tange/parallel
1
0
Fork 0
mirror of https://git.savannah.gnu.org/git/parallel.git synced 2024-11-22 05:57:54 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity
parallel/src/parallel_alternatives.pod
Ole Tange b5b3d5dc3e release procedure updated. 
parallel: --plus --onall now works.
parallel: --blocktimeout must be >= 1.
2020-05-31 16:42:04 +02:00

3028 lines
87 KiB
Plaintext
Raw Permalink Blame History

#!/usr/bin/perl -w
=encoding utf8
=head1 NAME
parallel_alternatives - Alternatives to GNU B<parallel>
=head1 DIFFERENCES BETWEEN GNU Parallel AND ALTERNATIVES
There are a lot programs with some of the functionality of GNU
B<parallel>. GNU B<parallel> strives to include the best of the
functionality without sacrificing ease of use.
B<parallel> has existed since 2002 and as GNU B<parallel> since
2010. A lot of the alternatives have not had the vitality to survive
that long, but have come and gone during that time.
GNU B<parallel> is actively maintained with a new release every month
since 2010. Most other alternatives are fleeting interests of the
developers with irregular releases and only maintained for a few
years.
=head2 SUMMARY TABLE
The following features are in some of the comparable tools:
B<Inputs>
I1. Arguments can be read from stdin
I2. Arguments can be read from a file
I3. Arguments can be read from multiple files
I4. Arguments can be read from command line
I5. Arguments can be read from a table
I6. Arguments can be read from the same file using #! (shebang)
I7. Line oriented input as default (Quoting of special chars not needed)
B<Manipulation of input>
M1. Composed command
M2. Multiple arguments can fill up an execution line
M3. Arguments can be put anywhere in the execution line
M4. Multiple arguments can be put anywhere in the execution line
M5. Arguments can be replaced with context
M6. Input can be treated as the complete command line
B<Outputs>
O1. Grouping output so output from different jobs do not mix
O2. Send stderr (standard error) to stderr (standard error)
O3. Send stdout (standard output) to stdout (standard output)
O4. Order of output can be same as order of input
O5. Stdout only contains stdout (standard output) from the command
O6. Stderr only contains stderr (standard error) from the command
O7. Buffering on disk
O8. Cleanup of file if killed
O9. Test if disk runs full during run
O10. Output of a line bigger than 4 GB
B<Execution>
E1. Running jobs in parallel
E2. List running jobs
E3. Finish running jobs, but do not start new jobs
E4. Number of running jobs can depend on number of cpus
E5. Finish running jobs, but do not start new jobs after first failure
E6. Number of running jobs can be adjusted while running
E7. Only spawn new jobs if load is less than a limit
B<Remote execution>
R1. Jobs can be run on remote computers
R2. Basefiles can be transferred
R3. Argument files can be transferred
R4. Result files can be transferred
R5. Cleanup of transferred files
R6. No config files needed
R7. Do not run more than SSHD's MaxStartups can handle
R8. Configurable SSH command
R9. Retry if connection breaks occasionally
B<Semaphore>
S1. Possibility to work as a mutex
S2. Possibility to work as a counting semaphore
B<Legend>
- = no
x = not applicable
ID = yes
As every new version of the programs are not tested the table may be
outdated. Please file a bug-report if you find errors (See REPORTING
BUGS).
parallel:
I1 I2 I3 I4 I5 I6 I7
M1 M2 M3 M4 M5 M6
O1 O2 O3 O4 O5 O6 O7 O8 O9 O10
E1 E2 E3 E4 E5 E6 E7
R1 R2 R3 R4 R5 R6 R7 R8 R9
S1 S2
find -exec:
- - - x - x -
- M2 M3 - - - -
- O2 O3 O4 O5 O6
- - - - - - -
- - - - - - - - -
x x
make -j:
- - - - - - -
- - - - - -
O1 O2 O3 - x O6
E1 - - - E5 -
- - - - - - - - -
- -
=head2 DIFFERENCES BETWEEN xargs AND GNU Parallel
Summary table (see legend above):
I1 I2 - - - - -
- M2 M3 - - -
- O2 O3 - O5 O6
E1 - - - - - -
- - - - - x - - -
- -
B<xargs> offers some of the same possibilities as GNU B<parallel>.
B<xargs> deals badly with special characters (such as space, \, ' and
"). To see the problem try this:
touch important_file
touch 'not important_file'
ls not* | xargs rm
mkdir -p "My brother's 12\" records"
ls | xargs rmdir
touch 'c:\windows\system32\clfs.sys'
echo 'c:\windows\system32\clfs.sys' | xargs ls -l
You can specify B<-0>, but many input generators are not optimized for
using B<NUL> as separator but are optimized for B<newline> as
separator. E.g. B<awk>, B<ls>, B<echo>, B<tar -v>, B<head> (requires
using B<-z>), B<tail> (requires using B<-z>), B<sed> (requires using
B<-z>), B<perl> (B<-0> and \0 instead of \n), B<locate> (requires
using B<-0>), B<find> (requires using B<-print0>), B<grep> (requires
using B<-z> or B<-Z>), B<sort> (requires using B<-z>).
GNU B<parallel>'s newline separation can be emulated with:
B<cat | xargs -d "\n" -n1 I<command>>
B<xargs> can run a given number of jobs in parallel, but has no
support for running number-of-cpu-cores jobs in parallel.
B<xargs> has no support for grouping the output, therefore output may
run together, e.g. the first half of a line is from one process and
the last half of the line is from another process. The example
B<Parallel grep> cannot be done reliably with B<xargs> because of
this. To see this in action try:
parallel perl -e '\$a=\"1\".\"{}\"x10000000\;print\ \$a,\"\\n\"' \
'>' {} ::: a b c d e f g h
# Serial = no mixing = the wanted result
# 'tr -s a-z' squeezes repeating letters into a single letter
echo a b c d e f g h | xargs -P1 -n1 grep 1 | tr -s a-z
# Compare to 8 jobs in parallel
parallel -kP8 -n1 grep 1 ::: a b c d e f g h | tr -s a-z
echo a b c d e f g h | xargs -P8 -n1 grep 1 | tr -s a-z
echo a b c d e f g h | xargs -P8 -n1 grep --line-buffered 1 | \
tr -s a-z
Or try this:
slow_seq() {
echo Count to "$@"
seq "$@" |
perl -ne '$|=1; for(split//){ print; select($a,$a,$a,0.100);}'
}
export -f slow_seq
# Serial = no mixing = the wanted result
seq 8 | xargs -n1 -P1 -I {} bash -c 'slow_seq {}'
# Compare to 8 jobs in parallel
seq 8 | parallel -P8 slow_seq {}
seq 8 | xargs -n1 -P8 -I {} bash -c 'slow_seq {}'
B<xargs> has no support for keeping the order of the output, therefore
if running jobs in parallel using B<xargs> the output of the second
job cannot be postponed till the first job is done.
B<xargs> has no support for running jobs on remote computers.
B<xargs> has no support for context replace, so you will have to create the
arguments.
If you use a replace string in B<xargs> (B<-I>) you can not force
B<xargs> to use more than one argument.
Quoting in B<xargs> works like B<-q> in GNU B<parallel>. This means
composed commands and redirection require using B<bash -c>.
ls | parallel "wc {} >{}.wc"
ls | parallel "echo {}; ls {}|wc"
becomes (assuming you have 8 cores and that none of the filenames
contain space, " or ').
ls | xargs -d "\n" -P8 -I {} bash -c "wc {} >{}.wc"
ls | xargs -d "\n" -P8 -I {} bash -c "echo {}; ls {}|wc"
https://www.gnu.org/software/findutils/
=head2 DIFFERENCES BETWEEN find -exec AND GNU Parallel
B<find -exec> offers some of the same possibilities as GNU B<parallel>.
B<find -exec> only works on files. Processing other input (such as
hosts or URLs) will require creating these inputs as files. B<find
-exec> has no support for running commands in parallel.
https://www.gnu.org/software/findutils/ (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN make -j AND GNU Parallel
B<make -j> can run jobs in parallel, but requires a crafted Makefile
to do this. That results in extra quoting to get filenames containing
newlines to work correctly.
B<make -j> computes a dependency graph before running jobs. Jobs run
by GNU B<parallel> does not depend on each other.
(Very early versions of GNU B<parallel> were coincidentally implemented
using B<make -j>).
https://www.gnu.org/software/make/ (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN ppss AND GNU Parallel
Summary table (see legend above):
I1 I2 - - - - I7
M1 - M3 - - M6
O1 - - x - -
E1 E2 ?E3 E4 - - -
R1 R2 R3 R4 - - ?R7 ? ?
- -
B<ppss> is also a tool for running jobs in parallel.
The output of B<ppss> is status information and thus not useful for
using as input for another command. The output from the jobs are put
into files.
The argument replace string ($ITEM) cannot be changed. Arguments must
be quoted - thus arguments containing special characters (space '"&!*)
may cause problems. More than one argument is not supported. Filenames
containing newlines are not processed correctly. When reading input
from a file null cannot be used as a terminator. B<ppss> needs to read
the whole input file before starting any jobs.
Output and status information is stored in ppss_dir and thus requires
cleanup when completed. If the dir is not removed before running
B<ppss> again it may cause nothing to happen as B<ppss> thinks the
task is already done. GNU B<parallel> will normally not need cleaning
up if running locally and will only need cleaning up if stopped
abnormally and running remote (B<--cleanup> may not complete if
stopped abnormally). The example B<Parallel grep> would require extra
postprocessing if written using B<ppss>.
For remote systems PPSS requires 3 steps: config, deploy, and
start. GNU B<parallel> only requires one step.
=head3 EXAMPLES FROM ppss MANUAL
Here are the examples from B<ppss>'s manual page with the equivalent
using GNU B<parallel>:
1$ ./ppss.sh standalone -d /path/to/files -c 'gzip '
1$ find /path/to/files -type f | parallel gzip
2$ ./ppss.sh standalone -d /path/to/files -c 'cp "$ITEM" /destination/dir '
2$ find /path/to/files -type f | parallel cp {} /destination/dir
3$ ./ppss.sh standalone -f list-of-urls.txt -c 'wget -q '
3$ parallel -a list-of-urls.txt wget -q
4$ ./ppss.sh standalone -f list-of-urls.txt -c 'wget -q "$ITEM"'
4$ parallel -a list-of-urls.txt wget -q {}
5$ ./ppss config -C config.cfg -c 'encode.sh ' -d /source/dir \
-m 192.168.1.100 -u ppss -k ppss-key.key -S ./encode.sh \
-n nodes.txt -o /some/output/dir --upload --download;
./ppss deploy -C config.cfg
./ppss start -C config
5$ # parallel does not use configs. If you want a different username put it in nodes.txt: user@hostname
find source/dir -type f |
parallel --sshloginfile nodes.txt --trc {.}.mp3 lame -a {} -o {.}.mp3 --preset standard --quiet
6$ ./ppss stop -C config.cfg
6$ killall -TERM parallel
7$ ./ppss pause -C config.cfg
7$ Press: CTRL-Z or killall -SIGTSTP parallel
8$ ./ppss continue -C config.cfg
8$ Enter: fg or killall -SIGCONT parallel
9$ ./ppss.sh status -C config.cfg
9$ killall -SIGUSR2 parallel
https://github.com/louwrentius/PPSS
=head2 DIFFERENCES BETWEEN pexec AND GNU Parallel
Summary table (see legend above):
I1 I2 - I4 I5 - -
M1 - M3 - - M6
O1 O2 O3 - O5 O6
E1 - - E4 - E6 -
R1 - - - - R6 - - -
S1 -
B<pexec> is also a tool for running jobs in parallel.
=head3 EXAMPLES FROM pexec MANUAL
Here are the examples from B<pexec>'s info page with the equivalent
using GNU B<parallel>:
1$ pexec -o sqrt-%s.dat -p "$(seq 10)" -e NUM -n 4 -c -- \
'echo "scale=10000;sqrt($NUM)" | bc'
1$ seq 10 | parallel -j4 'echo "scale=10000;sqrt({})" | \
bc > sqrt-{}.dat'
2$ pexec -p "$(ls myfiles*.ext)" -i %s -o %s.sort -- sort
2$ ls myfiles*.ext | parallel sort {} ">{}.sort"
3$ pexec -f image.list -n auto -e B -u star.log -c -- \
'fistar $B.fits -f 100 -F id,x,y,flux -o $B.star'
3$ parallel -a image.list \
'fistar {}.fits -f 100 -F id,x,y,flux -o {}.star' 2>star.log
4$ pexec -r *.png -e IMG -c -o - -- \
'convert $IMG ${IMG%.png}.jpeg ; "echo $IMG: done"'
4$ ls *.png | parallel 'convert {} {.}.jpeg; echo {}: done'
5$ pexec -r *.png -i %s -o %s.jpg -c 'pngtopnm | pnmtojpeg'
5$ ls *.png | parallel 'pngtopnm < {} | pnmtojpeg > {}.jpg'
6$ for p in *.png ; do echo ${p%.png} ; done | \
pexec -f - -i %s.png -o %s.jpg -c 'pngtopnm | pnmtojpeg'
6$ ls *.png | parallel 'pngtopnm < {} | pnmtojpeg > {.}.jpg'
7$ LIST=$(for p in *.png ; do echo ${p%.png} ; done)
pexec -r $LIST -i %s.png -o %s.jpg -c 'pngtopnm | pnmtojpeg'
7$ ls *.png | parallel 'pngtopnm < {} | pnmtojpeg > {.}.jpg'
8$ pexec -n 8 -r *.jpg -y unix -e IMG -c \
'pexec -j -m blockread -d $IMG | \
jpegtopnm | pnmscale 0.5 | pnmtojpeg | \
pexec -j -m blockwrite -s th_$IMG'
8$ # Combining GNU B<parallel> and GNU B<sem>.
ls *jpg | parallel -j8 'sem --id blockread cat {} | jpegtopnm |' \
'pnmscale 0.5 | pnmtojpeg | sem --id blockwrite cat > th_{}'
# If reading and writing is done to the same disk, this may be
# faster as only one process will be either reading or writing:
ls *jpg | parallel -j8 'sem --id diskio cat {} | jpegtopnm |' \
'pnmscale 0.5 | pnmtojpeg | sem --id diskio cat > th_{}'
https://www.gnu.org/software/pexec/
=head2 DIFFERENCES BETWEEN xjobs AND GNU Parallel
B<xjobs> is also a tool for running jobs in parallel. It only supports
running jobs on your local computer.
B<xjobs> deals badly with special characters just like B<xargs>. See
the section B<DIFFERENCES BETWEEN xargs AND GNU Parallel>.
=head3 EXAMPLES FROM xjobs MANUAL
Here are the examples from B<xjobs>'s man page with the equivalent
using GNU B<parallel>:
1$ ls -1 *.zip | xjobs unzip
1$ ls *.zip | parallel unzip
2$ ls -1 *.zip | xjobs -n unzip
2$ ls *.zip | parallel unzip >/dev/null
3$ find . -name '*.bak' | xjobs gzip
3$ find . -name '*.bak' | parallel gzip
4$ ls -1 *.jar | sed 's/\(.*\)/\1 > \1.idx/' | xjobs jar tf
4$ ls *.jar | parallel jar tf {} '>' {}.idx
5$ xjobs -s script
5$ cat script | parallel
6$ mkfifo /var/run/my_named_pipe;
xjobs -s /var/run/my_named_pipe &
echo unzip 1.zip >> /var/run/my_named_pipe;
echo tar cf /backup/myhome.tar /home/me >> /var/run/my_named_pipe
6$ mkfifo /var/run/my_named_pipe;
cat /var/run/my_named_pipe | parallel &
echo unzip 1.zip >> /var/run/my_named_pipe;
echo tar cf /backup/myhome.tar /home/me >> /var/run/my_named_pipe
http://www.maier-komor.de/xjobs.html (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN prll AND GNU Parallel
B<prll> is also a tool for running jobs in parallel. It does not
support running jobs on remote computers.
B<prll> encourages using BASH aliases and BASH functions instead of
scripts. GNU B<parallel> supports scripts directly, functions if they
are exported using B<export -f>, and aliases if using B<env_parallel>.
B<prll> generates a lot of status information on stderr (standard
error) which makes it harder to use the stderr (standard error) output
of the job directly as input for another program.
=head3 EXAMPLES FROM prll's MANUAL
Here is the example from B<prll>'s man page with the equivalent
using GNU B<parallel>:
1$ prll -s 'mogrify -flip $1' *.jpg
1$ parallel mogrify -flip ::: *.jpg
https://github.com/exzombie/prll (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN dxargs AND GNU Parallel
B<dxargs> is also a tool for running jobs in parallel.
B<dxargs> does not deal well with more simultaneous jobs than SSHD's
MaxStartups. B<dxargs> is only built for remote run jobs, but does not
support transferring of files.
https://web.archive.org/web/20120518070250/http://www.
semicomplete.com/blog/geekery/distributed-xargs.html (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN mdm/middleman AND GNU Parallel
middleman(mdm) is also a tool for running jobs in parallel.
=head3 EXAMPLES FROM middleman's WEBSITE
Here are the shellscripts of
https://web.archive.org/web/20110728064735/http://mdm.
berlios.de/usage.html ported to GNU B<parallel>:
1$ seq 19 | parallel buffon -o - | sort -n > result
cat files | parallel cmd
find dir -execdir sem cmd {} \;
https://github.com/cklin/mdm (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN xapply AND GNU Parallel
B<xapply> can run jobs in parallel on the local computer.
=head3 EXAMPLES FROM xapply's MANUAL
Here are the examples from B<xapply>'s man page with the equivalent
using GNU B<parallel>:
1$ xapply '(cd %1 && make all)' */
1$ parallel 'cd {} && make all' ::: */
2$ xapply -f 'diff %1 ../version5/%1' manifest | more
2$ parallel diff {} ../version5/{} < manifest | more
3$ xapply -p/dev/null -f 'diff %1 %2' manifest1 checklist1
3$ parallel --link diff {1} {2} :::: manifest1 checklist1
4$ xapply 'indent' *.c
4$ parallel indent ::: *.c
5$ find ~ksb/bin -type f ! -perm -111 -print | \
xapply -f -v 'chmod a+x' -
5$ find ~ksb/bin -type f ! -perm -111 -print | \
parallel -v chmod a+x
6$ find */ -... | fmt 960 1024 | xapply -f -i /dev/tty 'vi' -
6$ sh <(find */ -... | parallel -s 1024 echo vi)
6$ find */ -... | parallel -s 1024 -Xuj1 vi
7$ find ... | xapply -f -5 -i /dev/tty 'vi' - - - - -
7$ sh <(find ... | parallel -n5 echo vi)
7$ find ... | parallel -n5 -uj1 vi
8$ xapply -fn "" /etc/passwd
8$ parallel -k echo < /etc/passwd
9$ tr ':' '\012' < /etc/passwd | \
xapply -7 -nf 'chown %1 %6' - - - - - - -
9$ tr ':' '\012' < /etc/passwd | parallel -N7 chown {1} {6}
10$ xapply '[ -d %1/RCS ] || echo %1' */
10$ parallel '[ -d {}/RCS ] || echo {}' ::: */
11$ xapply -f '[ -f %1 ] && echo %1' List | ...
11$ parallel '[ -f {} ] && echo {}' < List | ...
https://web.archive.org/web/20160702211113/
http://carrera.databits.net/~ksb/msrc/local/bin/xapply/xapply.html
=head2 DIFFERENCES BETWEEN AIX apply AND GNU Parallel
B<apply> can build command lines based on a template and arguments -
very much like GNU B<parallel>. B<apply> does not run jobs in
parallel. B<apply> does not use an argument separator (like B<:::>);
instead the template must be the first argument.
=head3 EXAMPLES FROM IBM's KNOWLEDGE CENTER
Here are the examples from IBM's Knowledge Center and the
corresponding command using GNU B<parallel>:
=head4 To obtain results similar to those of the B<ls> command, enter:
1$ apply echo *
1$ parallel echo ::: *
=head4 To compare the file named a1 to the file named b1, and
the file named a2 to the file named b2, enter:
2$ apply -2 cmp a1 b1 a2 b2
2$ parallel -N2 cmp ::: a1 b1 a2 b2
=head4 To run the B<who> command five times, enter:
3$ apply -0 who 1 2 3 4 5
3$ parallel -N0 who ::: 1 2 3 4 5
=head4 To link all files in the current directory to the directory
/usr/joe, enter:
4$ apply 'ln %1 /usr/joe' *
4$ parallel ln {} /usr/joe ::: *
https://www-01.ibm.com/support/knowledgecenter/
ssw_aix_71/com.ibm.aix.cmds1/apply.htm (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN paexec AND GNU Parallel
B<paexec> can run jobs in parallel on both the local and remote computers.
B<paexec> requires commands to print a blank line as the last
output. This means you will have to write a wrapper for most programs.
B<paexec> has a job dependency facility so a job can depend on another
job to be executed successfully. Sort of a poor-man's B<make>.
=head3 EXAMPLES FROM paexec's EXAMPLE CATALOG
Here are the examples from B<paexec>'s example catalog with the equivalent
using GNU B<parallel>:
=head4 1_div_X_run
1$ ../../paexec -s -l -c "`pwd`/1_div_X_cmd" -n +1 <<EOF [...]
1$ parallel echo {} '|' `pwd`/1_div_X_cmd <<EOF [...]
=head4 all_substr_run
2$ ../../paexec -lp -c "`pwd`/all_substr_cmd" -n +3 <<EOF [...]
2$ parallel echo {} '|' `pwd`/all_substr_cmd <<EOF [...]
=head4 cc_wrapper_run
3$ ../../paexec -c "env CC=gcc CFLAGS=-O2 `pwd`/cc_wrapper_cmd" \
-n 'host1 host2' \
-t '/usr/bin/ssh -x' <<EOF [...]
3$ parallel echo {} '|' "env CC=gcc CFLAGS=-O2 `pwd`/cc_wrapper_cmd" \
-S host1,host2 <<EOF [...]
# This is not exactly the same, but avoids the wrapper
parallel gcc -O2 -c -o {.}.o {} \
-S host1,host2 <<EOF [...]
=head4 toupper_run
4$ ../../paexec -lp -c "`pwd`/toupper_cmd" -n +10 <<EOF [...]
4$ parallel echo {} '|' ./toupper_cmd <<EOF [...]
# Without the wrapper:
parallel echo {} '| awk {print\ toupper\(\$0\)}' <<EOF [...]
https://github.com/cheusov/paexec
=head2 DIFFERENCES BETWEEN map(sitaramc) AND GNU Parallel
Summary table (see legend above):
I1 - - I4 - - (I7)
M1 (M2) M3 (M4) M5 M6
- O2 O3 - O5 - - N/A N/A O10
E1 - - - - - -
- - - - - - - - -
- -
(I7): Only under special circumstances. See below.
(M2+M4): Only if there is a single replacement string.
B<map> rejects input with special characters:
echo "The Cure" > My\ brother\'s\ 12\"\ records
ls | map 'echo %; wc %'
It works with GNU B<parallel>:
ls | parallel 'echo {}; wc {}'
Under some circumstances it also works with B<map>:
ls | map 'echo % works %'
But tiny changes make it reject the input with special characters:
ls | map 'echo % does not work "%"'
This means that many UTF-8 characters will be rejected. This is by
design. From the web page: "As such, programs that I<quietly handle
them, with no warnings at all,> are doing their users a disservice."
B<map> delays each job by 0.01 s. This can be emulated by using
B<parallel --delay 0.01>.
B<map> prints '+' on stderr when a job starts, and '-' when a job
finishes. This cannot be disabled. B<parallel> has B<--bar> if you
need to see progress.
B<map>'s replacement strings (% %D %B %E) can be simulated in GNU
B<parallel> by putting this in B<~/.parallel/config>:
--rpl '%'
--rpl '%D $_=Q(::dirname($_));'
--rpl '%B s:.*/::;s:\.[^/.]+$::;'
--rpl '%E s:.*\.::'
B<map> does not have an argument separator on the command line, but
uses the first argument as command. This makes quoting harder which again
may affect readability. Compare:
map -p 2 'perl -ne '"'"'/^\S+\s+\S+$/ and print $ARGV,"\n"'"'" *
parallel -q perl -ne '/^\S+\s+\S+$/ and print $ARGV,"\n"' ::: *
B<map> can do multiple arguments with context replace, but not without
context replace:
parallel --xargs echo 'BEGIN{'{}'}END' ::: 1 2 3
map "echo 'BEGIN{'%'}END'" 1 2 3
B<map> has no support for grouping. So this gives the wrong results:
parallel perl -e '\$a=\"1{}\"x10000000\;print\ \$a,\"\\n\"' '>' {} \
::: a b c d e f
ls -l a b c d e f
parallel -kP4 -n1 grep 1 ::: a b c d e f > out.par
map -n1 -p 4 'grep 1' a b c d e f > out.map-unbuf
map -n1 -p 4 'grep --line-buffered 1' a b c d e f > out.map-linebuf
map -n1 -p 1 'grep --line-buffered 1' a b c d e f > out.map-serial
ls -l out*
md5sum out*
=head3 EXAMPLES FROM map's WEBSITE
Here are the examples from B<map>'s web page with the equivalent using
GNU B<parallel>:
1$ ls *.gif | map convert % %B.png # default max-args: 1
1$ ls *.gif | parallel convert {} {.}.png
2$ map "mkdir %B; tar -C %B -xf %" *.tgz # default max-args: 1
2$ parallel 'mkdir {.}; tar -C {.} -xf {}' ::: *.tgz
3$ ls *.gif | map cp % /tmp # default max-args: 100
3$ ls *.gif | parallel -X cp {} /tmp
4$ ls *.tar | map -n 1 tar -xf %
4$ ls *.tar | parallel tar -xf
5$ map "cp % /tmp" *.tgz
5$ parallel cp {} /tmp ::: *.tgz
6$ map "du -sm /home/%/mail" alice bob carol
6$ parallel "du -sm /home/{}/mail" ::: alice bob carol
or if you prefer running a single job with multiple args:
6$ parallel -Xj1 "du -sm /home/{}/mail" ::: alice bob carol
7$ cat /etc/passwd | map -d: 'echo user %1 has shell %7'
7$ cat /etc/passwd | parallel --colsep : 'echo user {1} has shell {7}'
8$ export MAP_MAX_PROCS=$(( `nproc` / 2 ))
8$ export PARALLEL=-j50%
https://github.com/sitaramc/map (Last checked: 2020-05)
=head2 DIFFERENCES BETWEEN ladon AND GNU Parallel
B<ladon> can run multiple jobs on files in parallel.
B<ladon> only works on files and the only way to specify files is
using a quoted glob string (such as \*.jpg). It is not possible to
list the files manually.
As replacement strings it uses FULLPATH DIRNAME BASENAME EXT RELDIR
RELPATH
These can be simulated using GNU B<parallel> by putting this in
B<~/.parallel/config>:
--rpl 'FULLPATH $_=Q($_);chomp($_=qx{readlink -f $_});'
--rpl 'DIRNAME $_=Q(::dirname($_));chomp($_=qx{readlink -f $_});'
--rpl 'BASENAME s:.*/::;s:\.[^/.]+$::;'
--rpl 'EXT s:.*\.::'
--rpl 'RELDIR $_=Q($_);chomp(($_,$c)=qx{readlink -f $_;pwd});
s:\Q$c/\E::;$_=::dirname($_);'
--rpl 'RELPATH $_=Q($_);chomp(($_,$c)=qx{readlink -f $_;pwd});
s:\Q$c/\E::;'
B<ladon> deals badly with filenames containing " and newline, and it
fails for output larger than 200k:
ladon '*' -- seq 36000 | wc
=head3 EXAMPLES FROM ladon MANUAL
It is assumed that the '--rpl's above are put in B<~/.parallel/config>
and that it is run under a shell that supports '**' globbing (such as B<zsh>):
1$ ladon "**/*.txt" -- echo RELPATH
1$ parallel echo RELPATH ::: **/*.txt
2$ ladon "~/Documents/**/*.pdf" -- shasum FULLPATH >hashes.txt
2$ parallel shasum FULLPATH ::: ~/Documents/**/*.pdf >hashes.txt
3$ ladon -m thumbs/RELDIR "**/*.jpg" -- convert FULLPATH \
-thumbnail 100x100^ -gravity center -extent 100x100 \
thumbs/RELPATH
3$ parallel mkdir -p thumbs/RELDIR\; convert FULLPATH
-thumbnail 100x100^ -gravity center -extent 100x100 \
thumbs/RELPATH ::: **/*.jpg
4$ ladon "~/Music/*.wav" -- lame -V 2 FULLPATH DIRNAME/BASENAME.mp3
4$ parallel lame -V 2 FULLPATH DIRNAME/BASENAME.mp3 ::: ~/Music/*.wav
https://github.com/danielgtaylor/ladon (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN jobflow AND GNU Parallel
B<jobflow> can run multiple jobs in parallel.
Just like B<xargs> output from B<jobflow> jobs running in parallel mix
together by default. B<jobflow> can buffer into files (placed in
/run/shm), but these are not cleaned up if B<jobflow> dies
unexpectedly (e.g. by Ctrl-C). If the total output is big (in the
order of RAM+swap) it can cause the system to slow to a crawl and
eventually run out of memory.
B<jobflow> gives no error if the command is unknown, and like B<xargs>
redirection and composed commands require wrapping with B<bash -c>.
Input lines can at most be 4096 bytes. You can at most have 16 {}'s in
the command template. More than that either crashes the program or
simple does not execute the command.
B<jobflow> has no equivalent for B<--pipe>, or B<--sshlogin>.
B<jobflow> makes it possible to set resource limits on the running
jobs. This can be emulated by GNU B<parallel> using B<bash>'s B<ulimit>:
jobflow -limits=mem=100M,cpu=3,fsize=20M,nofiles=300 myjob
parallel 'ulimit -v 102400 -t 3 -f 204800 -n 300 myjob'
=head3 EXAMPLES FROM jobflow README
1$ cat things.list | jobflow -threads=8 -exec ./mytask {}
1$ cat things.list | parallel -j8 ./mytask {}
2$ seq 100 | jobflow -threads=100 -exec echo {}
2$ seq 100 | parallel -j100 echo {}
3$ cat urls.txt | jobflow -threads=32 -exec wget {}
3$ cat urls.txt | parallel -j32 wget {}
4$ find . -name '*.bmp' | \
jobflow -threads=8 -exec bmp2jpeg {.}.bmp {.}.jpg
4$ find . -name '*.bmp' | \
parallel -j8 bmp2jpeg {.}.bmp {.}.jpg
https://github.com/rofl0r/jobflow
=head2 DIFFERENCES BETWEEN gargs AND GNU Parallel
B<gargs> can run multiple jobs in parallel.
Older versions cache output in memory. This causes it to be extremely
slow when the output is larger than the physical RAM, and can cause
the system to run out of memory.
See more details on this in B<man parallel_design>.
Newer versions cache output in files, but leave files in $TMPDIR if it
is killed.
Output to stderr (standard error) is changed if the command fails.
=head3 EXAMPLES FROM gargs WEBSITE
1$ seq 12 -1 1 | gargs -p 4 -n 3 "sleep {0}; echo {1} {2}"
1$ seq 12 -1 1 | parallel -P 4 -n 3 "sleep {1}; echo {2} {3}"
2$ cat t.txt | gargs --sep "\s+" \
-p 2 "echo '{0}:{1}-{2}' full-line: \'{}\'"
2$ cat t.txt | parallel --colsep "\\s+" \
-P 2 "echo '{1}:{2}-{3}' full-line: \'{}\'"
https://github.com/brentp/gargs
=head2 DIFFERENCES BETWEEN orgalorg AND GNU Parallel
B<orgalorg> can run the same job on multiple machines. This is related
to B<--onall> and B<--nonall>.
B<orgalorg> supports entering the SSH password - provided it is the
same for all servers. GNU B<parallel> advocates using B<ssh-agent>
instead, but it is possible to emulate B<orgalorg>'s behavior by
setting SSHPASS and by using B<--ssh "sshpass ssh">.
To make the emulation easier, make a simple alias:
alias par_emul="parallel -j0 --ssh 'sshpass ssh' --nonall --tag --lb"
If you want to supply a password run:
SSHPASS=`ssh-askpass`
or set the password directly:
SSHPASS=P4$$w0rd!
If the above is set up you can then do:
orgalorg -o frontend1 -o frontend2 -p -C uptime
par_emul -S frontend1 -S frontend2 uptime
orgalorg -o frontend1 -o frontend2 -p -C top -bid 1
par_emul -S frontend1 -S frontend2 top -bid 1
orgalorg -o frontend1 -o frontend2 -p -er /tmp -n \
'md5sum /tmp/bigfile' -S bigfile
par_emul -S frontend1 -S frontend2 --basefile bigfile \
--workdir /tmp md5sum /tmp/bigfile
B<orgalorg> has a progress indicator for the transferring of a
file. GNU B<parallel> does not.
https://github.com/reconquest/orgalorg
=head2 DIFFERENCES BETWEEN Rust parallel AND GNU Parallel
Rust parallel focuses on speed. It is almost as fast as B<xargs>. It
implements a few features from GNU B<parallel>, but lacks many
functions. All these fail:
# Read arguments from file
parallel -a file echo
# Changing the delimiter
parallel -d _ echo ::: a_b_c_
These do something different from GNU B<parallel>
# -q to protect quoted $ and space
parallel -q perl -e '$a=shift; print "$a"x10000000' ::: a b c
# Generation of combination of inputs
parallel echo {1} {2} ::: red green blue ::: S M L XL XXL
# {= perl expression =} replacement string
parallel echo '{= s/new/old/ =}' ::: my.new your.new
# --pipe
seq 100000 | parallel --pipe wc
# linked arguments
parallel echo ::: S M L :::+ sml med lrg ::: R G B :::+ red grn blu
# Run different shell dialects
zsh -c 'parallel echo \={} ::: zsh && true'
csh -c 'parallel echo \$\{\} ::: shell && true'
bash -c 'parallel echo \$\({}\) ::: pwd && true'
# Rust parallel does not start before the last argument is read
(seq 10; sleep 5; echo 2) | time parallel -j2 'sleep 2; echo'
tail -f /var/log/syslog | parallel echo
Most of the examples from the book GNU Parallel 2018 do not work, thus
Rust parallel is not close to being a compatible replacement.
Rust parallel has no remote facilities.
It uses /tmp/parallel for tmp files and does not clean up if
terminated abruptly. If another user on the system uses Rust parallel,
then /tmp/parallel will have the wrong permissions and Rust parallel
will fail. A malicious user can setup the right permissions and
symlink the output file to one of the user's files and next time the
user uses Rust parallel it will overwrite this file.
attacker$ mkdir /tmp/parallel
attacker$ chmod a+rwX /tmp/parallel
# Symlink to the file the attacker wants to zero out
attacker$ ln -s ~victim/.important-file /tmp/parallel/stderr_1
victim$ seq 1000 | parallel echo
# This file is now overwritten with stderr from 'echo'
victim$ cat ~victim/.important-file
If /tmp/parallel runs full during the run, Rust parallel does not
report this, but finishes with success - thereby risking data loss.
https://github.com/mmstick/parallel
=head2 DIFFERENCES BETWEEN Rush AND GNU Parallel
B<rush> (https://github.com/shenwei356/rush) is written in Go and
based on B<gargs>.
Just like GNU B<parallel> B<rush> buffers in temporary files. But
opposite GNU B<parallel> B<rush> does not clean up, if the process
dies abnormally.
B<rush> has some string manipulations that can be emulated by putting
this into ~/.parallel/config (/ is used instead of %, and % is used
instead of ^ as that is closer to bash's ${var%postfix}):
--rpl '{:} s:(\.[^/]+)*$::'
--rpl '{:%([^}]+?)} s:$$1(\.[^/]+)*$::'
--rpl '{/:%([^}]*?)} s:.*/(.*)$$1(\.[^/]+)*$:$1:'
--rpl '{/:} s:(.*/)?([^/.]+)(\.[^/]+)*$:$2:'
--rpl '{@(.*?)} /$$1/ and $_=$1;'
=head3 EXAMPLES FROM rush's WEBSITE
Here are the examples from B<rush>'s website with the equivalent
command in GNU B<parallel>.
B<1. Simple run, quoting is not necessary>
$ seq 1 3 | rush echo {}
$ seq 1 3 | parallel echo {}
B<2. Read data from file (`-i`)>
$ rush echo {} -i data1.txt -i data2.txt
$ cat data1.txt data2.txt | parallel echo {}
B<3. Keep output order (`-k`)>
$ seq 1 3 | rush 'echo {}' -k
$ seq 1 3 | parallel -k echo {}
B<4. Timeout (`-t`)>
$ time seq 1 | rush 'sleep 2; echo {}' -t 1
$ time seq 1 | parallel --timeout 1 'sleep 2; echo {}'
B<5. Retry (`-r`)>
$ seq 1 | rush 'python unexisted_script.py' -r 1
$ seq 1 | parallel --retries 2 'python unexisted_script.py'
Use B<-u> to see it is really run twice:
$ seq 1 | parallel -u --retries 2 'python unexisted_script.py'
B<6. Dirname (`{/}`) and basename (`{%}`) and remove custom
suffix (`{^suffix}`)>
$ echo dir/file_1.txt.gz | rush 'echo {/} {%} {^_1.txt.gz}'
$ echo dir/file_1.txt.gz |
parallel --plus echo {//} {/} {%_1.txt.gz}
B<7. Get basename, and remove last (`{.}`) or any (`{:}`) extension>
$ echo dir.d/file.txt.gz | rush 'echo {.} {:} {%.} {%:}'
$ echo dir.d/file.txt.gz | parallel 'echo {.} {:} {/.} {/:}'
B<8. Job ID, combine fields index and other replacement strings>
$ echo 12 file.txt dir/s_1.fq.gz |
rush 'echo job {#}: {2} {2.} {3%:^_1}'
$ echo 12 file.txt dir/s_1.fq.gz |
parallel --colsep ' ' 'echo job {#}: {2} {2.} {3/:%_1}'
B<9. Capture submatch using regular expression (`{@regexp}`)>
$ echo read_1.fq.gz | rush 'echo {@(.+)_\d}'
$ echo read_1.fq.gz | parallel 'echo {@(.+)_\d}'
B<10. Custom field delimiter (`-d`)>
$ echo a=b=c | rush 'echo {1} {2} {3}' -d =
$ echo a=b=c | parallel -d = echo {1} {2} {3}
B<11. Send multi-lines to every command (`-n`)>
$ seq 5 | rush -n 2 -k 'echo "{}"; echo'
$ seq 5 |
parallel -n 2 -k \
'echo {=-1 $_=join"\n",@arg[1..$#arg] =}; echo'
$ seq 5 | rush -n 2 -k 'echo "{}"; echo' -J ' '
$ seq 5 | parallel -n 2 -k 'echo {}; echo'
B<12. Custom record delimiter (`-D`), note that empty records are not used.>
$ echo a b c d | rush -D " " -k 'echo {}'
$ echo a b c d | parallel -d " " -k 'echo {}'
$ echo abcd | rush -D "" -k 'echo {}'
Cannot be done by GNU Parallel
$ cat fasta.fa
>seq1
tag
>seq2
cat
gat
>seq3
attac
a
cat
$ cat fasta.fa | rush -D ">" \
'echo FASTA record {#}: name: {1} sequence: {2}' -k -d "\n"
# rush fails to join the multiline sequences
$ cat fasta.fa | (read -n1 ignore_first_char;
parallel -d '>' --colsep '\n' echo FASTA record {#}: \
name: {1} sequence: '{=2 $_=join"",@arg[2..$#arg]=}'
)
B<13. Assign value to variable, like `awk -v` (`-v`)>
$ seq 1 |
rush 'echo Hello, {fname} {lname}!' -v fname=Wei -v lname=Shen
$ seq 1 |
parallel -N0 \
'fname=Wei; lname=Shen; echo Hello, ${fname} ${lname}!'
$ for var in a b; do \
$ seq 1 3 | rush -k -v var=$var 'echo var: {var}, data: {}'; \
$ done
In GNU B<parallel> you would typically do:
$ seq 1 3 | parallel -k echo var: {1}, data: {2} ::: a b :::: -
If you I<really> want the var:
$ seq 1 3 |
parallel -k var={1} ';echo var: $var, data: {}' ::: a b :::: -
If you I<really> want the B<for>-loop:
$ for var in a b; do
> export var;
> seq 1 3 | parallel -k 'echo var: $var, data: {}';
> done
Contrary to B<rush> this also works if the value is complex like:
My brother's 12" records
B<14. B<Preset variable> (`-v`), avoid repeatedly writing verbose replacement strings>
# naive way
$ echo read_1.fq.gz | rush 'echo {:^_1} {:^_1}_2.fq.gz'
$ echo read_1.fq.gz | parallel 'echo {:%_1} {:%_1}_2.fq.gz'
# macro + removing suffix
$ echo read_1.fq.gz |
rush -v p='{:^_1}' 'echo {p} {p}_2.fq.gz'
$ echo read_1.fq.gz |
parallel 'p={:%_1}; echo $p ${p}_2.fq.gz'
# macro + regular expression
$ echo read_1.fq.gz | rush -v p='{@(.+?)_\d}' 'echo {p} {p}_2.fq.gz'
$ echo read_1.fq.gz | parallel 'p={@(.+?)_\d}; echo $p ${p}_2.fq.gz'
Contrary to B<rush> GNU B<parallel> works with complex values:
echo "My brother's 12\"read_1.fq.gz" |
parallel 'p={@(.+?)_\d}; echo $p ${p}_2.fq.gz'
B<15. Interrupt jobs by `Ctrl-C`, rush will stop unfinished commands and exit.>
$ seq 1 20 | rush 'sleep 1; echo {}'
^C
$ seq 1 20 | parallel 'sleep 1; echo {}'
^C
B<16. Continue/resume jobs (`-c`). When some jobs failed (by
execution failure, timeout, or canceling by user with `Ctrl + C`),
please switch flag `-c/--continue` on and run again, so that `rush`
can save successful commands and ignore them in I<NEXT> run.>
$ seq 1 3 | rush 'sleep {}; echo {}' -t 3 -c
$ cat successful_cmds.rush
$ seq 1 3 | rush 'sleep {}; echo {}' -t 3 -c
$ seq 1 3 | parallel --joblog mylog --timeout 2 \
'sleep {}; echo {}'
$ cat mylog
$ seq 1 3 | parallel --joblog mylog --retry-failed \
'sleep {}; echo {}'
Multi-line jobs:
$ seq 1 3 | rush 'sleep {}; echo {}; \
echo finish {}' -t 3 -c -C finished.rush
$ cat finished.rush
$ seq 1 3 | rush 'sleep {}; echo {}; \
echo finish {}' -t 3 -c -C finished.rush
$ seq 1 3 |
parallel --joblog mylog --timeout 2 'sleep {}; echo {}; \
echo finish {}'
$ cat mylog
$ seq 1 3 |
parallel --joblog mylog --retry-failed 'sleep {}; echo {}; \
echo finish {}'
B<17. A comprehensive example: downloading 1K+ pages given by
three URL list files using `phantomjs save_page.js` (some page
contents are dynamically generated by Javascript, so `wget` does not
work). Here I set max jobs number (`-j`) as `20`, each job has a max
running time (`-t`) of `60` seconds and `3` retry changes
(`-r`). Continue flag `-c` is also switched on, so we can continue
unfinished jobs. Luckily, it's accomplished in one run :)>
$ for f in $(seq 2014 2016); do \
$ /bin/rm -rf $f; mkdir -p $f; \
$ cat $f.html.txt | rush -v d=$f -d = \
'phantomjs save_page.js "{}" > {d}/{3}.html' \
-j 20 -t 60 -r 3 -c; \
$ done
GNU B<parallel> can append to an existing joblog with '+':
$ rm mylog
$ for f in $(seq 2014 2016); do
/bin/rm -rf $f; mkdir -p $f;
cat $f.html.txt |
parallel -j20 --timeout 60 --retries 4 --joblog +mylog \
--colsep = \
phantomjs save_page.js {1}={2}={3} '>' $f/{3}.html
done
B<18. A bioinformatics example: mapping with `bwa`, and
processing result with `samtools`:>
$ ref=ref/xxx.fa
$ threads=25
$ ls -d raw.cluster.clean.mapping/* \
| rush -v ref=$ref -v j=$threads -v p='{}/{%}' \
'bwa mem -t {j} -M -a {ref} {p}_1.fq.gz {p}_2.fq.gz >{p}.sam;\
samtools view -bS {p}.sam > {p}.bam; \
samtools sort -T {p}.tmp -@ {j} {p}.bam -o {p}.sorted.bam; \
samtools index {p}.sorted.bam; \
samtools flagstat {p}.sorted.bam > {p}.sorted.bam.flagstat; \
/bin/rm {p}.bam {p}.sam;' \
-j 2 --verbose -c -C mapping.rush
GNU B<parallel> would use a function:
$ ref=ref/xxx.fa
$ export ref
$ thr=25
$ export thr
$ bwa_sam() {
p="$1"
bam="$p".bam
sam="$p".sam
sortbam="$p".sorted.bam
bwa mem -t $thr -M -a $ref ${p}_1.fq.gz ${p}_2.fq.gz > "$sam"
samtools view -bS "$sam" > "$bam"
samtools sort -T ${p}.tmp -@ $thr "$bam" -o "$sortbam"
samtools index "$sortbam"
samtools flagstat "$sortbam" > "$sortbam".flagstat
/bin/rm "$bam" "$sam"
}
$ export -f bwa_sam
$ ls -d raw.cluster.clean.mapping/* |
parallel -j 2 --verbose --joblog mylog bwa_sam
=head3 Other B<rush> features
B<rush> has:
=over 4
=item * B<awk -v> like custom defined variables (B<-v>)
With GNU B<parallel> you would simply set a shell variable:
parallel 'v={}; echo "$v"' ::: foo
echo foo | rush -v v={} 'echo {v}'
Also B<rush> does not like special chars. So these B<do not work>:
echo does not work | rush -v v=\" 'echo {v}'
echo "My brother's 12\" records" | rush -v v={} 'echo {v}'
Whereas the corresponding GNU B<parallel> version works:
parallel 'v=\"; echo "$v"' ::: works
parallel 'v={}; echo "$v"' ::: "My brother's 12\" records"
=item * Exit on first error(s) (-e)
This is called B<--halt now,fail=1> (or shorter: B<--halt 2>) when
used with GNU B<parallel>.
=item * Settable records sending to every command (B<-n>, default 1)
This is also called B<-n> in GNU B<parallel>.
=item * Practical replacement strings
=over 4
=item {:} remove any extension
With GNU B<parallel> this can be emulated by:
parallel --plus echo '{/\..*/}' ::: foo.ext.bar.gz
=item {^suffix}, remove suffix
With GNU B<parallel> this can be emulated by:
parallel --plus echo '{%.bar.gz}' ::: foo.ext.bar.gz
=item {@regexp}, capture submatch using regular expression
With GNU B<parallel> this can be emulated by:
parallel --rpl '{@(.*?)} /$$1/ and $_=$1;' \
echo '{@\d_(.*).gz}' ::: 1_foo.gz
=item {%.}, {%:}, basename without extension
With GNU B<parallel> this can be emulated by:
parallel echo '{= s:.*/::;s/\..*// =}' ::: dir/foo.bar.gz
And if you need it often, you define a B<--rpl> in
B<$HOME/.parallel/config>:
--rpl '{%.} s:.*/::;s/\..*//'
--rpl '{%:} s:.*/::;s/\..*//'
Then you can use them as:
parallel echo {%.} {%:} ::: dir/foo.bar.gz
=back
=item * Preset variable (macro)
E.g.
echo foosuffix | rush -v p={^suffix} 'echo {p}_new_suffix'
With GNU B<parallel> this can be emulated by:
echo foosuffix |
parallel --plus 'p={%suffix}; echo ${p}_new_suffix'
Opposite B<rush> GNU B<parallel> works fine if the input contains
double space, ' and ":
echo "1'6\" foosuffix" |
parallel --plus 'p={%suffix}; echo "${p}"_new_suffix'
=item * Commands of multi-lines
While you I<can> use multi-lined commands in GNU B<parallel>, to
improve readability GNU B<parallel> discourages the use of multi-line
commands. In most cases it can be written as a function:
seq 1 3 |
parallel --timeout 2 --joblog my.log 'sleep {}; echo {}; \
echo finish {}'
Could be written as:
doit() {
sleep "$1"
echo "$1"
echo finish "$1"
}
export -f doit
seq 1 3 | parallel --timeout 2 --joblog my.log doit
The failed commands can be resumed with:
seq 1 3 |
parallel --resume-failed --joblog my.log 'sleep {}; echo {};\
echo finish {}'
=back
https://github.com/shenwei356/rush
=head2 DIFFERENCES BETWEEN ClusterSSH AND GNU Parallel
ClusterSSH solves a different problem than GNU B<parallel>.
ClusterSSH opens a terminal window for each computer and using a
master window you can run the same command on all the computers. This
is typically used for administrating several computers that are almost
identical.
GNU B<parallel> runs the same (or different) commands with different
arguments in parallel possibly using remote computers to help
computing. If more than one computer is listed in B<-S> GNU B<parallel> may
only use one of these (e.g. if there are 8 jobs to be run and one
computer has 8 cores).
GNU B<parallel> can be used as a poor-man's version of ClusterSSH:
B<parallel --nonall -S server-a,server-b do_stuff foo bar>
https://github.com/duncs/clusterssh
=head2 DIFFERENCES BETWEEN coshell AND GNU Parallel
B<coshell> only accepts full commands on standard input. Any quoting
needs to be done by the user.
Commands are run in B<sh> so any B<bash>/B<tcsh>/B<zsh> specific
syntax will not work.
Output can be buffered by using B<-d>. Output is buffered in memory,
so big output can cause swapping and therefore be terrible slow or
even cause out of memory.
https://github.com/gdm85/coshell (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN spread AND GNU Parallel
B<spread> runs commands on all directories.
It can be emulated with GNU B<parallel> using this Bash function:
spread() {
_cmds() {
perl -e '$"=" && ";print "@ARGV"' "cd {}" "$@"
}
parallel $(_cmds "$@")'|| echo exit status $?' ::: */
}
This works except for the B<--exclude> option.
(Last checked: 2017-11)
=head2 DIFFERENCES BETWEEN pyargs AND GNU Parallel
B<pyargs> deals badly with input containing spaces. It buffers stdout,
but not stderr. It buffers in RAM. {} does not work as replacement
string. It does not support running functions.
B<pyargs> does not support composed commands if run with B<--lines>,
and fails on B<pyargs traceroute gnu.org fsf.org>.
=head3 Examples
seq 5 | pyargs -P50 -L seq
seq 5 | parallel -P50 --lb seq
seq 5 | pyargs -P50 --mark -L seq
seq 5 | parallel -P50 --lb \
--tagstring OUTPUT'[{= $_=$job->replaced()=}]' seq
# Similar, but not precisely the same
seq 5 | parallel -P50 --lb --tag seq
seq 5 | pyargs -P50 --mark command
# Somewhat longer with GNU Parallel due to the special
# --mark formatting
cmd="$(echo "command" | parallel --shellquote)"
wrap_cmd() {
echo "MARK $cmd $@================================" >&3
echo "OUTPUT START[$cmd $@]:"
eval $cmd "$@"
echo "OUTPUT END[$cmd $@]"
}
(seq 5 | env_parallel -P2 wrap_cmd) 3>&1
# Similar, but not exactly the same
seq 5 | parallel -t --tag command
(echo '1 2 3';echo 4 5 6) | pyargs --stream seq
(echo '1 2 3';echo 4 5 6) | perl -pe 's/\n/ /' |
parallel -r -d' ' seq
# Similar, but not exactly the same
parallel seq ::: 1 2 3 4 5 6
https://github.com/robertblackwell/pyargs (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN concurrently AND GNU Parallel
B<concurrently> runs jobs in parallel.
The output is prepended with the job number, and may be incomplete:
$ concurrently 'seq 100000' | (sleep 3;wc -l)
7165
When pretty printing it caches output in memory. Output mixes by using
test MIX below whether or not output is cached.
There seems to be no way of making a template command and have
B<concurrently> fill that with different args. The full commands must
be given on the command line.
There is also no way of controlling how many jobs should be run in
parallel at a time - i.e. "number of jobslots". Instead all jobs are
simply started in parallel.
https://github.com/kimmobrunfeldt/concurrently (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN map(soveran) AND GNU Parallel
B<map> does not run jobs in parallel by default. The README suggests using:
... | map t 'sleep $t && say done &'
But this fails if more jobs are run in parallel than the number of
available processes. Since there is no support for parallelization in
B<map> itself, the output also mixes:
seq 10 | map i 'echo start-$i && sleep 0.$i && echo end-$i &'
The major difference is that GNU B<parallel> is built for parallelization
and B<map> is not. So GNU B<parallel> has lots of ways of dealing with the
issues that parallelization raises:
=over 4
=item *
Keep the number of processes manageable
=item *
Make sure output does not mix
=item *
Make Ctrl-C kill all running processes
=back
=head3 EXAMPLES FROM maps WEBSITE
Here are the 5 examples converted to GNU Parallel:
1$ ls *.c | map f 'foo $f'
1$ ls *.c | parallel foo
2$ ls *.c | map f 'foo $f; bar $f'
2$ ls *.c | parallel 'foo {}; bar {}'
3$ cat urls | map u 'curl -O $u'
3$ cat urls | parallel curl -O
4$ printf "1\n1\n1\n" | map t 'sleep $t && say done'
4$ printf "1\n1\n1\n" | parallel 'sleep {} && say done'
4$ parallel 'sleep {} && say done' ::: 1 1 1
5$ printf "1\n1\n1\n" | map t 'sleep $t && say done &'
5$ printf "1\n1\n1\n" | parallel -j0 'sleep {} && say done'
5$ parallel -j0 'sleep {} && say done' ::: 1 1 1
https://github.com/soveran/map (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN loop AND GNU Parallel
B<loop> mixes stdout and stderr:
loop 'ls /no-such-file' >/dev/null
B<loop>'s replacement string B<$ITEM> does not quote strings:
echo 'two spaces' | loop 'echo $ITEM'
B<loop> cannot run functions:
myfunc() { echo joe; }
export -f myfunc
loop 'myfunc this fails'
=head3 EXAMPLES FROM loop's WEBSITE
Some of the examples from https://github.com/Miserlou/Loop/ can be
emulated with GNU B<parallel>:
# A couple of functions will make the code easier to read
$ loopy() {
yes | parallel -uN0 -j1 "$@"
}
$ export -f loopy
$ time_out() {
parallel -uN0 -q --timeout "$@" ::: 1
}
$ match() {
perl -0777 -ne 'grep /'"$1"'/,$_ and print or exit 1'
}
$ export -f match
$ loop 'ls' --every 10s
$ loopy --delay 10s ls
$ loop 'touch $COUNT.txt' --count-by 5
$ loopy touch '{= $_=seq()*5 =}'.txt
$ loop --until-contains 200 -- \
./get_response_code.sh --site mysite.biz`
$ loopy --halt now,success=1 \
'./get_response_code.sh --site mysite.biz | match 200'
$ loop './poke_server' --for-duration 8h
$ time_out 8h loopy ./poke_server
$ loop './poke_server' --until-success
$ loopy --halt now,success=1 ./poke_server
$ cat files_to_create.txt | loop 'touch $ITEM'
$ cat files_to_create.txt | parallel touch {}
$ loop 'ls' --for-duration 10min --summary
# --joblog is somewhat more verbose than --summary
$ time_out 10m loopy --joblog my.log ./poke_server; cat my.log
$ loop 'echo hello'
$ loopy echo hello
$ loop 'echo $COUNT'
# GNU Parallel counts from 1
$ loopy echo {#}
# Counting from 0 can be forced
$ loopy echo '{= $_=seq()-1 =}'
$ loop 'echo $COUNT' --count-by 2
$ loopy echo '{= $_=2*(seq()-1) =}'
$ loop 'echo $COUNT' --count-by 2 --offset 10
$ loopy echo '{= $_=10+2*(seq()-1) =}'
$ loop 'echo $COUNT' --count-by 1.1
# GNU Parallel rounds 3.3000000000000003 to 3.3
$ loopy echo '{= $_=1.1*(seq()-1) =}'
$ loop 'echo $COUNT $ACTUALCOUNT' --count-by 2
$ loopy echo '{= $_=2*(seq()-1) =} {#}'
$ loop 'echo $COUNT' --num 3 --summary
# --joblog is somewhat more verbose than --summary
$ seq 3 | parallel --joblog my.log echo; cat my.log
$ loop 'ls -foobarbatz' --num 3 --summary
# --joblog is somewhat more verbose than --summary
$ seq 3 | parallel --joblog my.log -N0 ls -foobarbatz; cat my.log
$ loop 'echo $COUNT' --count-by 2 --num 50 --only-last
# Can be emulated by running 2 jobs
$ seq 49 | parallel echo '{= $_=2*(seq()-1) =}' >/dev/null
$ echo 50| parallel echo '{= $_=2*(seq()-1) =}'
$ loop 'date' --every 5s
$ loopy --delay 5s date
$ loop 'date' --for-duration 8s --every 2s
$ time_out 8s loopy --delay 2s date
$ loop 'date -u' --until-time '2018-05-25 20:50:00' --every 5s
$ seconds=$((`date -d 2019-05-25T20:50:00 +%s` - `date +%s`))s
$ time_out $seconds loopy --delay 5s date -u
$ loop 'echo $RANDOM' --until-contains "666"
$ loopy --halt now,success=1 'echo $RANDOM | match 666'
$ loop 'if (( RANDOM % 2 )); then
(echo "TRUE"; true);
else
(echo "FALSE"; false);
fi' --until-success
$ loopy --halt now,success=1 'if (( $RANDOM % 2 )); then
(echo "TRUE"; true);
else
(echo "FALSE"; false);
fi'
$ loop 'if (( RANDOM % 2 )); then
(echo "TRUE"; true);
else
(echo "FALSE"; false);
fi' --until-error
$ loopy --halt now,fail=1 'if (( $RANDOM % 2 )); then
(echo "TRUE"; true);
else
(echo "FALSE"; false);
fi'
$ loop 'date' --until-match "(\d{4})"
$ loopy --halt now,success=1 'date | match [0-9][0-9][0-9][0-9]'
$ loop 'echo $ITEM' --for red,green,blue
$ parallel echo ::: red green blue
$ cat /tmp/my-list-of-files-to-create.txt | loop 'touch $ITEM'
$ cat /tmp/my-list-of-files-to-create.txt | parallel touch
$ ls | loop 'cp $ITEM $ITEM.bak'; ls
$ ls | parallel cp {} {}.bak; ls
$ loop 'echo $ITEM | tr a-z A-Z' -i
$ parallel 'echo {} | tr a-z A-Z'
# Or more efficiently:
$ parallel --pipe tr a-z A-Z
$ loop 'echo $ITEM' --for "`ls`"
$ parallel echo {} ::: "`ls`"
$ ls | loop './my_program $ITEM' --until-success;
$ ls | parallel --halt now,success=1 ./my_program {}
$ ls | loop './my_program $ITEM' --until-fail;
$ ls | parallel --halt now,fail=1 ./my_program {}
$ ./deploy.sh;
loop 'curl -sw "%{http_code}" http://coolwebsite.biz' \
--every 5s --until-contains 200;
./announce_to_slack.sh
$ ./deploy.sh;
loopy --delay 5s --halt now,success=1 \
'curl -sw "%{http_code}" http://coolwebsite.biz | match 200';
./announce_to_slack.sh
$ loop "ping -c 1 mysite.com" --until-success; ./do_next_thing
$ loopy --halt now,success=1 ping -c 1 mysite.com; ./do_next_thing
$ ./create_big_file -o my_big_file.bin;
loop 'ls' --until-contains 'my_big_file.bin';
./upload_big_file my_big_file.bin
# inotifywait is a better tool to detect file system changes.
# It can even make sure the file is complete
# so you are not uploading an incomplete file
$ inotifywait -qmre MOVED_TO -e CLOSE_WRITE --format %w%f . |
grep my_big_file.bin
$ ls | loop 'cp $ITEM $ITEM.bak'
$ ls | parallel cp {} {}.bak
$ loop './do_thing.sh' --every 15s --until-success --num 5
$ parallel --retries 5 --delay 15s ::: ./do_thing.sh
https://github.com/Miserlou/Loop/ (Last checked: 2018-10)
=head2 DIFFERENCES BETWEEN lorikeet AND GNU Parallel
B<lorikeet> can run jobs in parallel. It does this based on a
dependency graph described in a file, so this is similar to B<make>.
https://github.com/cetra3/lorikeet (Last checked: 2018-10)
=head2 DIFFERENCES BETWEEN spp AND GNU Parallel
B<spp> can run jobs in parallel. B<spp> does not use a command
template to generate the jobs, but requires jobs to be in a
file. Output from the jobs mix.
https://github.com/john01dav/spp (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN paral AND GNU Parallel
B<paral> prints a lot of status information and stores the output from
the commands run into files. This means it cannot be used the middle
of a pipe like this
paral "echo this" "echo does not" "echo work" | wc
Instead it puts the output into files named like
B<out_#_I<command>.out.log>. To get a very similar behaviour with GNU
B<parallel> use B<--results
'out_{#}_{=s/[^\sa-z_0-9]//g;s/\s+/_/g=}.log' --eta>
B<paral> only takes arguments on the command line and each argument
should be a full command. Thus it does not use command templates.
This limits how many jobs it can run in total, because they all need
to fit on a single command line.
B<paral> has no support for running jobs remotely.
=head3 EXAMPLES FROM README.markdown
The examples from B<README.markdown> and the corresponding command run
with GNU B<parallel> (B<--results
'out_{#}_{=s/[^\sa-z_0-9]//g;s/\s+/_/g=}.log' --eta> is omitted from
the GNU B<parallel> command):
1$ paral "command 1" "command 2 --flag" "command arg1 arg2"
1$ parallel ::: "command 1" "command 2 --flag" "command arg1 arg2"
2$ paral "sleep 1 && echo c1" "sleep 2 && echo c2" \
"sleep 3 && echo c3" "sleep 4 && echo c4" "sleep 5 && echo c5"
2$ parallel ::: "sleep 1 && echo c1" "sleep 2 && echo c2" \
"sleep 3 && echo c3" "sleep 4 && echo c4" "sleep 5 && echo c5"
# Or shorter:
parallel "sleep {} && echo c{}" ::: {1..5}
3$ paral -n=0 "sleep 5 && echo c5" "sleep 4 && echo c4" \
"sleep 3 && echo c3" "sleep 2 && echo c2" "sleep 1 && echo c1"
3$ parallel ::: "sleep 5 && echo c5" "sleep 4 && echo c4" \
"sleep 3 && echo c3" "sleep 2 && echo c2" "sleep 1 && echo c1"
# Or shorter:
parallel -j0 "sleep {} && echo c{}" ::: 5 4 3 2 1
4$ paral -n=1 "sleep 5 && echo c5" "sleep 4 && echo c4" \
"sleep 3 && echo c3" "sleep 2 && echo c2" "sleep 1 && echo c1"
4$ parallel -j1 "sleep {} && echo c{}" ::: 5 4 3 2 1
5$ paral -n=2 "sleep 5 && echo c5" "sleep 4 && echo c4" \
"sleep 3 && echo c3" "sleep 2 && echo c2" "sleep 1 && echo c1"
5$ parallel -j2 "sleep {} && echo c{}" ::: 5 4 3 2 1
6$ paral -n=5 "sleep 5 && echo c5" "sleep 4 && echo c4" \
"sleep 3 && echo c3" "sleep 2 && echo c2" "sleep 1 && echo c1"
6$ parallel -j5 "sleep {} && echo c{}" ::: 5 4 3 2 1
7$ paral -n=1 "echo a && sleep 0.5 && echo b && sleep 0.5 && \
echo c && sleep 0.5 && echo d && sleep 0.5 && \
echo e && sleep 0.5 && echo f && sleep 0.5 && \
echo g && sleep 0.5 && echo h"
7$ parallel ::: "echo a && sleep 0.5 && echo b && sleep 0.5 && \
echo c && sleep 0.5 && echo d && sleep 0.5 && \
echo e && sleep 0.5 && echo f && sleep 0.5 && \
echo g && sleep 0.5 && echo h"
https://github.com/amattn/paral (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN concurr AND GNU Parallel
B<concurr> is built to run jobs in parallel using a client/server
model.
=head3 EXAMPLES FROM README.md
The examples from B<README.md>:
1$ concurr 'echo job {#} on slot {%}: {}' : arg1 arg2 arg3 arg4
1$ parallel 'echo job {#} on slot {%}: {}' ::: arg1 arg2 arg3 arg4
2$ concurr 'echo job {#} on slot {%}: {}' :: file1 file2 file3
2$ parallel 'echo job {#} on slot {%}: {}' :::: file1 file2 file3
3$ concurr 'echo {}' < input_file
3$ parallel 'echo {}' < input_file
4$ cat file | concurr 'echo {}'
4$ cat file | parallel 'echo {}'
B<concurr> deals badly empty input files and with output larger than
64 KB.
https://github.com/mmstick/concurr (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN lesser-parallel AND GNU Parallel
B<lesser-parallel> is the inspiration for B<parallel --embed>. Both
B<lesser-parallel> and B<parallel --embed> define bash functions that
can be included as part of a bash script to run jobs in parallel.
B<lesser-parallel> implements a few of the replacement strings, but
hardly any options, whereas B<parallel --embed> gives you the full
GNU B<parallel> experience.
https://github.com/kou1okada/lesser-parallel (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN npm-parallel AND GNU Parallel
B<npm-parallel> can run npm tasks in parallel.
There are no examples and very little documentation, so it is hard to
compare to GNU B<parallel>.
https://github.com/spion/npm-parallel (Last checked: 2019-01)
=head2 DIFFERENCES BETWEEN machma AND GNU Parallel
B<machma> runs tasks in parallel. It gives time stamped
output. It buffers in RAM.
=head3 EXAMPLES FROM README.md
The examples from README.md:
1$ # Put shorthand for timestamp in config for the examples
echo '--rpl '\
\''{time} $_=::strftime("%Y-%m-%d %H:%M:%S",localtime())'\' \
> ~/.parallel/machma
echo '--line-buffer --tagstring "{#} {time} {}"' \
>> ~/.parallel/machma
2$ find . -iname '*.jpg' |
machma -- mogrify -resize 1200x1200 -filter Lanczos {}
find . -iname '*.jpg' |
parallel --bar -Jmachma mogrify -resize 1200x1200 \
-filter Lanczos {}
3$ cat /tmp/ips | machma -p 2 -- ping -c 2 -q {}
3$ cat /tmp/ips | parallel -j2 -Jmachma ping -c 2 -q {}
4$ cat /tmp/ips |
machma -- sh -c 'ping -c 2 -q $0 > /dev/null && echo alive' {}
4$ cat /tmp/ips |
parallel -Jmachma 'ping -c 2 -q {} > /dev/null && echo alive'
5$ find . -iname '*.jpg' |
machma --timeout 5s -- mogrify -resize 1200x1200 \
-filter Lanczos {}
5$ find . -iname '*.jpg' |
parallel --timeout 5s --bar mogrify -resize 1200x1200 \
-filter Lanczos {}
6$ find . -iname '*.jpg' -print0 |
machma --null -- mogrify -resize 1200x1200 -filter Lanczos {}
6$ find . -iname '*.jpg' -print0 |
parallel --null --bar mogrify -resize 1200x1200 \
-filter Lanczos {}
https://github.com/fd0/machma (Last checked: 2019-06)
=head2 DIFFERENCES BETWEEN interlace AND GNU Parallel
Summary table (see legend above):
- I2 I3 I4 - - -
M1 - M3 - - M6
- O2 O3 - - - - x x
E1 E2 - - - - -
- - - - - - - - -
- -
B<interlace> is built for network analysis to run network tools in parallel.
B<interface> does not buffer output, so output from different jobs mixes.
The overhead for each target is O(n*n), so with 1000 targets it
becomes very slow with an overhead in the order of 500ms/target.
=head3 EXAMPLES FROM interlace's WEBSITE
Using B<prips> most of the examples from
https://github.com/codingo/Interlace can be run with GNU B<parallel>:
Blocker
commands.txt:
mkdir -p _output_/_target_/scans/
_blocker_
nmap _target_ -oA _output_/_target_/scans/_target_-nmap
interlace -tL ./targets.txt -cL commands.txt -o $output
parallel -a targets.txt \
mkdir -p $output/{}/scans/\; nmap {} -oA $output/{}/scans/{}-nmap
Blocks
commands.txt:
_block:nmap_
mkdir -p _target_/output/scans/
nmap _target_ -oN _target_/output/scans/_target_-nmap
_block:nmap_
nikto --host _target_
interlace -tL ./targets.txt -cL commands.txt
_nmap() {
mkdir -p $1/output/scans/
nmap $1 -oN $1/output/scans/$1-nmap
}
export -f _nmap
parallel ::: _nmap "nikto --host" :::: targets.txt
Run Nikto Over Multiple Sites
interlace -tL ./targets.txt -threads 5 \
-c "nikto --host _target_ > ./_target_-nikto.txt" -v
parallel -a targets.txt -P5 nikto --host {} \> ./{}_-nikto.txt
Run Nikto Over Multiple Sites and Ports
interlace -tL ./targets.txt -threads 5 -c \
"nikto --host _target_:_port_ > ./_target_-_port_-nikto.txt" \
-p 80,443 -v
parallel -P5 nikto --host {1}:{2} \> ./{1}-{2}-nikto.txt \
:::: targets.txt ::: 80 443
Run a List of Commands against Target Hosts
commands.txt:
nikto --host _target_:_port_ > _output_/_target_-nikto.txt
sslscan _target_:_port_ > _output_/_target_-sslscan.txt
testssl.sh _target_:_port_ > _output_/_target_-testssl.txt
interlace -t example.com -o ~/Engagements/example/ \
-cL ./commands.txt -p 80,443
parallel --results ~/Engagements/example/{2}:{3}{1} {1} {2}:{3} \
::: "nikto --host" sslscan testssl.sh ::: example.com ::: 80 443
CIDR notation with an application that doesn't support it
interlace -t 192.168.12.0/24 -c "vhostscan _target_ \
-oN _output_/_target_-vhosts.txt" -o ~/scans/ -threads 50
prips 192.168.12.0/24 |
parallel -P50 vhostscan {} -oN ~/scans/{}-vhosts.txt
Glob notation with an application that doesn't support it
interlace -t 192.168.12.* -c "vhostscan _target_ \
-oN _output_/_target_-vhosts.txt" -o ~/scans/ -threads 50
# Glob is not supported in prips
prips 192.168.12.0/24 |
parallel -P50 vhostscan {} -oN ~/scans/{}-vhosts.txt
Dash (-) notation with an application that doesn't support it
interlace -t 192.168.12.1-15 -c \
"vhostscan _target_ -oN _output_/_target_-vhosts.txt" \
-o ~/scans/ -threads 50
# Dash notation is not supported in prips
prips 192.168.12.1 192.168.12.15 |
parallel -P50 vhostscan {} -oN ~/scans/{}-vhosts.txt
Threading Support for an application that doesn't support it
interlace -tL ./target-list.txt -c \
"vhostscan -t _target_ -oN _output_/_target_-vhosts.txt" \
-o ~/scans/ -threads 50
cat ./target-list.txt |
parallel -P50 vhostscan -t {} -oN ~/scans/{}-vhosts.txt
alternatively
./vhosts-commands.txt:
vhostscan -t $target -oN _output_/_target_-vhosts.txt
interlace -cL ./vhosts-commands.txt -tL ./target-list.txt \
-threads 50 -o ~/scans
./vhosts-commands.txt:
vhostscan -t "$1" -oN "$2"
parallel -P50 ./vhosts-commands.txt {} ~/scans/{}-vhosts.txt \
:::: ./target-list.txt
Exclusions
interlace -t 192.168.12.0/24 -e 192.168.12.0/26 -c \
"vhostscan _target_ -oN _output_/_target_-vhosts.txt" \
-o ~/scans/ -threads 50
prips 192.168.12.0/24 | grep -xv -Ff <(prips 192.168.12.0/26) |
parallel -P50 vhostscan {} -oN ~/scans/{}-vhosts.txt
Run Nikto Using Multiple Proxies
interlace -tL ./targets.txt -pL ./proxies.txt -threads 5 -c \
"nikto --host _target_:_port_ -useproxy _proxy_ > \
./_target_-_port_-nikto.txt" -p 80,443 -v
parallel -j5 \
"nikto --host {1}:{2} -useproxy {3} > ./{1}-{2}-nikto.txt" \
:::: ./targets.txt ::: 80 443 :::: ./proxies.txt
https://github.com/codingo/Interlace (Last checked: 2019-09)
=head2 DIFFERENCES BETWEEN otonvm Parallel AND GNU Parallel
I have been unable to get the code to run at all. It seems unfinished.
https://github.com/otonvm/Parallel (Last checked: 2019-02)
=head2 DIFFERENCES BETWEEN k-bx par AND GNU Parallel
B<par> requires Haskell to work. This limits the number of platforms
this can work on.
B<par> does line buffering in memory. The memory usage is 3x the
longest line (compared to 1x for B<parallel --lb>). Commands must be
given as arguments. There is no template.
These are the examples from https://github.com/k-bx/par with the
corresponding GNU B<parallel> command.
par "echo foo; sleep 1; echo foo; sleep 1; echo foo" \
"echo bar; sleep 1; echo bar; sleep 1; echo bar" && echo "success"
parallel --lb ::: "echo foo; sleep 1; echo foo; sleep 1; echo foo" \
"echo bar; sleep 1; echo bar; sleep 1; echo bar" && echo "success"
par "echo foo; sleep 1; foofoo" \
"echo bar; sleep 1; echo bar; sleep 1; echo bar" && echo "success"
parallel --lb --halt 1 ::: "echo foo; sleep 1; foofoo" \
"echo bar; sleep 1; echo bar; sleep 1; echo bar" && echo "success"
par "PARPREFIX=[fooechoer] echo foo" "PARPREFIX=[bar] echo bar"
parallel --lb --colsep , --tagstring {1} {2} \
::: "[fooechoer],echo foo" "[bar],echo bar"
par --succeed "foo" "bar" && echo 'wow'
parallel "foo" "bar"; true && echo 'wow'
https://github.com/k-bx/par (Last checked: 2019-02)
=head2 DIFFERENCES BETWEEN parallelshell AND GNU Parallel
B<parallelshell> does not allow for composed commands:
# This does not work
parallelshell 'echo foo;echo bar' 'echo baz;echo quuz'
Instead you have to wrap that in a shell:
parallelshell 'sh -c "echo foo;echo bar"' 'sh -c "echo baz;echo quuz"'
It buffers output in RAM. All commands must be given on the command
line and all commands are started in parallel at the same time. This
will cause the system to freeze if there are so many jobs that there
is not enough memory to run them all at the same time.
https://github.com/keithamus/parallelshell (Last checked: 2019-02)
https://github.com/darkguy2008/parallelshell (Last checked: 2019-03)
=head2 DIFFERENCES BETWEEN shell-executor AND GNU Parallel
B<shell-executor> does not allow for composed commands:
# This does not work
sx 'echo foo;echo bar' 'echo baz;echo quuz'
Instead you have to wrap that in a shell:
sx 'sh -c "echo foo;echo bar"' 'sh -c "echo baz;echo quuz"'
It buffers output in RAM. All commands must be given on the command
line and all commands are started in parallel at the same time. This
will cause the system to freeze if there are so many jobs that there
is not enough memory to run them all at the same time.
https://github.com/royriojas/shell-executor (Last checked: 2019-02)
=head2 DIFFERENCES BETWEEN non-GNU par AND GNU Parallel
B<par> buffers in memory to avoid mixing of jobs. It takes 1s per 1
million output lines.
B<par> needs to have all commands before starting the first job. The
jobs are read from stdin (standard input) so any quoting will have to
be done by the user.
Stdout (standard output) is prepended with o:. Stderr (standard error)
is sendt to stdout (standard output) and prepended with e:.
For short jobs with little output B<par> is 20% faster than GNU
B<parallel> and 60% slower than B<xargs>.
http://savannah.nongnu.org/projects/par (Last checked: 2019-02)
=head2 DIFFERENCES BETWEEN fd AND GNU Parallel
B<fd> does not support composed commands, so commands must be wrapped
in B<sh -c>.
It buffers output in RAM.
It only takes file names from the filesystem as input (similar to B<find>).
https://github.com/sharkdp/fd (Last checked: 2019-02)
=head2 DIFFERENCES BETWEEN lateral AND GNU Parallel
B<lateral> is very similar to B<sem>: It takes a single command and
runs it in the background. The design means that output from parallel
running jobs may mix. If it dies unexpectly it leaves a socket in
~/.lateral/socket.PID.
B<lateral> deals badly with too long command lines. This makes the
B<lateral> server crash:
lateral run echo `seq 100000| head -c 1000k`
Any options will be read by B<lateral> so this does not work
(B<lateral> interprets the B<-l>):
lateral run ls -l
Composed commands do not work:
lateral run pwd ';' ls
Functions do not work:
myfunc() { echo a; }
export -f myfunc
lateral run myfunc
Running B<emacs> in the terminal causes the parent shell to die:
echo '#!/bin/bash' > mycmd
echo emacs -nw >> mycmd
chmod +x mycmd
lateral start
lateral run ./mycmd
Here are the examples from https://github.com/akramer/lateral with the
corresponding GNU B<sem> and GNU B<parallel> commands:
1$ lateral start
1$ for i in $(cat /tmp/names); do
1$ lateral run -- some_command $i
1$ done
1$ lateral wait
1$
1$ for i in $(cat /tmp/names); do
1$ sem some_command $i
1$ done
1$ sem --wait
1$
1$ parallel some_command :::: /tmp/names
2$ lateral start
2$ for i in $(seq 1 100); do
2$ lateral run -- my_slow_command < workfile$i > /tmp/logfile$i
2$ done
2$ lateral wait
2$
2$ for i in $(seq 1 100); do
2$ sem my_slow_command < workfile$i > /tmp/logfile$i
2$ done
2$ sem --wait
2$
2$ parallel 'my_slow_command < workfile{} > /tmp/logfile{}' \
::: {1..100}
3$ lateral start -p 0 # yup, it will just queue tasks
3$ for i in $(seq 1 100); do
3$ lateral run -- command_still_outputs_but_wont_spam inputfile$i
3$ done
3$ # command output spam can commence
3$ lateral config -p 10; lateral wait
3$
3$ for i in $(seq 1 100); do
3$ echo "command inputfile$i" >> joblist
3$ done
3$ parallel -j 10 :::: joblist
3$
3$ echo 1 > /tmp/njobs
3$ parallel -j /tmp/njobs command inputfile{} \
::: {1..100} &
3$ echo 10 >/tmp/njobs
3$ wait
https://github.com/akramer/lateral (Last checked: 2019-03)
=head2 DIFFERENCES BETWEEN with-this AND GNU Parallel
The examples from https://github.com/amritb/with-this.git and the
corresponding GNU B<parallel> command:
with -v "$(cat myurls.txt)" "curl -L this"
parallel curl -L ::: myurls.txt
with -v "$(cat myregions.txt)" \
"aws --region=this ec2 describe-instance-status"
parallel aws --region={} ec2 describe-instance-status \
:::: myregions.txt
with -v "$(ls)" "kubectl --kubeconfig=this get pods"
ls | parallel kubectl --kubeconfig={} get pods
with -v "$(ls | grep config)" "kubectl --kubeconfig=this get pods"
ls | grep config | parallel kubectl --kubeconfig={} get pods
with -v "$(echo {1..10})" "echo 123"
parallel -N0 echo 123 ::: {1..10}
Stderr is merged with stdout. B<with-this> buffers in RAM. It uses 3x
the output size, so you cannot have output larger than 1/3rd the
amount of RAM. The input values cannot contain spaces. Composed
commands do not work.
B<with-this> gives some additional information, so the output has to
be cleaned before piping it to the next command.
https://github.com/amritb/with-this.git (Last checked: 2019-03)
=head2 DIFFERENCES BETWEEN Tollef's parallel (moreutils) AND GNU Parallel
Summary table (see legend above):
- - - I4 - - I7
- - M3 - - M6
- O2 O3 - O5 O6 - x x
E1 - - - - - E7
- x x x x x x x x
- -
=head3 EXAMPLES FROM Tollef's parallel MANUAL
B<Tollef> parallel sh -c "echo hi; sleep 2; echo bye" -- 1 2 3
B<GNU> parallel "echo hi; sleep 2; echo bye" ::: 1 2 3
B<Tollef> parallel -j 3 ufraw -o processed -- *.NEF
B<GNU> parallel -j 3 ufraw -o processed ::: *.NEF
B<Tollef> parallel -j 3 -- ls df "echo hi"
B<GNU> parallel -j 3 ::: ls df "echo hi"
(Last checked: 2019-08)
=head2 DIFFERENCES BETWEEN rargs AND GNU Parallel
Summary table (see legend above):
I1 - - - - - I7
- - M3 M4 - -
- O2 O3 - O5 O6 - O8 -
E1 - - E4 - - -
- - - - - - - - -
- -
B<rargs> has elegant ways of doing named regexp capture and field ranges.
With GNU B<parallel> you can use B<--rpl> to get a similar
functionality as regexp capture gives, and use B<join> and B<@arg> to
get the field ranges. But the syntax is longer. This:
--rpl '{r(\d+)\.\.(\d+)} $_=join"$opt::colsep",@arg[$$1..$$2]'
would make it possible to use:
{1r3..6}
for field 3..6.
For full support of {n..m:s} including negative numbers use a dynamic
replacement string like this:
PARALLEL=--rpl\ \''{r((-?\d+)?)\.\.((-?\d+)?)((:([^}]*))?)}
$a = defined $$2 ? $$2 < 0 ? 1+$#arg+$$2 : $$2 : 1;
$b = defined $$4 ? $$4 < 0 ? 1+$#arg+$$4 : $$4 : $#arg+1;
$s = defined $$6 ? $$7 : " ";
$_ = join $s,@arg[$a..$b]'\'
export PARALLEL
You can then do:
head /etc/passwd | parallel --colsep : echo ..={1r..} ..3={1r..3} \
4..={1r4..} 2..4={1r2..4} 3..3={1r3..3} ..3:-={1r..3:-} \
..3:/={1r..3:/} -1={-1} -5={-5} -6={-6} -3..={1r-3..}
=head3 EXAMPLES FROM rargs MANUAL
ls *.bak | rargs -p '(.*)\.bak' mv {0} {1}
ls *.bak | parallel mv {} {.}
cat download-list.csv | rargs -p '(?P<url>.*),(?P<filename>.*)' wget {url} -O {filename}
cat download-list.csv | parallel --csv wget {1} -O {2}
# or use regexps:
cat download-list.csv |
parallel --rpl '{url} s/,.*//' --rpl '{filename} s/.*?,//' wget {url} -O {filename}
cat /etc/passwd | rargs -d: echo -e 'id: "{1}"\t name: "{5}"\t rest: "{6..::}"'
cat /etc/passwd |
parallel -q --colsep : echo -e 'id: "{1}"\t name: "{5}"\t rest: "{=6 $_=join":",@arg[6..$#arg]=}"'
https://github.com/lotabout/rargs (Last checked: 2020-01)
=head2 DIFFERENCES BETWEEN threader AND GNU Parallel
Summary table (see legend above):
I1 - - - - - -
M1 - M3 - - M6
O1 - O3 - O5 - - N/A N/A
E1 - - E4 - - -
- - - - - - - - -
- -
Newline separates arguments, but newline at the end of file is treated
as an empty argument. So this runs 2 jobs:
echo two_jobs | threader -run 'echo "$THREADID"'
B<threader> ignores stderr, so any output to stderr is
lost. B<threader> buffers in RAM, so output bigger than the machine's
virtual memory will cause the machine to crash.
https://github.com/voodooEntity/threader (Last checked: 2020-04)
=head2 DIFFERENCES BETWEEN runp AND GNU Parallel
Summary table (see legend above):
I1 I2 - - - - -
M1 - (M3) - - M6
O1 O2 O3 - O5 O6 - N/A N/A -
E1 - - - - - -
- - - - - - - - -
- -
(M3): You can add a prefix and a postfix to the input, so it means you can
only insert the argument on the command line once.
B<runp> runs 10 jobs in parallel by default. B<runp> blocks if output
of a command is > 64 Kbytes. Quoting of input is needed. It adds
output to stderr (this can be prevented with -q)
=head3 Examples as GNU Parallel
base='https://images-api.nasa.gov/search'
query='jupiter'
desc='planet'
type='image'
url="$base?q=$query&description=$desc&media_type=$type"
# Download the images in parallel using runp
curl -s $url | jq -r .collection.items[].href | \
runp -p 'curl -s' | jq -r .[] | grep large | \
runp -p 'curl -s -L -O'
time curl -s $url | jq -r .collection.items[].href | \
runp -g 1 -q -p 'curl -s' | jq -r .[] | grep large | \
runp -g 1 -q -p 'curl -s -L -O'
# Download the images in parallel
curl -s $url | jq -r .collection.items[].href | \
parallel curl -s | jq -r .[] | grep large | \
parallel curl -s -L -O
time curl -s $url | jq -r .collection.items[].href | \
parallel -j 1 curl -s | jq -r .[] | grep large | \
parallel -j 1 curl -s -L -O
=head4 Run some test commands (read from file)
# Create a file containing commands to run in parallel.
cat << EOF > /tmp/test-commands.txt
sleep 5
sleep 3
blah # this will fail
ls $PWD # PWD shell variable is used here
EOF
# Run commands from the file.
runp /tmp/test-commands.txt > /dev/null
parallel -a /tmp/test-commands.txt > /dev/null
=head4 Ping several hosts and see packet loss (read from stdin)
# First copy this line and press Enter
runp -p 'ping -c 5 -W 2' -s '| grep loss'
localhost
1.1.1.1
8.8.8.8
# Press Enter and Ctrl-D when done entering the hosts
# First copy this line and press Enter
parallel ping -c 5 -W 2 {} '| grep loss'
localhost
1.1.1.1
8.8.8.8
# Press Enter and Ctrl-D when done entering the hosts
=head4 Get directories' sizes (read from stdin)
echo -e "$HOME\n/etc\n/tmp" | runp -q -p 'sudo du -sh'
echo -e "$HOME\n/etc\n/tmp" | parallel sudo du -sh
# or:
parallel sudo du -sh ::: "$HOME" /etc /tmp
=head4 Compress files
find . -iname '*.txt' | runp -p 'gzip --best'
find . -iname '*.txt' | parallel gzip --best
=head4 Measure HTTP request + response time
export CURL="curl -w 'time_total: %{time_total}\n'"
CURL="$CURL -o /dev/null -s https://golang.org/"
perl -wE 'for (1..10) { say $ENV{CURL} }' |
runp -q # Make 10 requests
perl -wE 'for (1..10) { say $ENV{CURL} }' | parallel
# or:
parallel -N0 "$CURL" ::: {1..10}
=head4 Find open TCP ports
cat << EOF > /tmp/host-port.txt
localhost 22
localhost 80
localhost 81
127.0.0.1 443
127.0.0.1 444
scanme.nmap.org 22
scanme.nmap.org 23
scanme.nmap.org 443
EOF
cat /tmp/host-port.txt | \
runp -q -p 'netcat -v -w2 -z' 2>&1 | egrep '(succeeded!|open)$'
# --colsep is needed to split the line
cat /tmp/host-port.txt | \
parallel --colsep ' ' netcat -v -w2 -z 2>&1 | egrep '(succeeded!|open)$'
# or use uq for unquoted:
cat /tmp/host-port.txt | \
parallel netcat -v -w2 -z {=uq=} 2>&1 | egrep '(succeeded!|open)$'
https://github.com/jreisinger/runp (Last checked: 2020-04)
=head2 DIFFERENCES BETWEEN papply AND GNU Parallel
Summary table (see legend above):
- - - I4 - - -
M1 - M3 - - M6
- - O3 - O5 - - N/A N/A O10
E1 - - E4 - - -
- - - - - - - - -
- -
B<papply> does not print the output if the command fails:
$ papply 'echo %F; false' foo
"echo foo; false" did not succeed
B<papply>'s replacement strings (%F %d %f %n %e %z) can be simulated in GNU
B<parallel> by putting this in B<~/.parallel/config>:
--rpl '%F'
--rpl '%d $_=Q(::dirname($_));'
--rpl '%f s:.*/::;'
--rpl '%n s:.*/::;s:\.[^/.]+$::;'
--rpl '%e s:.*\.:.:'
--rpl '%z $_=""'
B<papply> buffers in RAM, and uses twice the amount of output. So
output of 5 GB takes 10 GB RAM.
The buffering is very CPU intensive: Buffering a line of 5 GB takes 40
seconds (compared to 10 seconds with GNU B<parallel>).
=head3 Examples as GNU Parallel
1$ papply gzip *.txt
1$ parallel gzip ::: *.txt
2$ papply "convert %F %n.jpg" *.png
2$ parallel convert {} {.}.jpg ::: *.png
https://pypi.org/project/papply/ (Last checked: 2020-04)
=head2 Todo
https://gitlab.com/netikras/bthread
https://github.com/JeiKeiLim/simple_distribute_job
https://github.com/reggi/pkgrun
https://github.com/benoror/better-npm-run - not obvious how to use
https://github.com/bahmutov/with-package
https://github.com/xuchenCN/go-pssh
https://github.com/flesler/parallel
https://github.com/Julian/Verge
https://github.com/ExpectationMax/simple_gpu_scheduler
simple_gpu_scheduler --gpus 0 1 2 < gpu_commands.txt
parallel -j3 --shuf CUDA_VISIBLE_DEVICES='{=1 $_=slot()-1 =} {=uq;=}' < gpu_commands.txt
simple_hypersearch "python3 train_dnn.py --lr {lr} --batch_size {bs}" -p lr 0.001 0.0005 0.0001 -p bs 32 64 128 | simple_gpu_scheduler --gpus 0,1,2
parallel --header : --shuf -j3 -v CUDA_VISIBLE_DEVICES='{=1 $_=slot()-1 =}' python3 train_dnn.py --lr {lr} --batch_size {bs} ::: lr 0.001 0.0005 0.0001 ::: bs 32 64 128
simple_hypersearch "python3 train_dnn.py --lr {lr} --batch_size {bs}" --n-samples 5 -p lr 0.001 0.0005 0.0001 -p bs 32 64 128 | simple_gpu_scheduler --gpus 0,1,2
parallel --header : --shuf CUDA_VISIBLE_DEVICES='{=1 $_=slot()-1; seq() > 5 and skip() =}' python3 train_dnn.py --lr {lr} --batch_size {bs} ::: lr 0.001 0.0005 0.0001 ::: bs 32 64 128
touch gpu.queue
tail -f -n 0 gpu.queue | simple_gpu_scheduler --gpus 0,1,2 &
echo "my_command_with | and stuff > logfile" >> gpu.queue
touch gpu.queue
tail -f -n 0 gpu.queue | parallel -j3 CUDA_VISIBLE_DEVICES='{=1 $_=slot()-1 =} {=uq;=}' &
# Needed to fill job slots once
seq 3 | parallel echo true >> gpu.queue
# Add jobs
echo "my_command_with | and stuff > logfile" >> gpu.queue
# Needed to flush output from completed jobs
seq 3 | parallel echo true >> gpu.queue
=head1 TESTING OTHER TOOLS
There are certain issues that are very common on parallelizing
tools. Here are a few stress tests. Be warned: If the tool is badly
coded it may overload your machine.
=head2 MIX: Output mixes
Output from 2 jobs should not mix. If the output is not used, this
does not matter; but if the output I<is> used then it is important
that you do not get half a line from one job followed by half a line
from another job.
If the tool does not buffer, output will most likely mix now and then.
This test stresses whether output mixes.
#!/bin/bash
paralleltool="parallel -j0"
cat <<-EOF > mycommand
#!/bin/bash
# If a, b, c, d, e, and f mix: Very bad
perl -e 'print STDOUT "a"x3000_000," "'
perl -e 'print STDERR "b"x3000_000," "'
perl -e 'print STDOUT "c"x3000_000," "'
perl -e 'print STDERR "d"x3000_000," "'
perl -e 'print STDOUT "e"x3000_000," "'
perl -e 'print STDERR "f"x3000_000," "'
echo
echo >&2
EOF
chmod +x mycommand
# Run 30 jobs in parallel
seq 30 |
$paralleltool ./mycommand > >(tr -s abcdef) 2> >(tr -s abcdef >&2)
# 'a c e' and 'b d f' should always stay together
# and there should only be a single line per job
=head2 STDERRMERGE: Stderr is merged with stdout
Output from stdout and stderr should not be merged, but kept separated.
This test shows whether stdout is mixed with stderr.
#!/bin/bash
paralleltool="parallel -j0"
cat <<-EOF > mycommand
#!/bin/bash
echo stdout
echo stderr >&2
echo stdout
echo stderr >&2
EOF
chmod +x mycommand
# Run one job
echo |
$paralleltool ./mycommand > stdout 2> stderr
cat stdout
cat stderr
=head2 RAM: Output limited by RAM
Some tools cache output in RAM. This makes them extremely slow if the
output is bigger than physical memory and crash if the output is
bigger than the virtual memory.
#!/bin/bash
paralleltool="parallel -j0"
cat <<'EOF' > mycommand
#!/bin/bash
# Generate 1 GB output
yes "`perl -e 'print \"c\"x30_000'`" | head -c 1G
EOF
chmod +x mycommand
# Run 20 jobs in parallel
# Adjust 20 to be > physical RAM and < free space on /tmp
seq 20 | time $paralleltool ./mycommand | wc -c
=head2 DISKFULL: Incomplete data if /tmp runs full
If caching is done on disk, the disk can run full during the run. Not
all programs discover this. GNU Parallel discovers it, if it stays
full for at least 2 seconds.
#!/bin/bash
paralleltool="parallel -j0"
# This should be a dir with less than 100 GB free space
smalldisk=/tmp/shm/parallel
TMPDIR="$smalldisk"
export TMPDIR
max_output() {
# Force worst case scenario:
# Make GNU Parallel only check once per second
sleep 10
# Generate 100 GB to fill $TMPDIR
# Adjust if /tmp is bigger than 100 GB
yes | head -c 100G >$TMPDIR/$$
# Generate 10 MB output that will not be buffered due to full disk
perl -e 'print "X"x10_000_000' | head -c 10M
echo This part is missing from incomplete output
sleep 2
rm $TMPDIR/$$
echo Final output
}
export -f max_output
seq 10 | $paralleltool max_output | tr -s X
=head2 CLEANUP: Leaving tmp files at unexpected death
Some tools do not clean up tmp files if they are killed. If the tool
buffers on disk, they may not clean up, if they are killed.
#!/bin/bash
paralleltool=parallel
ls /tmp >/tmp/before
seq 10 | $paralleltool sleep &
pid=$!
# Give the tool time to start up
sleep 1
# Kill it without giving it a chance to cleanup
kill -9 $!
# Should be empty: No files should be left behind
diff <(ls /tmp) /tmp/before
=head2 SPCCHAR: Dealing badly with special file names.
It is not uncommon for users to create files like:
My brother's 12" *** record (costs $$$).jpg
Some tools break on this.
#!/bin/bash
paralleltool=parallel
touch "My brother's 12\" *** record (costs \$\$\$).jpg"
ls My*jpg | $paralleltool ls -l
=head2 COMPOSED: Composed commands do not work
Some tools require you to wrap composed commands into B<bash -c>.
echo bar | $paralleltool echo foo';' echo {}
=head2 ONEREP: Only one replacement string allowed
Some tools can only insert the argument once.
echo bar | $paralleltool echo {} foo {}
=head2 INPUTSIZE: Length of input should not be limited
Some tools limit the length of the input lines artificially with no good
reason. GNU B<parallel> does not:
perl -e 'print "foo."."x"x100_000_000' | parallel echo {.}
GNU B<parallel> limits the command to run to 128 KB due to execve(1):
perl -e 'print "x"x131_000' | parallel echo {} | wc
=head2 NUMWORDS: Speed depends on number of words
Some tools become very slow if output lines have many words.
#!/bin/bash
paralleltool=parallel
cat <<-EOF > mycommand
#!/bin/bash
# 10 MB of lines with 1000 words
yes "`seq 1000`" | head -c 10M
EOF
chmod +x mycommand
# Run 30 jobs in parallel
seq 30 | time $paralleltool -j0 ./mycommand > /dev/null
=head2 4GB: Output with a line > 4GB should be OK
#!/bin/bash
paralleltool="parallel -j0"
cat <<-EOF > mycommand
#!/bin/bash
perl -e '\$a="a"x1000_000; for(1..5000) { print \$a }'
EOF
chmod +x mycommand
# Run 1 job
seq 1 | $paralleltool ./mycommand | LC_ALL=C wc
=head1 AUTHOR
When using GNU B<parallel> for a publication please cite:
O. Tange (2011): GNU Parallel - The Command-Line Power Tool, ;login:
The USENIX Magazine, February 2011:42-47.
This helps funding further development; and it won't cost you a cent.
If you pay 10000 EUR you should feel free to use GNU Parallel without citing.
Copyright (C) 2007-10-18 Ole Tange, http://ole.tange.dk
Copyright (C) 2008-2010 Ole Tange, http://ole.tange.dk
Copyright (C) 2010-2020 Ole Tange, http://ole.tange.dk and Free
Software Foundation, Inc.
Parts of the manual concerning B<xargs> compatibility is inspired by
the manual of B<xargs> from GNU findutils 4.4.2.
=head1 LICENSE
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or
at your option any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
=head2 Documentation license I
Permission is granted to copy, distribute and/or modify this documentation
under the terms of the GNU Free Documentation License, Version 1.3 or
any later version published by the Free Software Foundation; with no
Invariant Sections, with no Front-Cover Texts, and with no Back-Cover
Texts. A copy of the license is included in the file fdl.txt.
=head2 Documentation license II
You are free:
=over 9
=item B<to Share>
to copy, distribute and transmit the work
=item B<to Remix>
to adapt the work
=back
Under the following conditions:
=over 9
=item B<Attribution>
You must attribute the work in the manner specified by the author or
licensor (but not in any way that suggests that they endorse you or
your use of the work).
=item B<Share Alike>
If you alter, transform, or build upon this work, you may distribute
the resulting work only under the same, similar or a compatible
license.
=back
With the understanding that:
=over 9
=item B<Waiver>
Any of the above conditions can be waived if you get permission from
the copyright holder.
=item B<Public Domain>
Where the work or any of its elements is in the public domain under
applicable law, that status is in no way affected by the license.
=item B<Other Rights>
In no way are any of the following rights affected by the license:
=over 2
=item *
Your fair dealing or fair use rights, or other applicable
copyright exceptions and limitations;
=item *
The author's moral rights;
=item *
Rights other persons may have either in the work itself or in
how the work is used, such as publicity or privacy rights.
=back
=back
=over 9
=item B<Notice>
For any reuse or distribution, you must make clear to others the
license terms of this work.
=back
A copy of the full license is included in the file as cc-by-sa.txt.
=head1 DEPENDENCIES
GNU B<parallel> uses Perl, and the Perl modules Getopt::Long,
IPC::Open3, Symbol, IO::File, POSIX, and File::Temp. For remote usage
it also uses rsync with ssh.
=head1 SEE ALSO
B<find>(1), B<xargs>(1), B<make>(1), B<pexec>(1), B<ppss>(1),
B<xjobs>(1), B<prll>(1), B<dxargs>(1), B<mdm>(1)
=cut
Reference in a new issue View git blame Copy permalink