1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
|
#!/usr/bin/env python
# spawns a ddd session for each one found in mpilistjobs
# nco-specific: sends SIGUSR1 to nodeid>0 processes
# $Id: ddd_mpd.py,v 1.2 2005/09/26 23:13:38 wangd Exp $
# Usage:
# ./ddd_mpd.py
# -- spawn ddd sessions for each MPI process found via
# mpdlistjobs for a particular MPI job. Before spawning, send SIGUSR1
# to the processes of rank > 0 (the non-manager nodes) to reduce user
# tedium. If multiple jobs are found on MPD, prompt user to select
# one of jobs.
# -- If no jobs owned by the current user are found in mpd,
# complain and exit.
#
# Prefer something else than ddd? Try changing the argument at the
# call to mainthing(...). Support hasn't been checked for
# command-line arguments other than ddd/gdb, so you may have to hack
# spawnDebugger(...) as a short-term solution.
#
# Please direct feedback to the sourceforge forums for NCO. Thanks, -Daniel
from os import environ, getuid, getpid, path, getcwd, popen3
import os
class Job:
pass
def newJob(jid, username, host, pid, sid, rank, path):
job = Job()
job.jobid = jid
job.username = username
job.host = host
job.pid = pid
job.sid = sid
job.rank = rank
job.pgm = path
return job
def readJobInfo(jobtext):
job = Job()
#print "got job ", jobtext
lines = jobtext.split("\n")
for line in lines:
pair = line.split("=")
if len(pair) > 1:
key,value = pair
#print key.strip(),value.strip()
exec "job."+key.strip()+"="+"\""+value.strip()+"\""
return job;
def readJobListMPD():
joblist = []
jobI, jobO, jobE = popen3("mpdlistjobs")
allinfo = jobO.read()
for jobtext in allinfo.split("\n\n"):
joblist.append(readJobInfo(jobtext))
return joblist
def makeSampleJobList(): ## dummy for testing
joblist = []
joblist.append(newJob( "21@dirt_3734", "wangd", "dirt",
"11931", "11926", "0",
"/home/wangd/nco/nco/mpi_bin/mpncwa"))
joblist.append(newJob( "21@dirt_3734", "wangd", "dirt",
"11929", "11927", "1",
"/home/wangd/nco/nco/mpi_bin/mpncwa"))
joblist.append(newJob( "21@dirt_3734", "wangd", "dirt",
"11930", "11928", "2",
"/home/wangd/nco/nco/mpi_bin/mpncwa"))
# next two jobs should be culled
joblist.append(newJob( "10@dirt_3734", "somebody", "dirt",
"112", "110", "0",
"/badpath/mpncbo"))
joblist.append(newJob( "10@dirt_3734", "somebody", "dirt",
"119", "111", "1",
"/badpath/mpncbo"))
joblist.append(newJob( "24@dirt_3734", "wangd", "dirt",
"11951", "11946", "0",
"/home/wangd/nco/nco/mpi_bin/mpncbo"))
joblist.append(newJob( "24@dirt_3734", "wangd", "dirt",
"11949", "11947", "1",
"/home/wangd/nco/nco/mpi_bin/mpncbo"))
joblist.append(newJob( "24@dirt_3734", "wangd", "dirt",
"11950", "11948", "2",
"/home/wangd/nco/nco/mpi_bin/mpncbo"))
return joblist
def cutoutJobs(joblist, jobids, paths):
print "There seems to be more than one job in MPD. Which would you like?"
pick = ""
index = -1
while not pick in jobids:
for i in range(len(jobids)):
print i,jobids[i], paths[i]
print "which would you like(0..",len(jobids)-1,")?",
typed = raw_input()
try:
index = int(typed)
except ValueError:
pass
if index not in range(len(jobids)):
print "Sorry, bad choice: ",typed, " Try another."
else:
pick = jobids[index]
print "Chose id:", pick, " path:",paths[index]
# remove job objects that do not match
joblist = filter((lambda j: j.jobid == pick), joblist)
return joblist
def spawnDebugger(progname, joblist, debug):
shI, shO, shE = popen3("sh")
children = filter(lambda j: j.rank != "0", joblist)
for c in children: # resume the children first
s = "kill -USR1 %s\n" % (c.pid)
if not debug: shI.write(s)
else: print s,
print "Node 0 is pid:",filter(lambda j: j.rank == "0",joblist)[0].pid
shI.write("\n")
for job in joblist:
try:
#print job.jobid, job.pid, job.rank, job.pgm
dummy = job.pgm
dummy = job.pid
s = "%s %s %s &\n" % (progname, job.pgm, job.pid)
if not debug: shI.write(s)
#else:
print s,
except AttributeError:
continue
shI.close()
print shO.read().strip()
print shE.read().strip()
def mainthing(dbgprogname):
joblist = readJobListMPD()
#joblist = makeSampleJobList() # for simple testing
# now, go ahead and spawn ddd jobs.
# for now, just spit out the command line so we don't have to deal
# with the process management. *sigh*
####print len(joblist)
jobids = []
paths = []
joblist = filter(lambda j: hasattr(j, "jobid"), joblist)
myname = os.popen("whoami").read().strip()
joblist = filter(lambda j: j.username == myname, joblist)
if len(joblist) < 1:
print "No acceptable jobs found."
return
for job in joblist: ## check to see how many jobs there are
if not job.jobid in jobids:
jobids.append( job.jobid )
paths.append( job.pgm )
if len(jobids) > 1:
joblist = cutoutJobs(joblist, jobids, paths)
spawnDebugger(dbgprogname, joblist, False) #False for no debug
## -----------------
## main program body
## -----------------
mainthing("ddd")
# if you like gdb or dbx, you may wish to try something like:
#
# mainthing ("xterm -e gdb")
#
# You will probably want to spawn xterms for each gdb to avoid
# managing multiple gdb sessions from a single terminal window.
|
