我一直在使用Perl::Net::SSH在我的远程机器上自动运行一些脚本。然而,其中一些脚本需要很长时间才能完成(一两个小时),有时,我停止从它们获取数据,而实际上并没有失去连接。
下面是我使用的代码:sub run_regression_tests {
for(my $i = 0; $i < @servers; $i++){
my $inner = $users[$i];
foreach(@$inner){
my $user = $_;
my $server = $servers[$i];
my $outFile;
open($outFile, ">" . $outputDir . $user . "@" . $server . ".log.txt");
print $outFile "Opening connection to $user at $server on " . localtime() . "nn";
close($outFile);
my $pid = $pm->start and next;
print "Connecting to $user@" . "$server...n";
my $hasWentToDownloadYet = 0;
my $ssh = Net::SSH::Perl->new($server, %sshParams);
$ssh->login($user, $password);
$ssh->register_handler("stdout", sub {
my($channel, $buffer) = @_;
my $outFile;
open($outFile, ">>", $outputDir . $user . "@" . $server . ".log.txt");
print $outFile $buffer->bytes;
close($outFile);
my @lines = split("n", $buffer->bytes);
foreach(@lines){
if($_ =~ m/REGRESSION TEST IS COMPLETE/){
$ssh->_disconnect();
if(!$hasWentToDownloadYet){
$hasWentToDownloadYet = 1;
print "Caught exit signal.n";
print("Regression tests for ${user}@${server} finised.n");
download_regression_results($user, $server);
$pm->finish;
}
}
}
});
$ssh->register_handler("stderr", sub {
my($channel, $buffer) = @_;
my $outFile;
open($outFile, ">>", $outputDir . $user . "@" . $server . ".log.txt");
print $outFile $buffer->bytes;
close($outFile);
});
if($debug){
$ssh->cmd('tail -fn 40 /GDS/gds/gdstest/t-gds-master/bin/comp.reg');
}else{
my ($stdout, $stderr, $exit) = $ssh->cmd('. ./.profile && cleanall && my.comp.reg');
if(!$exit){
print "SSH connection failed for ${user}@${server} finised.n";
}
}
#$ssh->cmd('. ./.profile');
if(!$hasWentToDownloadYet){
$hasWentToDownloadYet = 1;
print("Regression tests for ${user}@${server} finised.n");
download_regression_results($user, $server);
}
$pm->finish;
}
}
sleep(1);
print "nnnAll tests started. Tests typically take 1 hour to complete.n";
print "If they take significantly less time, there could be an error.n";
print "nnNo output will be printed until all commands have executed and finished.n";
print "If you wish to watch the progress tail -f one of the logs this script produces.n Example:nt" . 'tail -f ./gds1@tdgds10.log.txt' . "n";
$pm->wait_all_children;
print "nnAll Tests are Finished. n";
}
这里是%sshParams:
my %sshParams = (
protocol => '2',
port => '22',
options => [
"TCPKeepAlive yes",
"ConenctTimeout 10",
"BatchMode yes"
]
);
有时候,某个长时间运行的命令只是随机地停止打印/触发stdout或stderr事件,并且永远不会退出。ssh连接不会死(据我所知),因为$ssh->cmd
仍然阻塞。
知道如何纠正这种行为吗?
在%sshParams散列中,您可能需要将"TCPKeepAlive yes"添加到您的选项中:
$sshParams{'options'} = ["BatchMode yes", "TCPKeepAlive yes"];
这些选项可能适合您,也可能不适合您,但我建议为任何长时间运行的SSH连接设置TCPKeepAlive。如果您的路径中有任何类型的有状态防火墙,如果它长时间没有通过连接传递流量,它可能会丢弃状态。
失败可能是由于您查看REGRESSION TEST IS COMPLETE
标记输出的方式。它可能被分散在两个不同的SSH数据包中,所以你的回调永远找不到它。
更好的方法是使用一个远程命令,该命令在完成后结束,就像这样一行:
perl -pe 'BEGIN {$p = open STDIN, "my.comp.reg |" or die $!}; kill TERM => -$p if /REGRESSION TEST IS COMPLETE/}'
否则,您将关闭远程连接,但不会停止将保持活动的远程进程。
除此之外,您应该尝试使用Net::OpenSSH或Net::OpenSSH::Parallel来代替Net::SSH::Perl:
use Net::OpenSSH::Parallel;
my $pssh = Net::OpenSSH::Parallel->new;
for my $i (0..$#server) {
my $server = $server[$i];
for my $user (@{$users[$ix]}) {
$pssh->add_host("$user@$server", password => $password);
}
}
if ($debug) {
$pssh->all(cmd => { stdout_file => "$outputDir%USER%@%HOST%.log.txt",
stderr_to_stdout => 1 },
'fail -fn 40 /GDS/gds/gdstest/t-gds-master/bin/comp.reg');
}
else {
$pssh->all(cmd => { stdout_file => "$outputDir%USER%@%HOST%.log.txt",
stderr_to_stdout => 1 },
'. ./.profile && cleanall && my.comp.reg');
}
$pssh->all(scp_get => $remote_regression_results_path, "regression_results/%USER%@%HOST%/");
$pssh->run;