Update of instant load calc on more platforms. Needs more tought, though.

This commit is contained in:
Ole Tange 2014-11-24 21:45:11 +01:00
parent 70d6afbbb9
commit 87f65b76fe
2 changed files with 89 additions and 37 deletions

View file

@ -226,34 +226,17 @@ cc:Tim Cuthbertson <tim3d.junk@gmail.com>,
Ryoichiro Suzuki <ryoichiro.suzuki@gmail.com>,
Jesse Alama <jesse.alama@gmail.com>
Subject: GNU Parallel 20141122 ('Rosetta') released
Subject: GNU Parallel 20141222 ('') released
GNU Parallel 20141122 ('Rosetta') has been released. It is available for download at: http://ftp.gnu.org/gnu/parallel/
GNU Parallel 20141222 ('') has been released. It is available for download at: http://ftp.gnu.org/gnu/parallel/
Haiku of the month:
Hadoop bit too much?
Want a simpler syntax now?
Use GNU Parallel.
-- Ole Tange
A central piece of command generation was rewritten making this release beta quality. As always it passes the testsuite, so most functionality clearly works.
<<>>
New in this release:
* Remote systems can be divided into hostgroups (e.g. web and db) by prepending '@groupname/' to the sshlogin. Multiple groups can be given by separating groups with '+'. E.g. @web/www1 @web+db/www2 @db/mariadb
* Remote execution can be restricted to servers that are part of one or more groups by '@groupname' as an sshlogin. Multiple groups can be given by separating groups with '+'. E.g. -S @web or -S @db+web
* With --hostgroup you can restrict arguments to certain hostgroups by appending '@groupname' to the argument. Multiple groups can be given by separating groups with '+'. E.g. my_web_arg@web db-or-web-arg@db+web db-only-arg@db Thanks to Michel Courtine for developing a prototype for this.
* GNU Parallel was cited in: HTSeq-Hadoop: Extending HTSeq for Massively Parallel Sequencing Data Analysis using Hadoop http://essenceofescience.se/wp-content/uploads/2014/11/Siretskiy.pdf
* GNU Parallel was cited in: SlideToolkit: An Assistive Toolset for the Histological Quantification of Whole Slide Images http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0110289#close
* GNU Parallel was cited in: Exploring a multiprocessor design space to analyze the impact of using STT-RAM in the memory hierarchy http://conservancy.umn.edu/bitstream/handle/11299/167286/Borse_umn_0130M_15431.pdf
* Command-Line OCR with Tesseract on Mac OS X https://ryanfb.github.io/etc/2014/11/13/command_line_ocr_on_mac_os_x.html
* GNU Parallel was cited in: Parallel post-processing with MPI-Bash http://dl.acm.org/citation.cfm?id=2691137
* Bug fixes and man page updates.

View file

@ -194,7 +194,6 @@ sub pipe_part_files {
# Remote exec should look like:
# ssh -oLogLevel=quiet lo 'eval `echo $SHELL | grep "/t\{0,1\}csh" > /dev/null && echo setenv PARALLEL_SEQ '$PARALLEL_SEQ'\; setenv PARALLEL_PID '$PARALLEL_PID' || echo PARALLEL_SEQ='$PARALLEL_SEQ'\;export PARALLEL_SEQ\; PARALLEL_PID='$PARALLEL_PID'\;export PARALLEL_PID` ;' tty\ \>/dev/null\ \&\&\ stty\ isig\ -onlcr\ -echo\;echo\ \$SHELL\ \|\ grep\ \"/t\\\{0,1\\\}csh\"\ \>\ /dev/null\ \&\&\ setenv\ FOO\ /tmp/foo\ \|\|\ export\ FOO=/tmp/foo\; \(wc\ -\ \$FOO\)
# ssh -tt not allowed. Remote will die due to broken pipe anyway.
# TODO test remote with --fifo / --cat
return @cat_partials;
}
@ -3134,15 +3133,15 @@ sub which {
(
'aix' => $sysv,
'cygwin' => $sysv,
'msys' => $sysv,
'dec_osf' => $sysv,
'darwin' => $bsd,
'dec_osf' => $sysv,
'dragonfly' => $bsd,
'freebsd' => $bsd,
'gnu' => $sysv,
'hpux' => $sysv,
'linux' => $sysv,
'mirbsd' => $bsd,
'msys' => $sysv,
'netbsd' => $bsd,
'nto' => $sysv,
'openbsd' => $bsd,
@ -3724,6 +3723,66 @@ sub loadavg_too_high {
$loadavg > $self->max_loadavg());
}
{
my $cmd;
sub loadavg_cmd {
if(not $cmd) {
# aix => "ps -ae -o state,command" # state wrong
# bsd => "ps ax -o state,command"
# sysv => "ps -ef -o s -o comm"
# dec_osf => bsd
# dragonfly => bsd
# freebsd => bsd
# gnu => bsd
# hpux => ps -el|awk '{print $2,$14,$15}'
# irix => ps -ef -o state -o comm
# linux => bsd
# minix => ps el|awk '{print \$1,\$11}'
# mirbsd => bsd
# netbsd => bsd
# openbsd => bsd
# solaris => sysv
# svr5 => sysv
# ultrix => ps -ax | awk '{print $3,$5}'
# unixware => ps -el|awk '{print $2,$14,$15}'
my $ps = q{
$sysv="ps -ef -o s -o comm";
$sysv2="ps -ef -o state -o comm";
$bsd="ps ax -o state,command";
$psel="ps -el|awk '{ print \$2,\$14,\$15 }'";
$dummy="echo S COMMAND;echo R dummy";
%ps=(
'aix' => "uptime",
'cygwin' => $sysv,
'darwin' => $bsd,
'dec_osf' => $sysv2,
'dragonfly' => $bsd,
'freebsd' => $bsd,
'gnu' => $bsd,
'hpux' => $psel,
'irix' => 'ps -ef -o state -o comm',
'linux' => $bsd,
'minix' => "ps el|awk '{print \$1,\$11}'",
'mirbsd' => $bsd,
'msys' => $sysv,
'netbsd' => $bsd,
'nto' => $dummy,
'openbsd' => $bsd,
'solaris' => $sysv,
'svr5' => $psel,
'ultrix' => "ps -ax | awk '{print \$3,\$5}'",
);
print `$ps{$^O}`;
};
$ps =~ s/[ \t\n]+/ /g;
$cmd = "perl -e ".::shell_quote_scalar($ps);
$cmd =~ s/\^/\\^/g;
}
return $cmd;
}
}
sub loadavg {
# If the currently know loadavg is too old:
# Recompute a new one in the background
@ -3741,21 +3800,28 @@ sub loadavg {
local $/ = undef;
my $load_out = <$load_fh>;
close $load_fh;
my $load =()= ($load_out=~/(^[DR]....[^\[])/gm);
my $load =()= ($load_out=~/(^\s?[DOR]\S* [^\[])/gm);
if($load > 0) {
# load is overestimated by 1
$self->{'loadavg'} = $load - 1;
::debug("load", "New loadavg: ", $self->{'loadavg'});
::debug("load", "New loadavg: ", $self->{'loadavg'},"\n");
} elsif ($load_out=~/average: (\d+.\d+)/) {
# Aix does not support instant load average
# 04:11AM up 21 days, 12:55, 1 user, load average: 1.85, 1.57, 1.55
$self->{'loadavg'} = $1;
} else {
::die_bug("loadavg_invalid_content: $load_out");
::die_bug("loadavg_invalid_content: " .
$self->{'loadavg_file'} . "\n$load_out");
}
::debug("load", "Last update: ", $self->{'last_loadavg_update'});
if(time - $self->{'last_loadavg_update'} > 10) {
# last loadavg was started 10 seconds ago
::debug("load", time - $self->{'last_loadavg_update'}, " secs old: ",
$self->{'loadavg_file'});
# Because of instant load average, it should not be delayed 10 secs
# The instant load does not give 2 R if there is only 1 cpu.
# ::debug("load", "Last update: ", $self->{'last_loadavg_update'});
# if(time - $self->{'last_loadavg_update'} > 10) {
# # last loadavg was started 10 seconds ago
# ::debug("load", time - $self->{'last_loadavg_update'}, " secs old: ",
# $self->{'loadavg_file'});
$update_loadavg_file = 1;
}
# }
} else {
::debug("load", "No loadavg file: ", $self->{'loadavg_file'});
$self->{'loadavg'} = undef;
@ -3768,12 +3834,15 @@ sub loadavg {
-e $ENV{'HOME'}."/.parallel/tmp" or mkdir $ENV{'HOME'}."/.parallel/tmp";
my $cmd = "";
if($self->{'string'} ne ":") {
$cmd = $self->sshcommand() . " " . $self->serverlogin() . " ";
$cmd = $self->sshcommand() . " " . $self->serverlogin() . " " .
::shell_quote_scalar(loadavg_cmd());
} else {
$cmd .= loadavg_cmd();
}
# TODO Is is called 'ps ax -o state,command' on other platforms?
$cmd .= "ps ax -o state,command";
# $cmd .= "ps ax -o state,command";
# As the command can take long to run if run remote
# save it to a tmp file before moving it to the correct file
::debug("load", "Cmd: ", $cmd);
my $file = $self->{'loadavg_file'};
my ($dummy_fh, $tmpfile) = ::tmpfile(SUFFIX => ".loa");
qx{ ($cmd > $tmpfile && mv $tmpfile $file || rm $tmpfile) & };