Mercurial > hg > cc > cirrus_work

annotate bin/_runme.sh @ 40:316495371bbc

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
add lots more debugging output, reinstate -t for thread allocation
author Henry S. Thompson <ht@inf.ed.ac.uk>
date Wed, 05 Jul 2023 15:12:07 +0100
parents e3c440666f1a
children 235004978b22
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
2
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
1 #!/bin/bash
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
2 # This runs on the compute nodes...
40
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
3 # Args: wd [-d] [-t nthreads] [-b CMDS] [-i input] CMDS
38
e3c440666f1a doc pointer
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 2
diff changeset
4 # See ug4/azure/notes.txt for documentation
2
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
5
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
6 N=$SLURM_JOB_NUM_NODES
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
7 n=$SLURM_NTASKS
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
8 c=$SLURM_CPUS_PER_TASK
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
9 nodename=$SLURMD_NODENAME
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
10 local=$SLURM_LOCALID
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
11 node=$SLURM_NODEID
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
12 task=$SLURM_PROCID
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
13
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
14 cd "$1"
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
15 shift
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
16
40
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
17 if [ "$1" = "-d" ]
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
18 then
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
19 shift
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
20 debug=1
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
21 fi
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
22
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
23 if [ "$1" = "-t" ]
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
24 then
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
25 shift
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
26 threadsPerTask=$1
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
27 shift
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
28 else
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
29 threadsPerTask=2
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
30 fi
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
31
2
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
32 pjobs=$((c / $threadsPerTask))
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
33
40
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
34 echo $(date) task $n.$task on $nodename:$N.$node start $pjobs jobs 1>&2
2
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
35
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
36 PATH=$W/$USER/bin:$W/shared/bin:$PATH
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
37 export task PATH n
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
38
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
39 if [ "$1" = "-b" ]
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
40 then
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
41 shift
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
42 eval "$1"
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
43 shift
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
44 fi
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
45
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
46 if [ "$1" = "-i" ]
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
47 then
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
48 shift
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
49 input="$1"
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
50 shift
40
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
51 if [ "$debug" ]
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
52 then
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
53 echo $(date) task $n.$task input "|$input|"
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
54 fi
2
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
55 fi
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
56
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
57 export cmd="$1"
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
58 shift
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
59
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
60 doit () {
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
61 arg="$1"
40
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
62 echo $(date) start $task $PARALLEL_SEQ "|$cmd|$arg|"
2
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
63 eval "$cmd"
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
64 echo $(date) end $task $PARALLEL_SEQ
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
65 }
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
66
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
67 export -f doit
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
68
40
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
69 if [ "$debug" ]
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
70 then
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
71 echo $(date) task $n.$task cmd "|$cmd|" doit $(type doit)
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
72 fi
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
73
2
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
74 eval "$input" | \
40
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
75 if [ "$debug" ]
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
76 then
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
77 tee >(cat 1>&2)
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
78 else
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
79 cat
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
80 fi | \
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
81 if [ $pjobs -le 1 ]
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
82 then
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
83 xargs -I ^ -n 1 bash -c "doit ^"
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
84 else
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
85 parallel --line-buffer -j $pjobs doit '{}'
316495371bbc add lots more debugging output,
Henry S. Thompson <ht@inf.ed.ac.uk>
parents: 38
diff changeset
86 fi
2
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
87
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
88 echo $(date) task $n.$task on $nodename:$N.$node end 1>&2
b4801f5696b2 compute node workers, see cirrus_home/bin repo for login node masters
Henry S. Thompson <ht@inf.ed.ac.uk>
parents:
diff changeset
89