
| 28 |
colors = [ "#A6CEE3", "#1F77B4", "#B2DF8A", "#33A02C", |
| 29 |
"#FB9A99", "#E31A1C", "#FDBF6F", "#FF7F00" ] |
| 30 |
|
| 31 |
colors.reverse() |
| 32 |
|
| 33 |
# for reference : ranges of the RRAs |
| 34 |
# base step size 60 sec |
| 35 |
# step 1 : 60 sec x 1600 points : 1600 min : 26,67 hr : 1.11 days |
| 36 |
# step 2 : 120 sec x 1200 points : 2400 min : 40 hr : 1,67 days |
| 37 |
# step 10 : 600 sec x 1800 points : 18 000 min : 300 hr : 12,5 days |
| 38 |
# step 30 : 1800 sec x 2500 points : 75 000 min : 1250 hr : 52.08 days |
| 39 |
# step 120 : 7200 sec x 1000 points : 120 000 min : 2000 hr : 83.33 days |
| 40 |
# step 480 : 28800 sec x 1000 points : 480 000 min : 8000 hr : 333.33 days |
| 41 |
# step 1440 : 86400 sec x 3650 points : 3650 days : 10 years |
| 42 |
|
| 43 |
# resolutions of RRAs |
| 44 |
|
| 45 |
# 1 : 60 sec : 1 min |
| 46 |
# 2 : 120 sec : 2 min |
| 47 |
# 3 : 600 sec : 10 min |
| 48 |
# 4 : 1 800 sec : 30 min |
| 49 |
# 5 : 7 200 sec : 120 min : 2 hr |
| 50 |
# 6 : 28 800 sec : 480 min : 8 hr |
| 51 |
# 7 : 86 400 sec : 1440 min : 24 hr : 1 day |
| 52 |
|
| 53 |
plotrangedef = { |
| 54 |
'hr' : { 'timeargs' : [ '-s', 'n-200min', '-e', 'n' ], |
| 55 |
'timetag' : 'hr', |
| 56 |
'avrange' : 24*3600, |
| 57 |
'avres' : 60, |
| 58 |
'sizeargs' : { 'small' : [ '--width', '200', '--height', '125', |
| 59 |
'--x-grid', |
| 60 |
'MINUTE:20:HOUR:1:HOUR:1:0:%H:00' |
| 61 |
], |
| 62 |
'large' : [ '--width', '800', '--height', '500' ] |
| 63 |
}, |
| 64 |
}, |
| 65 |
'day' : { 'timeargs' : [ '-s', 'n-2000min', '-e', 'n' ], |
| 66 |
'timetag' : 'day', |
| 67 |
'avrange' : 24*3600, |
| 68 |
'avres' : 60, |
| 69 |
'sizeargs' : { 'small' : [ '--width', '200', '--height', '125', |
| 70 |
'--x-grid', |
| 71 |
'HOUR:6:DAY:1:HOUR:12:0:%a %H:00' |
| 72 |
], |
| 73 |
'large' : [ '--width', '1000', '--height', '625' ] |
| 74 |
}, |
| 75 |
}, |
| 76 |
'week' : { 'timeargs' : [ '-s', 'n-288hr', '-e', 'n' ], |
| 77 |
'timetag' : 'week', |
| 78 |
'avrange' : 7*24*3600, |
| 79 |
'avres' : 600, |
| 80 |
'sizeargs' : { 'small' : [ '--width', '576', '--height', '125' ], |
| 81 |
'large' : [ '--width', '1728', '--height', '375', |
| 82 |
'-n', 'DEFAULT:16:'] |
| 83 |
}, |
| 84 |
}, |
| 85 |
|
| 86 |
# note the construction "repr(576*n)" here --- this is because the plot is |
| 87 |
# (a multiple of) 576 pixels, and 120 min is one of the RRAs, so choosing |
| 88 |
# a lower limit of 576*120 gives us a plot with one pixel per RRA bin. |
| 89 |
|
| 90 |
'month' : { 'timeargs' : [ '-s', 'n-'+repr(576*120)+'min', '-e', 'n'], |
| 91 |
'timetag' : 'month', |
| 92 |
'avrange' : 31*24*3600, |
| 93 |
'avres' : 1800, |
| 94 |
'sizeargs' : { 'small' : [ '--width', '576', '--height', '105'], |
| 95 |
'large' : [ '--width', '2304', '--height', '420', |
| 96 |
'-n', 'DEFAULT:18:', |
| 97 |
'--x-grid', |
| 98 |
'HOUR:12:DAY:1:DAY:3:86400:%d-%b' |
| 99 |
] |
| 100 |
}, |
| 101 |
}, |
| 102 |
|
| 103 |
'year' : { 'timeargs' : [ '-s', 'n-'+repr(576*1440)+'min', '-e', 'n'], |
| 104 |
'timetag' : 'year', |
| 105 |
'avrange' : 365*24*3600, |
| 106 |
'avres' : 86400, |
| 107 |
'sizeargs' : { 'small' : [ '--width', '576', '--height', '105'], |
| 108 |
'large' : [ '--width', '2304', '--height', '420'] |
| 109 |
}, |
| 110 |
}, |
| 111 |
|
| 112 |
# adjusting for nice plot ... now 15/4 of a year |
| 113 |
|
| 114 |
'alltime' : { 'timeargs' : [ '-s', 'n-'+repr(15*365*1440/4)+'min', '-e', 'n'], |
| 115 |
'timetag' : 'alltime', |
| 116 |
'avrange' : 15*365*24*3600/4, |
| 117 |
'avres' : 86400, |
| 118 |
'sizeargs' : { 'small' : [ '--width', '576', '--height', '105', |
| 119 |
'--x-grid', |
| 120 |
'MONTH:3:YEAR:1:YEAR:1:31536000:%Y' |
| 121 |
], |
| 122 |
'large' : [ '--width', '2304', '--height', '420', |
| 123 |
'--x-grid', |
| 124 |
'MONTH:1:YEAR:1:MONTH:3:2592000:%b-%Y' |
| 125 |
] |
| 126 |
}, |
| 127 |
}, |
| 128 |
|
| 129 |
|
| 130 |
} |
| 131 |
|
| 132 |
commonargs = ['--imgformat', 'PNG', |
| 133 |
'--legend-position=east', '--legend-direction=bottomup'] |
| 134 |
|
| 135 |
import rrdtool |
| 136 |
import time |
| 137 |
import glob |
| 138 |
|
| 139 |
### function definitions |
| 140 |
|
| 141 |
def doplot(glist, dbtype, psize, timetag, sizeargs, timeargs, pcents, ranktype): |
| 142 |
|
| 143 |
grouplist = glist[:] |
| 144 |
defs = list() |
| 145 |
plots = list() |
| 146 |
|
| 147 |
data_defs = list() |
| 148 |
plot_defs = list() |
| 149 |
|
| 150 |
gcolors = dict() |
| 151 |
|
| 152 |
for idx in range(len(grouplist)): |
| 153 |
gcolors[grouplist[idx]] = colors[idx] |
| 154 |
|
| 155 |
if ranktype == 'bottom': |
| 156 |
if 'unused' in grouplist: |
| 157 |
grouplist.remove('unused') |
| 158 |
grouplist.insert(0,'unused') |
| 159 |
if 'offline' in grouplist: |
| 160 |
grouplist.remove('offline') |
| 161 |
grouplist.insert(0,'offline') |
| 162 |
|
| 163 |
if dbtype == 'queued': |
| 164 |
if 'unused' in grouplist: grouplist.remove('unused') |
| 165 |
if 'offline' in grouplist: grouplist.remove('offline') |
| 166 |
|
| 167 |
if ranktype == 'top': |
| 168 |
for group in (grouplist + ['total']): |
| 169 |
data_defs.append('DEF:'+group+'='+DATADIR+group+'.'+dbtype+'.rrd:'+dbtype+':AVERAGE') |
| 170 |
otherstring = 'CDEF:other=total,' |
| 171 |
for group in grouplist: |
| 172 |
otherstring += group + ',' |
| 173 |
otherstring += (len(grouplist)-1) * '+,' + '-' |
| 174 |
|
| 175 |
# print otherstring |
| 176 |
data_defs.append(otherstring) |
| 177 |
elif ranktype == 'bottom': |
| 178 |
for group in grouplist: |
| 179 |
data_defs.append('DEF:'+group+'='+DATADIR+group+'.'+dbtype+'.rrd:'+dbtype+':AVERAGE') |
| 180 |
else: |
| 181 |
print 'Unknown ranktype detected:', ranktype |
| 182 |
sys.exit(2) |
| 183 |
|
| 184 |
sumshown = 0 |
| 185 |
for idx in range(len(grouplist)): |
| 186 |
group = grouplist[idx] |
| 187 |
if group == 'unused': |
| 188 |
acolor = '#d8d8d8' |
| 189 |
elif group == "offline": |
| 190 |
acolor = "#790ead" |
| 191 |
else: |
| 192 |
acolor = gcolors[group] |
| 193 |
pdefstr = 'AREA' ':' + group + acolor + ':' + "%8s" % (group) |
| 194 |
if pcents: |
| 195 |
pdefstr = pdefstr + ' (' + "%4.1f" % (pcents[group]) + ')' |
| 196 |
sumshown += float(pcents[group]) |
| 197 |
pdefstr = pdefstr + '\\n' |
| 198 |
pdefstr = pdefstr + ':STACK' |
| 199 |
plot_defs.append(pdefstr) |
| 200 |
|
| 201 |
if ranktype == 'top': |
| 202 |
pdefstr = 'AREA' ':' + 'other' + '#794044' + ':' + ' other' |
| 203 |
if pcents: |
| 204 |
pdefstr = pdefstr + ' (' + "%4.1f" % (100 - sumshown) + ')' |
| 205 |
pdefstr = pdefstr + '\\n' |
| 206 |
plot_defs.insert(0,pdefstr) |
| 207 |
plot_defs.append("LINE:total#000000") # :total") |
| 208 |
|
| 209 |
pargs = [ PLOTDIR + dbtype + '-' + timetag + '-' + ranktype + '-' + \ |
| 210 |
psize + '.png'] + commonargs + ['-l', '0'] + sizeargs[psize] + \ |
| 211 |
timeargs + data_defs + plot_defs |
| 212 |
rrdtool.graph( *pargs ) |
| 213 |
|
| 214 |
def doplot_wait(glist, dbtype, psize, timetag, sizeargs, timeargs, ranktype): |
| 215 |
|
| 216 |
grouplist = glist[:] |
| 217 |
defs = list() |
| 218 |
plots = list() |
| 219 |
|
| 220 |
data_defs = list() |
| 221 |
plot_defs = list() |
| 222 |
|
| 223 |
if dbtype == 'waittime': |
| 224 |
if 'unused' in grouplist: grouplist.remove('unused') |
| 225 |
if 'offline' in grouplist: grouplist.remove('offline') |
| 226 |
|
| 227 |
for group in (grouplist + ['rollover','lastroll']): |
| 228 |
data_defs.append('DEF:'+group+'='+DATADIR+group+'.'+dbtype+'.rrd:'+dbtype+':AVERAGE') |
| 229 |
|
| 230 |
for idx in range(len(grouplist)): |
| 231 |
group = grouplist[idx] |
| 232 |
if group == 'unused': |
| 233 |
acolor = '#d8d8d8' |
| 234 |
elif group == "offline": |
| 235 |
acolor = "#790ead" |
| 236 |
else: |
| 237 |
acolor = colors[idx] |
| 238 |
pdefstr = 'LINE3' ':' + group + acolor + ':' + group |
| 239 |
pdefstr = pdefstr + '\\n' |
| 240 |
plot_defs.append(pdefstr) |
| 241 |
|
| 242 |
plot_defs.append('LINE2:rollover#660198') |
| 243 |
plot_defs.append('LINE2:lastroll#000000') |
| 244 |
|
| 245 |
pargs = [ PLOTDIR + dbtype + '-' + timetag + '-' + ranktype + '-' + \ |
| 246 |
psize + '.png'] + ['--slope-mode', '-o'] + commonargs + \ |
| 247 |
sizeargs[psize] + timeargs + data_defs + plot_defs |
| 248 |
rrdtool.graph( *pargs ) |
| 249 |
|
| 250 |
def makeplots(prangedef): |
| 251 |
|
| 252 |
resolu = prangedef['avres'] |
| 253 |
|
| 254 |
# first need to find "top eight" list |
| 255 |
# base it on running jobs |
| 256 |
|
| 257 |
now=int(time.mktime(time.localtime())) |
| 258 |
end = (now / resolu) * resolu |
| 259 |
start = end - (prangedef['avrange']) + resolu |
| 260 |
|
| 261 |
### block finding 'top N' group list ### |
| 262 |
|
| 263 |
tgroup = dict() # structure tgroup[groupname] = total of hourly average |
| 264 |
|
| 265 |
running_files = glob.glob(DATADIR+'*.running.rrd') |
| 266 |
for db in running_files: |
| 267 |
group = db[len(DATADIR):db.find('.running.rrd')] |
| 268 |
tup = rrdtool.fetch(db,'AVERAGE','-r', repr(resolu), |
| 269 |
'-s', repr(start), '-e', repr(end)) |
| 270 |
vallist = [0] # start with zero, in case no vals returned, get zero as answer |
| 271 |
for tup2 in tup[2]: |
| 272 |
val = tup2[0] |
| 273 |
if val: |
| 274 |
vallist.append(val) |
| 275 |
|
| 276 |
# put numbers in meaningful units now. result returned is an |
| 277 |
# integration over the time range, of averages over "resolu" |
| 278 |
# ... native resolution is in minutes, so multiplying by |
| 279 |
# (resolu / 60) puts the answer in core-minutes; dividing by |
| 280 |
# the number of minutes in the range gives the average number |
| 281 |
# of cores occupied, over the range |
| 282 |
|
| 283 |
if group == "total": |
| 284 |
totval = sum(vallist) * (resolu / 60) / ( (end-start) / 60. ) |
| 285 |
else: |
| 286 |
tgroup[group] = sum(vallist) * (resolu / 60) / ( (end-start) / 60. ) |
| 287 |
|
| 288 |
pgroup = dict() |
| 289 |
for g in tgroup.keys(): |
| 290 |
pgroup[g] = 100*tgroup[g]/totval |
| 291 |
|
| 292 |
### start block 'top N' and 'next N' group list ### |
| 293 |
|
| 294 |
groups_sorted = sorted(tgroup, key=tgroup.get, reverse=False) # increasing useage towards end of list |
| 295 |
|
| 296 |
topgroups = groups_sorted[ -NUMGROUPS: ] |
| 297 |
bottgroups = groups_sorted[-2*NUMGROUPS:-NUMGROUPS] |
| 298 |
|
| 299 |
for glist in [topgroups, bottgroups]: |
| 300 |
if 'unused' in glist: |
| 301 |
glist.remove('unused') |
| 302 |
glist.append('unused') # makes it always appear at top of plot (except see offline) |
| 303 |
if 'offline' in glist: |
| 304 |
glist.remove('offline') |
| 305 |
glist.append('offline') # makes it always appear at top of plot |
| 306 |
|
| 307 |
if opts.rankonly and prangedef['timetag'] != 'hr' : |
| 308 |
print "Ranked average running jobs over period", prangedef['timetag'] |
| 309 |
print "%10s %12d" % ('total', totval) |
| 310 |
rank = 0 |
| 311 |
for g in reversed(groups_sorted): |
| 312 |
if tgroup[g] > 0: |
| 313 |
rank += 1 |
| 314 |
print "%2d. %10s %9.4f %9.4f%%" % (rank, g, tgroup[g], pgroup[g]) |
| 315 |
return |
| 316 |
|
| 317 |
### end block 'top N' group list ### |
| 318 |
|
| 319 |
### block generating plots ### |
| 320 |
|
| 321 |
|
| 322 |
for dbtype in ['queued', 'running', 'waittime']: |
| 323 |
|
| 324 |
### this is a bit of a hack : we only want pgroups for when it's |
| 325 |
### a 'running' database and we don't want it for timetag hour. |
| 326 |
### fix this up here |
| 327 |
|
| 328 |
if dbtype == 'running' and prangedef['timetag'] != 'hr': |
| 329 |
percents = pgroup |
| 330 |
else: |
| 331 |
percents = None |
| 332 |
for grps in [ ('top', topgroups), ('bottom', bottgroups) ]: |
| 333 |
for psize in ['small', 'large']: |
| 334 |
if dbtype == 'waittime': |
| 335 |
doplot_wait(grps[1], dbtype, psize, prangedef['timetag'], |
| 336 |
prangedef['sizeargs'], |
| 337 |
prangedef['timeargs'], grps[0] |
| 338 |
) |
| 339 |
else: |
| 340 |
doplot(grps[1], dbtype, psize, prangedef['timetag'], |
| 341 |
prangedef['sizeargs'], |
| 342 |
prangedef['timeargs'], |
| 343 |
percents, grps[0] |
| 344 |
) |
| 345 |
|
| 346 |
for k in ['hr', 'day', 'week', 'month', 'year', 'alltime']: # plotrangedef.keys(): |
| 347 |
makeplots(plotrangedef[k]) |
| 348 |
|
| 349 |
import sys |
| 350 |
sys.exit(0) |
Parent Directory
| 
Revision Log