|
1.3.7
by Gennaro Oliva
Import upstream version 2.1.0 |
1 |
#!/usr/bin/expect
|
2 |
############################################################################
|
|
3 |
# Purpose: Test of SLURM functionality
|
|
4 |
# Test of sinfo cpu total and allocated
|
|
5 |
#
|
|
6 |
# Output: "TEST: #.#" followed by "SUCCESS" if test was successful, OR
|
|
7 |
# "FAILURE: ..." otherwise with an explanation of the failure, OR
|
|
8 |
# anything else indicates a failure mode that must be investigated.
|
|
9 |
############################################################################
|
|
10 |
# Copyright (C) 2009 Lawrence Livermore National Security.
|
|
11 |
# Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
|
|
12 |
# Written by Joseph Donaghy <donaghy1@llnl.gov>
|
|
13 |
# CODE-OCEC-09-009. All rights reserved.
|
|
14 |
#
|
|
15 |
# This file is part of SLURM, a resource management program.
|
|
16 |
# For details, see <https://computing.llnl.gov/linux/slurm/>.
|
|
17 |
# Please also read the included file: DISCLAIMER.
|
|
18 |
#
|
|
19 |
# SLURM is free software; you can redistribute it and/or modify it under
|
|
20 |
# the terms of the GNU General Public License as published by the Free
|
|
21 |
# Software Foundation; either version 2 of the License, or (at your option)
|
|
22 |
# any later version.
|
|
23 |
#
|
|
24 |
# SLURM is distributed in the hope that it will be useful, but WITHOUT ANY
|
|
25 |
# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
|
|
26 |
# FOR A PARTICULAR PURPOSE. See the GNU General Public License for more
|
|
27 |
# details.
|
|
28 |
#
|
|
29 |
# You should have received a copy of the GNU General Public License along
|
|
30 |
# with SLURM; if not, write to the Free Software Foundation, Inc.,
|
|
31 |
# 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
|
32 |
############################################################################
|
|
33 |
source ./globals
|
|
34 |
||
35 |
set test_id "4.12" |
|
36 |
set file_in "test$test_id.in" |
|
37 |
set exit_code 0 |
|
38 |
set pnumsc 0 |
|
39 |
set pnumsi 0 |
|
40 |
set aprocsc 0 |
|
41 |
set aprocsi 0 |
|
42 |
set inode_found 0 |
|
43 |
set inode_name "" |
|
44 |
set smallest 1 |
|
45 |
set layout "static" |
|
46 |
set select_type "" |
|
47 |
set node_scaling 0 |
|
48 |
||
49 |
proc scontrol_test { node proc_cnt } {
|
|
50 |
global scontrol number prompt |
|
51 |
upvar spawn_id spawn_id |
|
52 |
||
53 |
set found 0 |
|
54 |
set rc 0 |
|
55 |
||
56 |
send "$scontrol show node $node\r" |
|
57 |
expect {
|
|
58 |
-re "CPUAlloc=($number)" {
|
|
59 |
set num_alloc $expect_out(1,string) |
|
60 |
set found 1 |
|
61 |
if {$proc_cnt != $num_alloc} {
|
|
62 |
send_user "\nFAILURE: requested $proc_cnt but got $num_alloc instead\n" |
|
63 |
set rc 1 |
|
64 |
} |
|
65 |
exp_continue |
|
66 |
} |
|
67 |
-re $prompt {
|
|
68 |
} |
|
69 |
timeout {
|
|
70 |
send_user "\nFAILURE: scontrol not responding\n" |
|
71 |
slow_kill $mypid |
|
72 |
return 1 |
|
73 |
} |
|
74 |
eof {
|
|
75 |
wait |
|
76 |
} |
|
77 |
} |
|
78 |
||
79 |
if {!$found} {
|
|
80 |
send_user "\nFAILURE: didn't get expected output from scontrol\n" |
|
81 |
set rc 1 |
|
82 |
} |
|
83 |
||
84 |
return $rc |
|
85 |
} |
|
86 |
||
87 |
proc sinfo_test_1 { node proc_cnt total_procs idle_cpus } {
|
|
88 |
global sinfo number prompt |
|
89 |
upvar spawn_id spawn_id |
|
90 |
||
91 |
set found 0 |
|
92 |
set rc 0 |
|
93 |
||
94 |
send "$sinfo -o \"%C %A %N\" -h -n $node\r" |
|
95 |
# make sure we get by the sinfo command so we don't |
|
96 |
# think the %'s are a prompt |
|
97 |
expect {
|
|
98 |
"%C %A %N" {
|
|
99 |
} |
|
100 |
} |
|
101 |
expect {
|
|
102 |
-re "($number)(K?).($number)(K?).($number)(K?).($number)(K?) ($number)(K?).($number)(K?) $node" {
|
|
103 |
set found 1 |
|
104 |
set num_alloc $expect_out(1,string) |
|
105 |
if {[string compare $expect_out(2,string) ""]} {
|
|
106 |
set num_alloc [expr $num_alloc * 1024] |
|
107 |
} |
|
108 |
set num_idle $expect_out(3,string) |
|
109 |
if {[string compare $expect_out(4,string) ""]} {
|
|
110 |
set num_idle [expr $num_idle * 1024] |
|
111 |
} |
|
112 |
set num_other $expect_out(5,string) |
|
113 |
if {[string compare $expect_out(6,string) ""]} {
|
|
114 |
set num_other [expr $num_other * 1024] |
|
115 |
} |
|
116 |
set num_total $expect_out(7,string) |
|
117 |
if {[string compare $expect_out(8,string) ""]} {
|
|
118 |
set num_total [expr $num_total * 1024] |
|
119 |
} |
|
120 |
||
121 |
if { $num_alloc != $proc_cnt } {
|
|
122 |
send_user "\nFAILURE: sinfo 1 allocated cpus wrong, got $num_alloc but needed $proc_cnt\n" |
|
123 |
set rc 1 |
|
124 |
} elseif { $num_idle != $idle_cpus } {
|
|
125 |
send_user "\nFAILURE: sinfo 1 idle cpus wrong, got $num_idle but needed $idle_cpus\n" |
|
126 |
set rc 1 |
|
127 |
} elseif { $num_total != $total_procs } {
|
|
128 |
send_user "\nFAILURE: sinfo 1 total cpus wrong, got $num_total but needed $total_procs\n" |
|
129 |
set rc 1 |
|
130 |
} |
|
131 |
exp_continue |
|
132 |
} |
|
133 |
-re $prompt {
|
|
134 |
} |
|
135 |
timeout {
|
|
136 |
send_user "\nFAILURE: sinfo not responding\n" |
|
137 |
slow_kill $mypid |
|
138 |
return 1 |
|
139 |
} |
|
140 |
eof {
|
|
141 |
wait |
|
142 |
} |
|
143 |
} |
|
144 |
||
145 |
if {!$found} {
|
|
146 |
send_user "\nFAILURE: didn't get expected output from sinfo\n" |
|
147 |
set rc 1 |
|
148 |
} |
|
149 |
||
150 |
return $rc |
|
151 |
} |
|
152 |
||
153 |
proc sinfo_test_2 { node proc_cnt total_procs } {
|
|
154 |
global sinfo number prompt node_scaling |
|
155 |
upvar spawn_id spawn_id |
|
156 |
||
157 |
set rc 0 |
|
158 |
set num_alloc 0 |
|
159 |
set num_idle 0 |
|
160 |
set alloc_nodes 1 |
|
161 |
set total_nodes 1 |
|
162 |
||
163 |
if {$node_scaling} {
|
|
164 |
set alloc_nodes [expr $proc_cnt / $node_scaling] |
|
165 |
set total_nodes [expr $total_procs / $node_scaling] |
|
166 |
} |
|
167 |
set idle_nodes [expr $total_nodes - $alloc_nodes] |
|
168 |
||
169 |
send "$sinfo -o \"%t %D %N\" -h -n $node\r" |
|
170 |
# make sure we get by the sinfo command so we don't |
|
171 |
# think the %'s are a prompt |
|
172 |
expect {
|
|
173 |
"%t %D %N" {
|
|
174 |
} |
|
175 |
} |
|
176 |
expect {
|
|
177 |
-re "alloc ($number)(K?) $node" {
|
|
178 |
set num_alloc $expect_out(1,string) |
|
179 |
if {[string compare $expect_out(2,string) ""]} {
|
|
180 |
set num_alloc [expr $inode_procs * 1024] |
|
181 |
} |
|
182 |
exp_continue |
|
183 |
} |
|
184 |
-re "idle ($number)(K?) $node" {
|
|
185 |
set num_idle $expect_out(1,string) |
|
186 |
if {[string compare $expect_out(2,string) ""]} {
|
|
187 |
set num_idle [expr $num_idle * 1024] |
|
188 |
} |
|
189 |
exp_continue |
|
190 |
} |
|
191 |
-re $prompt {
|
|
192 |
} |
|
193 |
timeout {
|
|
194 |
send_user "\nFAILURE: sinfo not responding\n" |
|
195 |
slow_kill $mypid |
|
196 |
return 1 |
|
197 |
} |
|
198 |
eof {
|
|
199 |
wait |
|
200 |
} |
|
201 |
} |
|
202 |
||
203 |
if { $num_alloc != $alloc_nodes } {
|
|
204 |
send_user "\nFAILURE: sinfo 2 allocated nodes wrong, got $num_alloc but needed $alloc_nodes\n" |
|
205 |
set rc 1 |
|
206 |
} elseif { $num_idle != $idle_nodes } {
|
|
207 |
send_user "\nFAILURE: sinfo 2 idle nodes wrong, got $num_idle but needed $idle_nodes\n" |
|
208 |
set rc 1 |
|
209 |
} |
|
210 |
||
211 |
return $rc |
|
212 |
} |
|
213 |
||
214 |
# allocate a set of nodes (node_cnt) and the quit right after |
|
215 |
proc allocate_and_quit { node proc_cnt total_procs } {
|
|
216 |
global salloc scontrol sinfo number alpha_numeric_under |
|
217 |
global prompt select_type procs_per_node |
|
218 |
||
219 |
set job_id 0 |
|
220 |
set num_alloc 0 |
|
221 |
set block "" |
|
222 |
set rc 0 |
|
223 |
set timeout 60 |
|
224 |
set idle_cpus [expr $total_procs - $proc_cnt] |
|
225 |
||
226 |
set mypid [spawn $salloc -w $node -N1 -n $proc_cnt bash] |
|
227 |
expect {
|
|
228 |
-re "Granted job allocation ($number)" {
|
|
229 |
set job_id $expect_out(1,string) |
|
230 |
exp_continue |
|
231 |
} |
|
232 |
||
233 |
-re $prompt {
|
|
234 |
# test for scontrol to give me the correct cpu count |
|
235 |
if { [scontrol_test $node $proc_cnt] } {
|
|
236 |
send "exit\r" |
|
237 |
return 1 |
|
238 |
} |
|
239 |
||
240 |
||
241 |
# test for sinfo to give me the correct cpu count |
|
242 |
if { [sinfo_test_1 $node $proc_cnt $total_procs $idle_cpus] } {
|
|
243 |
send "exit\r" |
|
244 |
return 1 |
|
245 |
} |
|
246 |
# test for sinfo to give me the correct node count |
|
247 |
if { [sinfo_test_2 $node $proc_cnt $total_procs] } {
|
|
248 |
send "exit\r" |
|
249 |
return 1 |
|
250 |
} |
|
251 |
send "exit\r" |
|
252 |
exp_continue |
|
253 |
} |
|
254 |
||
255 |
-re "Unable to contact" {
|
|
256 |
send_user "\nFAILURE: slurm appears to be down\n" |
|
257 |
exp_continue |
|
258 |
} |
|
259 |
timeout {
|
|
260 |
send_user "\nFAILURE: salloc not responding\n" |
|
261 |
if {$job_id != 0} {
|
|
262 |
cancel_job $job_id |
|
263 |
} |
|
264 |
slow_kill $mypid |
|
265 |
return 1 |
|
266 |
} |
|
267 |
eof {
|
|
268 |
wait |
|
269 |
} |
|
270 |
} |
|
271 |
||
272 |
return $rc |
|
273 |
} |
|
274 |
||
275 |
############################################################################ |
|
276 |
# test starts here |
|
277 |
############################################################################ |
|
278 |
||
279 |
print_header $test_id |
|
280 |
||
281 |
# find the default partition |
|
|
1.4.6
by Gennaro Oliva
Import upstream version 2.1.11 |
282 |
set def_part [default_partition] |
|
1.3.7
by Gennaro Oliva
Import upstream version 2.1.0 |
283 |
|
284 |
# find the nodes in the default partition |
|
285 |
log_user 0 |
|
286 |
set def_hostlist "" |
|
287 |
set part_exclusive 0 |
|
288 |
spawn $scontrol show part $def_part |
|
289 |
expect {
|
|
290 |
-re " Shared=EXCLUSIVE" {
|
|
291 |
set part_exclusive 1 |
|
292 |
exp_continue |
|
293 |
} |
|
294 |
-re " Nodes=($alpha_numeric_nodelist)" {
|
|
295 |
set def_hostlist $expect_out(1,string) |
|
296 |
exp_continue |
|
297 |
} |
|
298 |
-re " BasePartitions=($alpha_numeric_nodelist)" {
|
|
299 |
set def_hostlist $expect_out(1,string) |
|
300 |
exp_continue |
|
301 |
} |
|
302 |
timeout {
|
|
303 |
send_user "\nFAILURE: scontrol not responding\n" |
|
304 |
set exit_code 1 |
|
305 |
} |
|
306 |
eof {
|
|
307 |
wait |
|
308 |
} |
|
309 |
} |
|
310 |
set host_cnt 0 |
|
311 |
spawn $scontrol show hostnames $def_hostlist |
|
312 |
expect {
|
|
313 |
-re "($alpha_numeric_under)" {
|
|
314 |
set host_name($host_cnt) $expect_out(1,string) |
|
315 |
incr host_cnt |
|
316 |
exp_continue |
|
317 |
} |
|
318 |
timeout {
|
|
319 |
send_user "\nFAILURE: scontrol not responding\n" |
|
320 |
set exit_code 1 |
|
321 |
} |
|
322 |
eof {
|
|
323 |
wait |
|
324 |
} |
|
325 |
} |
|
326 |
log_user 1 |
|
327 |
if {$host_cnt == 0} {
|
|
328 |
send_user "\nFAILURE: could not find any nodes in default partition\n" |
|
329 |
exit 1 |
|
330 |
} |
|
331 |
||
332 |
# find me an idle node in default partition |
|
333 |
log_user 0 |
|
|
1.4.6
by Gennaro Oliva
Import upstream version 2.1.11 |
334 |
set inode_name "" |
335 |
set inode_cores_per_socket 0 |
|
|
1.3.7
by Gennaro Oliva
Import upstream version 2.1.0 |
336 |
set inode_procs 0 |
|
1.4.6
by Gennaro Oliva
Import upstream version 2.1.11 |
337 |
set units "" |
338 |
set inode_sockets 0 |
|
339 |
set inode_threads_per_core 0 |
|
340 |
||
341 |
set fd [open "|$scontrol --oneliner show node $def_hostlist"] |
|
342 |
exp_internal 1 |
|
343 |
while {[gets $fd line] != -1} {
|
|
344 |
if {[regexp {NodeName=(\w+).*CoresPerSocket=(\d+).*CPUTot=(\d+)(K?).*Sockets=(\d+) State=IDLE ThreadsPerCore=(\d+)} $line frag inode_name inode_cores_per_socket inode_procs units inode_sockets inode_threads_per_core] == 1} {
|
|
345 |
break |
|
346 |
} |
|
347 |
} |
|
348 |
exp_internal 0 |
|
349 |
if {[string compare $units ""]} {
|
|
350 |
set inode_procs [expr $inode_procs * 1024] |
|
351 |
} |
|
352 |
catch {close $fd}
|
|
353 |
||
|
1.3.7
by Gennaro Oliva
Import upstream version 2.1.0 |
354 |
log_user 1 |
355 |
||
356 |
if {!$inode_procs} {
|
|
357 |
send_user "\nFAILURE: couldn't find an idle node in the default partition\n" |
|
358 |
exit 1 |
|
359 |
} |
|
360 |
||
361 |
send_user "found idle node $inode_name with $inode_procs processors\n" |
|
362 |
||
363 |
# figure out the select plugin we are using |
|
364 |
set select_type [test_select_type] |
|
365 |
if {![string compare $select_type "bluegene"]} {
|
|
366 |
# figure out some things if a bluegene system |
|
367 |
set layout [get_bluegene_layout] |
|
368 |
if {$layout == 0} {
|
|
369 |
send_user "\nFAILURE: No layout mode found for this system\n" |
|
370 |
exit 1 |
|
371 |
} |
|
372 |
set psets [get_bluegene_psets] |
|
373 |
||
374 |
if {$psets == 0} {
|
|
375 |
send_user "\nFAILURE: No psets are set on this system\n" |
|
376 |
exit 1 |
|
377 |
} |
|
378 |
set type [get_bluegene_type] |
|
379 |
||
380 |
if {$type == 0} {
|
|
381 |
send_user "\nFAILURE: No bluegene type found \n" |
|
382 |
exit 1 |
|
383 |
} |
|
384 |
||
385 |
if {![string compare $type "P"]} {
|
|
386 |
if {$psets >= 32} {
|
|
387 |
set smallest 16 |
|
388 |
} elseif {$psets >= 16} {
|
|
389 |
set smallest 32 |
|
390 |
} elseif {$psets >= 8} {
|
|
391 |
set smallest 64 |
|
392 |
} else {
|
|
393 |
set smallest 128 |
|
394 |
} |
|
395 |
} elseif {![string compare $type "L"]} {
|
|
396 |
if {$psets >= 16} {
|
|
397 |
set smallest 32 |
|
398 |
} else {
|
|
399 |
set smallest 128 |
|
400 |
} |
|
401 |
} else {
|
|
402 |
send_user "\nFAILURE: unknown bluegene system type '$type'\n"; |
|
403 |
exit 1 |
|
404 |
} |
|
405 |
set node_scaling [get_bluegene_procs_per_cnode] |
|
406 |
set smallest [expr $smallest * $node_scaling] |
|
407 |
} elseif {![string compare $select_type "linear"]} {
|
|
408 |
set smallest $inode_procs |
|
409 |
} else {
|
|
410 |
set select_params [test_select_type_params] |
|
411 |
if {$part_exclusive == 1} {
|
|
412 |
set smallest $inode_procs |
|
413 |
} elseif {![string compare $select_params "CR_CPU"]} {
|
|
414 |
set smallest $inode_threads_per_core |
|
415 |
} elseif {![string compare $select_params "CR_CPU_MEMORY"]} {
|
|
416 |
set smallest $inode_threads_per_core |
|
417 |
} elseif {![string compare $select_params "NONE"]} {
|
|
418 |
set smallest $inode_threads_per_core |
|
419 |
} elseif {![string compare $select_params "CR_CORE"]} {
|
|
420 |
set smallest $inode_threads_per_core |
|
421 |
} elseif {![string compare $select_params "CR_CORE_MEMORY"]} {
|
|
422 |
set smallest $inode_threads_per_core |
|
423 |
} elseif {![string compare $select_params "CR_SOCKET"]} {
|
|
424 |
set smallest [expr $inode_cores_per_socket *$inode_threads_per_core] |
|
425 |
} elseif {![string compare $select_params "CR_SOCKET_MEMORY"]} {
|
|
426 |
set smallest [expr $inode_cores_per_socket *$inode_threads_per_core] |
|
427 |
} else {
|
|
428 |
send_user "\nWARNING: failed to parse SelectTypeParameters '$select_params'\n" |
|
429 |
set smallest $inode_procs |
|
430 |
} |
|
431 |
} |
|
432 |
||
433 |
set exit_code [allocate_and_quit $inode_name $smallest $inode_procs] |
|
434 |
if {!$exit_code && $smallest != $inode_procs} {
|
|
|
1.6.4
by Gennaro Oliva
Import upstream version 2.2.0 |
435 |
# just to make sure we get a clean state we will sleep a bit |
436 |
sleep 1 |
|
437 |
set exit_code [allocate_and_quit $inode_name $inode_procs $inode_procs] |
|
|
1.3.7
by Gennaro Oliva
Import upstream version 2.1.0 |
438 |
} |
439 |
if {$exit_code == 0} {
|
|
440 |
send_user "\nSUCCESS\n" |
|
441 |
} else {
|
|
442 |
exit $exit_code |
|
443 |
} |