static void makeflow_node_export_variables( struct dag *d, struct dag_node *n )
{
struct jx *j = dag_node_env_create(d,n);
if(j) {
jx_export(j);
jx_delete(j);
}
}
static batch_job_id_t batch_job_condor_submit (struct batch_queue *q, const char *cmd, const char *extra_input_files, const char *extra_output_files, struct jx *envlist, const struct rmsummary *resources )
{
FILE *file;
int njobs;
int jobid;
const char *options = hash_table_lookup(q->options, "batch-options");
if(setup_condor_wrapper("condor.sh") < 0) {
debug(D_BATCH, "could not create condor.sh: %s", strerror(errno));
return -1;
}
if(!string_istrue(hash_table_lookup(q->options, "skip-afs-check"))) {
char *cwd = path_getcwd();
if(!strncmp(cwd, "/afs", 4)) {
debug(D_NOTICE|D_BATCH, "The working directory is '%s':", cwd);
debug(D_NOTICE|D_BATCH, "This won't work because Condor is not able to write to files in AFS.");
debug(D_NOTICE|D_BATCH, "Instead, run makeflow from a local disk like /tmp.");
debug(D_NOTICE|D_BATCH, "Or, use the Work Queue with -T wq and condor_submit_workers.");
free(cwd);
exit(EXIT_FAILURE);
}
free(cwd);
}
file = fopen("condor.submit", "w");
if(!file) {
debug(D_BATCH, "could not create condor.submit: %s", strerror(errno));
return -1;
}
fprintf(file, "universe = vanilla\n");
fprintf(file, "executable = condor.sh\n");
char *escaped = string_escape_condor(cmd);
fprintf(file, "arguments = %s\n", escaped);
free(escaped);
if(extra_input_files)
fprintf(file, "transfer_input_files = %s\n", extra_input_files);
// Note that we do not use transfer_output_files, because that causes the job
// to get stuck in a system hold if the files are not created.
fprintf(file, "should_transfer_files = yes\n");
fprintf(file, "when_to_transfer_output = on_exit\n");
fprintf(file, "notification = never\n");
fprintf(file, "copy_to_spool = true\n");
fprintf(file, "transfer_executable = true\n");
fprintf(file, "keep_claim_idle = 30\n");
fprintf(file, "log = %s\n", q->logfile);
const char *c_req = batch_queue_get_option(q, "condor-requirements");
char *bexp = blacklisted_expression(q);
if(c_req && bexp) {
fprintf(file, "requirements = %s && %s\n", c_req, bexp);
} else if(c_req) {
fprintf(file, "requirements = %s\n", c_req);
} else if(bexp) {
fprintf(file, "requirements = %s\n", bexp);
}
if(bexp)
free(bexp);
/*
Getting environment variables formatted for a condor submit
file is very hairy, due to some strange quoting rules.
To avoid problems, we simply export vars to the environment,
and then tell condor getenv=true, which pulls in the environment.
*/
fprintf(file, "getenv = true\n");
if(envlist) {
jx_export(envlist);
}
if(options)
fprintf(file, "%s\n", options);
/* set same deafults as condor_submit_workers */
int64_t cores = 1;
int64_t memory = 1024;
int64_t disk = 1024;
if(resources) {
cores = resources->cores > -1 ? resources->cores : cores;
memory = resources->memory > -1 ? resources->memory : memory;
disk = resources->disk > -1 ? resources->disk : disk;
}
/* convert disk to KB */
disk *= 1024;
if(batch_queue_get_option(q, "autosize")) {
fprintf(file, "request_cpus = ifThenElse(%" PRId64 " > TotalSlotCpus, %" PRId64 ", TotalSlotCpus)\n", cores, cores);
fprintf(file, "request_memory = ifThenElse(%" PRId64 " > TotalSlotMemory, %" PRId64 ", TotalSlotMemory)\n", memory, memory);
fprintf(file, "request_disk = ifThenElse((%" PRId64 ") > TotalSlotDisk, (%" PRId64 "), TotalSlotDisk)\n", disk, disk);
}
else {
fprintf(file, "request_cpus = %" PRId64 "\n", cores);
fprintf(file, "request_memory = %" PRId64 "\n", memory);
fprintf(file, "request_disk = %" PRId64 "\n", disk);
}
fprintf(file, "queue\n");
fclose(file);
file = popen("condor_submit condor.submit", "r");
if(!file)
return -1;
char line[BATCH_JOB_LINE_MAX];
while(fgets(line, sizeof(line), file)) {
if(sscanf(line, "%d job(s) submitted to cluster %d", &njobs, &jobid) == 2) {
pclose(file);
debug(D_BATCH, "job %d submitted to condor", jobid);
struct batch_job_info *info;
info = malloc(sizeof(*info));
memset(info, 0, sizeof(*info));
info->submitted = time(0);
itable_insert(q->job_table, jobid, info);
return jobid;
}
}
pclose(file);
debug(D_BATCH, "failed to submit job to condor!");
return -1;
}
static batch_job_id_t batch_job_cluster_submit (struct batch_queue * q, const char *cmd, const char *extra_input_files, const char *extra_output_files, struct jx *envlist, const struct rmsummary *resources )
{
batch_job_id_t jobid;
struct batch_job_info *info;
const char *options = hash_table_lookup(q->options, "batch-options");
if(!setup_batch_wrapper(q, cluster_name)) {
debug(D_NOTICE|D_BATCH,"couldn't setup wrapper file: %s",strerror(errno));
return 0;
}
/* Use the first word in the command line as a name for the job. */
char *name = xxstrdup(cmd);
{
char *s = strchr(name, ' ');
if(s)
*s = 0;
}
/*
Experiment shows that passing environment variables
through the command-line doesn't work, due to multiple
levels of quote interpretation. So, we export all
variables into the environment, and rely upon the -V
option to load the environment into the job.
*/
if(envlist) {
jx_export(envlist);
}
/*
Pass the command to run through the environment as well.
*/
setenv("BATCH_JOB_COMMAND", cmd, 1);
char *command = string_format("%s %s %s '%s' %s %s.wrapper",
cluster_submit_cmd,
cluster_options,
cluster_jobname_var,
path_basename(name),
options ? options : "",
cluster_name);
free(name);
debug(D_BATCH, "%s", command);
FILE *file = popen(command, "r");
free(command);
if(!file) {
debug(D_BATCH, "couldn't submit job: %s", strerror(errno));
return -1;
}
char line[BATCH_JOB_LINE_MAX] = "";
while(fgets(line, sizeof(line), file)) {
if(sscanf(line, "Your job %" SCNbjid, &jobid) == 1
|| sscanf(line, "Submitted batch job %" SCNbjid, &jobid) == 1
|| sscanf(line, "%" SCNbjid, &jobid) == 1 ) {
debug(D_BATCH, "job %" PRIbjid " submitted", jobid);
pclose(file);
info = malloc(sizeof(*info));
memset(info, 0, sizeof(*info));
info->submitted = time(0);
itable_insert(q->job_table, jobid, info);
return jobid;
}
}
if(strlen(line)) {
debug(D_NOTICE, "job submission failed: %s", line);
} else {
debug(D_NOTICE, "job submission failed: no output from %s", cluster_name);
}
pclose(file);
return -1;
}
