Changeset 37102 for trunk/ippScripts/scripts/sc_remote_exec.pl

Timestamp:

Jul 22, 2014, 3:50:46 PM (12 years ago)

Author:

watersc1

Message:

path change, trace drop, config typo fix, stack psf fix, exec force poll addition, exec drop/fails (likely needs further tuning), transfer shift to rsync return handler.

File:

• trunk/ippScripts/scripts/sc_remote_exec.pl (modified) (14 diffs)

trunk/ippScripts/scripts/sc_remote_exec.pl

@@ -18 +18 @@
 my $SEC_HOST = 'mu-fe4';
 my $IPP_PATH = '/turquoise/usr/projects/cosmo/mswarren/ipp/';
-my $remote_root  = '/scratch3/watersc1/';
-
+#my $remote_root  = '/scratch3/watersc1/';
+my $remote_root = '/lustre/scratch1/turquoise/watersc1/ps1/';
 # tools
 my $missing_tools;
@@ -33 +33 @@
 
 # Options
-my ($remote_id,$path_base,$policy,$poll,$job_id,$dbname,$verbose,$no_update,$camera);
-
+my ($remote_id,$path_base,$policy,$poll,$force,$job_id,$dbname,$verbose,$no_update,$camera);
+$force = 0;
+$verbose = 0;
 GetOptions(
     'remote_id=s'   => \$remote_id,
@@ -41 +42 @@
     'policy=s'      => \$policy,
     'poll'          => \$poll,
+    'force-poll'    => \$force,
     'camera=s'      => \$camera,
     'dbname=s'      => \$dbname,
@@ -61 +63 @@
 
 print "passed authentication challenge.\n";
+
+# Phase 1.5: Grab the information about this run;
+my $rt_cmd = "$remotetool -listrun -remote_id $remote_id ";
+$rt_cmd   .= " -dbname $dbname " if defined($dbname);
+my ( $rt_success, $rt_error_code, $rt_full_buf, $rt_stdout_buf, $rt_stderr_buf ) =
+    run(command => $rt_cmd, verbose => $verbose);
+unless ($rt_success) {
+    $rt_error_code = (($rt_error_code >> 8) or $PS_EXIT_PROG_ERROR);
+    &my_die("UNable to run remotetool to determine remote run status",
+            $remote_id,$rt_error_code);
+}
+my $rrData = $mdcParser->parse(join "", @$rt_stdout_buf) or 
+    &my_die("UNable to run remotetool to determine remote run status",
+            $remote_id,$rt_error_code);
+my $rrData2 = parse_md_list($rrData);
+my $runData = ${ $rrData2 }[0]; # There should be only one
 
 # These are operations we only need to do on the first call, not on subsequent poll operations.
@@ -92 +110 @@
     my (undef,$remote_check) = uri_convert($uri_check);
     my (undef,$remote_transfer)= uri_convert($uri_transfer);
-#    my $ssh_check_stdout = ssh_exec_command("${remote_root}/sc_validate_files.pl --check_list $remote_check --transfer_list $remote_transfer");
+
     my $ssh_check_stdout = ssh_exec_command("${remote_root}/sc_transfer_tool.pl --input $remote_check --fetch $remote_transfer");
     # We no longer need to parse this output, as it retrieves files itself.
@@ -130 +148 @@
 # Poll the job status?
 my $poll_count = 0;
-my $poll_max   = 5;
+my $poll_max   = 1; # If it isn't finished now, it's unlikely to magically finish in the next five minutes.
 my $poll_sleep = 60;
 my $poll_response = 0;
-while (($poll_count < $poll_max)&&($poll_response != 1)&&($poll_response != -1)) {
+if ($force) {
+    $poll_response = 1;
+}
+my $poll_word  = "nothing";
+while (($poll_count < $poll_max)&&($poll_response != 1)&&($poll_response != -1)&&($force != 1)) {
     if ($verbose) {
         print "Polling code.  Sleeping.  $poll_count of $poll_max $poll_response\n";
@@ -140 +162 @@
         sleep($poll_sleep);
     }
-    $poll_response = poll_job($job_id);
+    ($poll_response,$poll_word) = poll_job($job_id);
 
     if ($poll_response == 1) { # This job has completed
@@ -146 +168 @@
     }
     elsif ($poll_response == -1) { # This job has an error
-        &my_die("The job exited incorrectly.",$remote_id, $PS_EXIT_PROG_ERROR);
+        warn("The job exited incorrectly: $remote_id $job_id $poll_word.  Attempting to recover what we can.");
+        $poll_response = 1; # To fool the check later into proceding
+#       &my_die("The job exited incorrectly.",$remote_id, $PS_EXIT_PROG_ERROR);
     }
     $poll_count++;
@@ -165 +189 @@
 my $uri_return  = $path_base . ".return";
 my ($ipp_return,$remote_return) = uri_convert($uri_return);
-scp_get($remote_return,$ipp_return);
-# Feed teh return list into the 
-my $return_push_output = ssh_exec_command("${remote_root}/sc_transfer_tool.pl --input $remote_return");
-print "$return_push_output\n";
-
-
+# We shouldn't need to get this file.  We made that file.
+#scp_get($remote_return,$ipp_return);
+
+# Check to see if we can resume.  Pack the files into an array
+my @returned_files = ();
+my @dbinfo_files   = ();
+my @dbindex        = ();
+my $iter           = 0;
 open(RETURN,   "$ipp_return") || &my_die("Couldn't open file? $ipp_return", $remote_id,$PS_EXIT_SYS_ERROR);
-while(<RETURN>) {
+while (<RETURN>) {
     chomp;
+    push @returned_files, $_;
     if ($_ =~ /dbinfo/) {
-        my $file = $_;
-#       $file =~ s/${remote_root}.tmp//; THis shouldn't be needed, as the transfer tool munges the files on that side.
-        open(DBF,"$file") || warn "Missing file $file\n";
-        my $cmd = <DBF>;
-        chomp($cmd);
-        $cmd =~ s/-/ -/g; # This is just a safety check for missing space.
-        my ( $success, $error_code, $full_buf, $stdout_buf, $stderr_buf ) =
-            run(command => $cmd, verbose => $verbose);
-        unless ($success) {
-            # This shouldn't fail, but we also can't suddenly abort if something does fail.  Maybe retry if this looks to be a continuing issue?
-            $error_code = (($error_code >> 8) or $PS_EXIT_PROG_ERROR);
-            warn("The command that shouldn't fail has failed: $cmd $remote_id $error_code");##,$remote_id,$error_code);
-        }
-        close(DBF);
-    }
+        push @dbinfo_files, $_;
+        push @dbindex, $iter;
+    }
+    $iter++;
+
 }
 close(RETURN);
+
+# Binary search it.  BUT: We're binary searching binary values.  So,
+# this has to be far more careful than a fully sorted case would be.
+# Only scan the dbinfo files as those are the things that are
+# generated after the processing, and so must be the last things to
+# exist.
+
+my $first = 0;
+my $last  = $#dbinfo_files;
+my $mid   = 0;
+my $range = 20;  # How wide of a chunk to look through to catch threading drops
+my $offset = 0;  # Offset to start from.
+if (( stat($dbinfo_files[$last]) )[7] != 0) { $offset = $#dbinfo_files;}
+else {
+    my $count = 0;
+    while ($first < $last - $range) {
+
+        $mid = int(($last + $first)/2);
+        for (my $i = 0; $i <= $range; $i++) {
+            my $size = (stat($dbinfo_files[$mid + $i]))[7];
+#           print STDERR "$first $last $mid $i $size $dbinfo_files[$mid + $i]\n";
+            if ($size == 0) { 
+                if ($last == $mid + $i) { 
+                    $first = $last - 1;
+                }
+                $last = $mid + $i; 
+                $i = $range + 10;
+            }
+        }
+        if ($last > $mid + $range) { $first = $mid + $range; }
+        $count++;
+        if ($count > 200) { 
+            &my_die("The iteration loop couldn't converge after 200 rounds, suggesting something has gone wrong.",
+                    $remote_id,4,"error_run");
+        }
+    }
+    $offset = $mid - $range; # Because I want to make sure I get to the beginning of the thread group.
+    $offset = $dbindex[$offset]; # Because we need to convert from dbinfo space to returned file space.
+}
+
+# Feed teh return list into the transfer tool
+# If this fails, we should go to an auth wait point.
+my $return_push_output = ssh_exec_command("${remote_root}/sc_transfer_tool.pl --input $remote_return --offset $offset");
+my $return_push_output_print = join "\n", @{ $return_push_output };
+print "$return_push_output_print\n";
+
+#die();
+
+foreach my $file (@dbinfo_files) {
+    open(DBF,"$file") || warn "Missing file $file\n";
+    my $cmd = <DBF>;
+    close(DBF);
+    chomp($cmd);
+    $cmd =~ s/-/ -/g; # This is just a safety check for missing space.
+    my ( $success, $error_code, $full_buf, $stdout_buf, $stderr_buf ) =
+        run(command => $cmd, verbose => $verbose);
+    unless ($success) {
+        # This shouldn't fail, but we also can't suddenly abort if something does fail.  
+        # Now we're going to drop the component that owns this dbinfo, which should force a retry.
+        $error_code = (($error_code >> 8) or $PS_EXIT_PROG_ERROR);
+        warn("The command that shouldn't fail has failed: $cmd $remote_id $error_code");
+        drop_component($remote_id,$cmd,$file);
+    }
+
+}
 
 # We're done, so set the state and exit.
 &my_die("Finished",$remote_id,0,"full");
-
 
 
@@ -209 +290 @@
         $error_code = (($error_code >> 8) or $PS_EXIT_PROG_ERROR);
         &my_die("Authorization check failed.",$remote_id,0,'auth');
-
-#       my $update_command = "$remotetool -updaterun -remote_id $remote_id ";
-#       $update_command .= " -set_state auth ";
-#       $update_command .= " -dbname $dbname " if defined $dbname;
-#       system($update_command);
-#       exit(0);  # This isn't an error.  It's a state we monitor and move to automatically.
     }
 }
@@ -250 +325 @@
 sub ssh_exec_command {
     my $cmd = shift;
-    $cmd = "$ssh $DMZ_HOST ssh ${SEC_HOST} $cmd";
+    $cmd = "$ssh  $DMZ_HOST ssh  ${SEC_HOST} $cmd";
     my ( $success, $error_code, $full_buf, $stdout_buf, $stderr_buf ) =
         run(command => $cmd, verbose => $verbose);
     unless ($success) {
         $error_code = (($error_code >> 8) or $PS_EXIT_PROG_ERROR);
+        # If we failed, see if we failed due to authorization.
+        check_ssh_connection(); 
         &my_die("Failed to execute command: >>$cmd<<"); 
     }
@@ -266 +343 @@
     
     if ($response =~ /State: Running/) { 
-        return(0);
+        return(0,"Running");
     }
     elsif ($response =~ /State: Completed/) {
-        return(1);
+        return(1,"Completed");
     }
     elsif ($response =~ /State: Idle/) {
-        return(-2);
+        return(-2,"Idle");
     }
     else {
-        return(-1);
+        my $response_word = $response;
+        $response_word =~ s/.*State: //;
+        $response_word =~ s/\n.*//;
+        return(-1,$response_word);
     }
 }
@@ -324 +404 @@
     return($local_uri);
 }
+
+sub drop_component {
+    my $remote_id = shift;
+    my $cmd = shift;
+    my $file = shift;
+
+    if ($cmd eq '') { 
+        # Zero byte file.  This dbinfo was never constructed properly.
+        
+        # Steal the stage_id from the filename.  This isn't something I'm happy about.
+        my $stage = $runData->{stage};
+        my $stage_id = $file;
+        if ($stage eq 'chip') { 
+            $stage_id =~ s/^.*ch\.//;
+        }
+        elsif ($stage eq 'camera') {
+            $stage_id =~ s/^.*cm\.//;
+        }
+        elsif ($stage eq 'warp') {
+            $stage_id =~ s/^.*wrp\.//;
+        }
+        elsif ($stage eq 'stack') {
+            $stage_id =~ s/^.*stk\.//;
+        }
+        $stage_id =~ s/\..*$//;
+        
+        # Drop this from the database
+        my $drop_cmd = "$remotetool -dropcomponent -remote_id $remote_id -stage_id $stage_id ";
+        $drop_cmd   .= " -dbname $dbname " if defined($dbname);
+        system($drop_cmd);
+    }
+
+}
+
 sub my_die {
     my $msg = shift;
@@ -347 +461 @@
 }
 
-# I don't think I need this anymore, but I don't want to burn it until I"m sure.
-
-# exit(1);
-# # Read this as a metadata object
-# my $ssh_exec_stdout = ssh_exec_command("sc_validate_processing.pl ${path_base}.out");
-# my $retrieveData = $mdcParser->parse(join "", @$ssh_exec_stdout);
-# my $retrieveData2= parse_md_list($retrieveData);
-
-# foreach my $retrieveEntry (@$retrieveData2) {
-#     my $process_fault = $retrieveEntry->{fault};
-#     my $dbcommand = $retrieveEntry->{dbcommand};
-#     my $filelist  = $retrieveEntry->{filelist};
-
-#     scp_get("${filelist}",${filelist});
-#     open(FL,${filelist});
-#     while(<FL>) {
-#       chomp;
-#       my $f = $_;
-#       scp_get("${f}",$f);
-#     }
-#     close(FL);
-
-#     my ( $success, $error_code, $full_buf, $stdout_buf, $stderr_buf ) =
-#       run(command => $dbcommand, verbose => $verbose);
-#     unless ($success) {
-#       $error_code = (($error_code >> 8) or $PS_EXIT_PROG_ERROR);
-#       &my_die("The database command returned did not execute correctly. $dbcommand $process_fault",
-#               $remote_id,$error_code); 
-#     }
-
-#     if ($process_fault != 0) {
-#       $fault = 1;
-#     }
-# }
-
-# if (($policy ne 'ignore')&&($fault != 0)) {
-#     &my_die("This job exited without all components finishing successfully.  Halting.",
-#           $remote_id,$fault,"halt");
-# }
-# else {
-#     print "This job exited without all components finishing successfully. $fault\n" if $fault;
-#     &my_die("This job finished.  Setting remote job to full.",
-#           $remote_id,0,"full");
-# }
-
-# Promote to finished.
-#&my_die(); # Not really, but I don't have the tool interface finished, so these are all just placeholders.
 # Quick review:
 # new -> pending -> run -> full