static int batch_fs_dryrun_getcwd (struct batch_queue *q, char *buf, size_t size) {
const char *cwd = batch_queue_get_option(q, "cwd");
size_t pathlength = strlen(cwd);
if (pathlength + 1 > size) {
errno = ERANGE;
return -1;
} else {
strcpy(buf, cwd);
return 0;
}
}
static int batch_queue_mesos_free(struct batch_queue *q)
{
FILE *fp;
fp = fopen(MESOS_DONE_FILE, "w");
if(fp == NULL) {
fatal("Fail to clean up batch queue. %s\n", strerror(errno));
}
int batch_queue_abort_flag = atoi(batch_queue_get_option(q, "batch-queue-abort-flag"));
int batch_queue_failed_flag = atoi(batch_queue_get_option(q, "batch-queue-failed-flag"));
if(batch_queue_abort_flag) {
fprintf(fp, "aborted");
} else if(batch_queue_failed_flag) {
fprintf(fp, "failed");
} else {
fprintf(fp, "finished");
}
fclose(fp);
return 0;
}
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_wq_wait (struct batch_queue * q, struct batch_job_info * info, time_t stoptime)
{
static int try_open_log = 0;
int timeout, taskid = -1;
if(!try_open_log)
{
try_open_log = 1;
if(!work_queue_specify_log(q->data, q->logfile))
{
return -1;
}
const char *transactions = batch_queue_get_option(q, "batch_log_transactions_name");
if(transactions) {
work_queue_specify_transactions_log(q->data, transactions);
}
}
if(stoptime == 0) {
timeout = WORK_QUEUE_WAITFORTASK;
} else {
timeout = MAX(0, stoptime - time(0));
}
struct work_queue_task *t = work_queue_wait(q->data, timeout);
if(t) {
info->submitted = t->time_when_submitted / 1000000;
info->started = t->time_when_commit_end / 1000000;
info->finished = t->time_when_done / 1000000;
info->exited_normally = 1;
info->exit_code = t->return_status;
info->exit_signal = 0;
info->disk_allocation_exhausted = t->disk_allocation_exhausted;
/*
If the standard ouput of the job is not empty,
then print it, because this is analogous to a Unix
job, and would otherwise be lost. Important for
capturing errors from the program.
*/
if(t->output && t->output[0]) {
if(t->output[1] || t->output[0] != '\n') {
string_chomp(t->output);
printf("%s\n", t->output);
}
}
char *outfile = itable_remove(q->output_table, t->taskid);
if(outfile) {
FILE *file = fopen(outfile, "w");
if(file) {
fwrite(t->output, strlen(t->output), 1, file);
fclose(file);
}
free(outfile);
}
taskid = t->taskid;
work_queue_task_delete(t);
}
if(taskid >= 0) {
return taskid;
}
if(work_queue_empty(q->data)) {
return 0;
} else {
return -1;
}
}
static batch_job_id_t batch_job_mesos_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 )
{
// Get the path to mesos python site-packages
if (!is_mesos_py_path_known) {
mesos_py_path = batch_queue_get_option(q, "mesos-path");
if (mesos_py_path != NULL) {
debug(D_INFO, "Get mesos_path %s from command line\n", mesos_py_path);
}
is_mesos_py_path_known = 1;
}
// Get the mesos master address
if (!is_mesos_master_known) {
mesos_master = batch_queue_get_option(q, "mesos-master");
if (mesos_master == NULL) {
fatal("Please specify the hostname of mesos master by using --mesos-master");
} else {
debug(D_INFO, "Get mesos_path %s from command line\n", mesos_py_path);
is_mesos_master_known = 1;
}
}
mesos_preload = batch_queue_get_option(q, "mesos-preload");
if (is_mesos_py_path_known &&
is_mesos_master_known &&
!is_scheduler_running ) {
// start mesos scheduler if it is not running
start_mesos_scheduler(q);
is_scheduler_running = 1;
}
int task_id = ++counter;
debug(D_BATCH, "task %d is ready", task_id);
struct batch_job_info *info = calloc(1, sizeof(*info));
info->started = time(0);
info->submitted = time(0);
itable_insert(q->job_table, task_id, info);
// write the ready task information as
// "task_id, task_cmd, inputs, outputs" to
// mesos_task_info, which will be scanned by
// mf_mesos_scheduler later.
FILE *task_info_fp;
if(access(FILE_TASK_INFO, F_OK) != -1) {
task_info_fp = fopen(FILE_TASK_INFO, "a+");
} else {
task_info_fp = fopen(FILE_TASK_INFO, "w+");
}
struct mesos_task *mt = mesos_task_create(task_id, cmd, extra_input_files, extra_output_files);
fprintf(task_info_fp, "%d,%s,", mt->task_id, mt->task_cmd);
if (extra_input_files != NULL && strlen(extra_input_files) != 0) {
int j = 0;
int num_input_files = text_list_size(mt->task_input_files);
for (j = 0; j < (num_input_files-1); j++) {
fprintf(task_info_fp, "%s ", text_list_get(mt->task_input_files, j));
}
fprintf(task_info_fp, "%s,", text_list_get(mt->task_input_files, num_input_files-1));
} else {
fprintf(task_info_fp, ",");
}
if (extra_output_files != NULL && strlen(extra_output_files) != 0) {
int j = 0;
int num_output_files = text_list_size(mt->task_output_files);
for (j = 0; j < (num_output_files-1); j++) {
fprintf(task_info_fp, "%s ", text_list_get(mt->task_output_files, j));
}
fprintf(task_info_fp, "%s,",text_list_get(mt->task_output_files, num_output_files-1));
} else {
fprintf(task_info_fp, ",");
}
// The default resource requirements for each task
int64_t cores = -1;
int64_t memory = -1;
int64_t disk = -1;
if (resources) {
cores = resources->cores > -1 ? resources->cores : cores;
memory = resources->memory > -1 ? resources->memory : memory;
disk = resources->disk > -1 ? resources->disk : disk;
}
fprintf(task_info_fp, "%" PRId64 ",%" PRId64 ",%" PRId64 ",", cores, memory, disk);
fputs("submitted\n", task_info_fp);
mesos_task_delete(mt);
fclose(task_info_fp);
return task_id;
}
