inline static dt::string call(vm_heap &heap, const dt::string &command) {
// Set stdout to be a temporary file.
FILE *tmpfp = tmpfile();
NP1_ASSERT(tmpfp, "Unable to create temporary file for meta.shell function");
::np1::io::file tmpf;
tmpf.from_handle(fileno(tmpfp));
int saved_stdout = dup(1);
dup2(tmpf.handle(), 1);
// Execute the command.
NP1_ASSERT(system(command.to_string().c_str()) >= 0, "system() failed for meta.shell");
// Set stdout back to what it was and read the whole file.
dup2(saved_stdout, 1);
close(saved_stdout);
tmpf.rewind();
::np1::io::mandatory_input_stream<np1::io::file> mandatory_tmpf(tmpf);
::np1::io::ext_heap_buffer_output_stream<vm_heap> command_output(heap, TEMP_FILE_READ_BUFFER_SIZE);
mandatory_tmpf.copy(command_output);
// Ensure that the command output is valid UTF-8.
char *command_output_p = (char *)command_output.ptr();
size_t command_output_size = command_output.size();
str::replace_invalid_utf8_sequences(command_output_p, command_output_size, '?');
return dt::string(command_output_p, command_output_size);
}
static bool
submit_try( ArgList &args, CondorID &condorID, bool prohibitMultiJobs )
{
MyString cmd; // for debug output
args.GetArgsStringForDisplay( &cmd );
FILE * fp = my_popen( args, "r", TRUE );
if (fp == NULL) {
debug_printf( DEBUG_NORMAL,
"ERROR: my_popen(%s) in submit_try() failed!\n",
cmd.Value() );
return false;
}
//----------------------------------------------------------------------
// Parse submit command output for a HTCondor job ID. This
// desperately needs to be replaced by HTCondor submit APIs.
//
// Typical condor_submit output for HTCondor v6 looks like:
//
// Submitting job(s).
// Logging submit event(s).
// 1 job(s) submitted to cluster 2267.
//----------------------------------------------------------------------
char buffer[UTIL_MAX_LINE_LENGTH];
buffer[0] = '\0';
// Configure what we look for in the command output according to
// which type of job we have.
const char *marker = NULL;
parse_submit_fnc parseFnc = NULL;
marker = " submitted to cluster ";
// Note: we *could* check how many jobs got submitted here, and
// correlate that with how many submit events we see later on.
// I'm not worrying about that for now... wenger 2006-02-07.
// We also have to check the number of jobs to get an accurate
// count of submitted jobs to report in the dagman.out file.
// We should also check whether we got more than one cluster, and
// either deal with it correctly or generate an error message.
parseFnc = parse_condor_submit;
// Take all of the output (both stdout and stderr) from condor_submit,
// and echo it to the dagman.out file. Look for
// the line (if any) containing the word "cluster" (HTCondor).
// If we don't find such a line, something
// went wrong with the submit, so we return false. The caller of this
// function can retry the submit by repeatedly calling this function.
MyString command_output("");
MyString keyLine("");
while (fgets(buffer, UTIL_MAX_LINE_LENGTH, fp)) {
MyString buf_line = buffer;
buf_line.chomp();
debug_printf(DEBUG_VERBOSE, "From submit: %s\n", buf_line.Value());
command_output += buf_line;
if (strstr(buffer, marker) != NULL) {
keyLine = buf_line;
}
}
{ // Relocated this curly bracket to its previous position to hopefully
// fix Coverity warning. Not sure why these curly brackets are here
// at all... wenger 2013-06-12
int status = my_pclose(fp) & 0xff;
if (keyLine == "") {
debug_printf(DEBUG_NORMAL, "failed while reading from pipe.\n");
debug_printf(DEBUG_NORMAL, "Read so far: %s\n", command_output.Value());
return false;
}
if (status != 0) {
debug_printf(DEBUG_NORMAL, "Read from pipe: %s\n",
command_output.Value());
debug_printf( DEBUG_QUIET, "ERROR while running \"%s\": "
"my_pclose() failed with status %d (errno %d, %s)!\n",
cmd.Value(), status, errno, strerror( errno ) );
return false;
}
}
int jobProcCount;
if ( !parseFnc( keyLine.Value(), jobProcCount, condorID._cluster) ) {
// We are going forward (do not return false here)
// Expectation is that higher levels will catch that we
// did not get a cluster initialized properly here, fail,
// and write a rescue DAG. gt3658 2013-06-03
//
// This is better than the old failure that would submit
// DAGMAN_MAX_SUBMIT_ATTEMPT copies of the same job.
debug_printf( DEBUG_NORMAL, "WARNING: submit returned 0, but "
"parsing submit output failed!\n" );
// Returning here so we don't try to process invalid values
// below. (This should really return something like "submit failed
// don't retry" -- see gittrac #3685.)
return true;
}
// Check for multiple job procs if configured to disallow that.
if ( prohibitMultiJobs && (jobProcCount > 1) ) {
debug_printf( DEBUG_NORMAL, "Submit generated %d job procs; "
"disallowed by DAGMAN_PROHIBIT_MULTI_JOBS setting\n",
jobProcCount );
main_shutdown_rescue( EXIT_ERROR, Dag::DAG_STATUS_ERROR );
}
return true;
}
