Merge branch 'master' into 6093-refresh-docs

[arvados.git] / sdk / cli / bin / crunch-job

diff --git a/sdk/cli/bin/crunch-job b/sdk/cli/bin/crunch-job
index 6ae04812325994678dad9e29d736b5556731f4f4..c748904105da7ade133920d7369d9da7655bb526 100755 (executable)
--- a/sdk/cli/bin/crunch-job
+++ b/sdk/cli/bin/crunch-job
@@ -597,19 +597,43 @@ else {
   my @execargs = ("sh", "-c",
                   "mkdir -p $ENV{CRUNCH_INSTALL} && cd $ENV{CRUNCH_TMP} && perl -");
 
+  $ENV{"CRUNCH_GIT_ARCHIVE_HASH"} = md5_hex($git_archive);
+  my ($install_stderr_r, $install_stderr_w);
+  pipe $install_stderr_r, $install_stderr_w or croak("pipe() failed: $!");
+  set_nonblocking($install_stderr_r);
   my $installpid = fork();
   if ($installpid == 0)
   {
+    close($install_stderr_r);
+    fcntl($install_stderr_w, F_SETFL, 0) or croak($!); # no close-on-exec
+    open(STDOUT, ">&", $install_stderr_w);
+    open(STDERR, ">&", $install_stderr_w);
     srun (\@srunargs, \@execargs, {}, $build_script . $git_archive);
     exit (1);
   }
-  while (1)
-  {
-    last if $installpid == waitpid (-1, WNOHANG);
+  close($install_stderr_w);
+  my $stderr_buf = '';
+  while ($installpid != waitpid(-1, WNOHANG)) {
     freeze_if_want_freeze ($installpid);
-    select (undef, undef, undef, 0.1);
+    # Wait up to 0.1 seconds for something to appear on stderr, then
+    # do a non-blocking read.
+    my $bits = fhbits($install_stderr_r);
+    select ($bits, undef, $bits, 0.1);
+    if (0 < sysread ($install_stderr_r, $stderr_buf, 8192, length($stderr_buf)))
+    {
+      while ($stderr_buf =~ /^(.*?)\n/) {
+        my $line = $1;
+        substr $stderr_buf, 0, 1+length($line), "";
+        Log(undef, "stderr $line");
+      }
+    }
   }
   my $install_exited = $?;
+  close($install_stderr_r);
+  if (length($stderr_buf) > 0) {
+    Log(undef, "stderr $stderr_buf")
+  }
+
   Log (undef, "Install script exited ".exit_status_s($install_exited));
   foreach my $tar_filename (map { tar_filename_n($_); } (1..$git_tar_count)) {
     unlink($tar_filename);
@@ -699,9 +723,8 @@ for (my $todo_ptr = 0; $todo_ptr <= $#jobstep_todo; $todo_ptr ++)
     next;
   }
 
-  pipe $reader{$id}, "writer" or croak ($!);
-  my $flags = fcntl ($reader{$id}, F_GETFL, 0) or croak ($!);
-  fcntl ($reader{$id}, F_SETFL, $flags | O_NONBLOCK) or croak ($!);
+  pipe $reader{$id}, "writer" or croak("pipe() failed: $!");
+  set_nonblocking($reader{$id});
 
   my $childslot = $freeslot[0];
   my $childnode = $slot[$childslot]->{node};
@@ -1861,6 +1884,12 @@ sub combined_git_archive {
   return $tar_contents;
 }
 
+sub set_nonblocking {
+  my $fh = shift;
+  my $flags = fcntl ($fh, F_GETFL, 0) or croak ($!);
+  fcntl ($fh, F_SETFL, $flags | O_NONBLOCK) or croak ($!);
+}
+
 __DATA__
 #!/usr/bin/perl
 #
@@ -1888,12 +1917,15 @@ use constant TASK_TEMPFAIL => 111;
 my %SDK_ENVVARS = ("perl/lib" => "PERLLIB", "ruby/lib" => "RUBYLIB");
 
 my $destdir = $ENV{"CRUNCH_SRC"};
-my $commit = $ENV{"CRUNCH_SRC_COMMIT"};
+my $archive_hash = $ENV{"CRUNCH_GIT_ARCHIVE_HASH"};
 my $repo = $ENV{"CRUNCH_SRC_URL"};
 my $install_dir = $ENV{"CRUNCH_INSTALL"} || (getcwd() . "/opt");
 my $job_work = $ENV{"JOB_WORK"};
 my $task_work = $ENV{"TASK_WORK"};
 
+open(STDOUT_ORIG, ">&", STDOUT);
+open(STDERR_ORIG, ">&", STDERR);
+
 for my $dir ($destdir, $job_work, $task_work) {
   if ($dir) {
     make_path $dir;
@@ -1905,11 +1937,6 @@ if ($task_work) {
   remove_tree($task_work, {keep_root => 1});
 }
 
-open(STDOUT_ORIG, ">&", STDOUT);
-open(STDERR_ORIG, ">&", STDERR);
-open(STDOUT, ">>", "$destdir.log");
-open(STDERR, ">&", STDOUT);
-
 ### Crunch script run mode
 if (@ARGV) {
   # We want to do routine logging during task 0 only.  This gives the user
@@ -1970,10 +1997,6 @@ if (@ARGV) {
     }
   }
 
-  close(STDOUT);
-  close(STDERR);
-  open(STDOUT, ">&", STDOUT_ORIG);
-  open(STDERR, ">&", STDERR_ORIG);
   exec(@ARGV);
   die "Cannot exec `@ARGV`: $!";
 }
@@ -1981,12 +2004,19 @@ if (@ARGV) {
 ### Installation mode
 open L, ">", "$destdir.lock" or die "$destdir.lock: $!";
 flock L, LOCK_EX;
-if (readlink ("$destdir.commit") eq $commit && -d $destdir) {
-  # This version already installed -> nothing to do.
+if (readlink ("$destdir.archive_hash") eq $archive_hash && -d $destdir) {
+  # This exact git archive (source + arvados sdk) is already installed
+  # here, so there's no need to reinstall it.
+
+  # We must consume our DATA section, though: otherwise the process
+  # feeding it to us will get SIGPIPE.
+  my $buf;
+  while (read(DATA, $buf, 65536)) { }
+
   exit(0);
 }
 
-unlink "$destdir.commit";
+unlink "$destdir.archive_hash";
 mkdir $destdir;
 
 if (!open(TARX, "|-", "tar", "-xC", $destdir)) {
@@ -2026,6 +2056,11 @@ if ((-d $python_dir) and can_run("python2.7") and
   close($pysdk_cfg);
 }
 
+# Hide messages from the install script (unless it fails: shell_or_die
+# will show $destdir.log in that case).
+open(STDOUT, ">>", "$destdir.log");
+open(STDERR, ">&", STDOUT);
+
 if (-e "$destdir/crunch_scripts/install") {
     shell_or_die (undef, "$destdir/crunch_scripts/install", $install_dir);
 } elsif (!-e "./install.sh" && -e "./tests/autotests.sh") {
@@ -2035,10 +2070,10 @@ if (-e "$destdir/crunch_scripts/install") {
     shell_or_die (undef, "./install.sh", $install_dir);
 }
 
-if ($commit) {
-    unlink "$destdir.commit.new";
-    symlink ($commit, "$destdir.commit.new") or die "$destdir.commit.new: $!";
-    rename ("$destdir.commit.new", "$destdir.commit") or die "$destdir.commit: $!";
+if ($archive_hash) {
+    unlink "$destdir.archive_hash.new";
+    symlink ($archive_hash, "$destdir.archive_hash.new") or die "$destdir.archive_hash.new: $!";
+    rename ("$destdir.archive_hash.new", "$destdir.archive_hash") or die "$destdir.archive_hash: $!";
 }
 
 close L;