struct batch_queue *batch_queue_create(batch_queue_type_t type)
{
int i;
struct batch_queue *q;
q = xxmalloc(sizeof(*q));
q->type = type;
strncpy(q->logfile, "", sizeof(q->logfile));
q->options = hash_table_create(0, NULL);
q->features = hash_table_create(0, NULL);
q->job_table = itable_create(0);
q->output_table = itable_create(0);
q->data = NULL;
batch_queue_set_feature(q, "local_job_queue", "yes");
batch_queue_set_feature(q, "batch_log_name", "%s.batchlog");
batch_queue_set_feature(q, "gc_size", "yes");
q->module = NULL;
for (i = 0; batch_queue_modules[i]->type != BATCH_QUEUE_TYPE_UNKNOWN; i++)
if (batch_queue_modules[i]->type == type)
q->module = batch_queue_modules[i];
if (q->module == NULL) {
batch_queue_delete(q);
return NULL;
}
if(q->module->create(q) == -1) {
batch_queue_delete(q);
return NULL;
}
debug(D_BATCH, "created queue %p (%s)", q, q->module->typestr);
return q;
}
int main(int argc, char *argv[])
{
batch_queue_type_t batch_queue_type = BATCH_QUEUE_TYPE_UNKNOWN;
catalog_host = CATALOG_HOST;
catalog_port = CATALOG_PORT;
batch_submit_options = getenv("BATCH_OPTIONS");
debug_config(argv[0]);
resources = rmsummary_create(-1);
int c;
while((c = getopt_long(argc, argv, "B:C:F:N:M:T:t:w:W:E:P:S:cd:o:O:vh", long_options, NULL)) > -1) {
switch (c) {
case 'B':
batch_submit_options = xxstrdup(optarg);
break;
case 'C':
config_file = xxstrdup(optarg);
break;
case 'F':
foremen_regex = xxstrdup(optarg);
break;
case 'N':
case 'M':
project_regex = xxstrdup(optarg);
break;
case 'T':
batch_queue_type = batch_queue_type_from_string(optarg);
if(batch_queue_type == BATCH_QUEUE_TYPE_UNKNOWN) {
fprintf(stderr, "unknown batch queue type: %s\n", optarg);
return EXIT_FAILURE;
}
break;
case 't':
worker_timeout = atoi(optarg);
break;
case 'w':
workers_min = atoi(optarg);
break;
case 'W':
workers_max = atoi(optarg);
break;
case LONG_OPT_WORKERS_PER_CYCLE:
workers_per_cycle = atoi(optarg);
break;
case LONG_OPT_TASKS_PER_WORKER:
tasks_per_worker = atof(optarg);
break;
case 'E':
extra_worker_args = xxstrdup(optarg);
break;
case LONG_OPT_CORES:
resources->cores = atoi(optarg);
break;
case LONG_OPT_AMAZON_CREDENTIALS:
amazon_credentials = xxstrdup(optarg);
break;
case LONG_OPT_AMAZON_AMI:
amazon_ami = xxstrdup(optarg);
break;
case LONG_OPT_MEMORY:
resources->memory = atoi(optarg);
break;
case LONG_OPT_DISK:
resources->disk = atoi(optarg);
break;
case LONG_OPT_GPUS:
resources->gpus = atoi(optarg);
break;
case LONG_OPT_AUTOSIZE:
autosize = 1;
break;
case LONG_OPT_FACTORY_TIMEOUT:
factory_timeout = MAX(0, atoi(optarg));
break;
case LONG_OPT_CONDOR_REQUIREMENTS:
if(condor_requirements) {
char *tmp = condor_requirements;
condor_requirements = string_format("(%s && (%s))", tmp, optarg);
free(tmp);
} else {
condor_requirements = string_format("(%s)", optarg);
}
break;
case LONG_OPT_WRAPPER:
wrapper_command = optarg;
break;
case LONG_OPT_WRAPPER_INPUT:
if(!wrapper_input) {
wrapper_input = strdup(optarg);
} else {
wrapper_input = string_format("%s,%s",wrapper_input,optarg);
}
break;
case 'P':
password_file = optarg;
break;
case 'S':
scratch_dir = optarg;
break;
case 'c':
consider_capacity = 1;
break;
case 'd':
debug_flags_set(optarg);
break;
case 'o':
debug_config_file(optarg);
break;
case 'O':
debug_config_file_size(string_metric_parse(optarg));
break;
case 'v':
cctools_version_print(stdout, argv[0]);
exit(EXIT_SUCCESS);
case 'h':
show_help(argv[0]);
exit(EXIT_SUCCESS);
default:
show_help(argv[0]);
return EXIT_FAILURE;
}
}
if(project_regex) {
using_catalog = 1;
}
else if((argc - optind) == 2) {
master_host = argv[optind];
master_port = atoi(argv[optind+1]);
}
else {
fprintf(stderr,"work_queue_factory: You must either give a project name with the -M option or master-name option with a configuration file, or give the master's host and port.\n");
show_help(argv[0]);
exit(1);
}
cctools_version_debug(D_DEBUG, argv[0]);
if(batch_queue_type == BATCH_QUEUE_TYPE_UNKNOWN) {
fprintf(stderr,"work_queue_factory: You must specify a batch type with the -T option.\n");
fprintf(stderr, "valid options:\n");
fprintf(stderr, "%s\n", batch_queue_type_string());
return 1;
}
if(config_file) {
char abs_path_name[PATH_MAX];
if(!realpath(config_file, abs_path_name)) {
fprintf(stderr, "work_queue_factory: could not resolve configuration file path: '%s'.\n", config_file);
exit(EXIT_FAILURE);
}
free(config_file);
/* From now on, read config_file from absolute path */
config_file = xxstrdup(abs_path_name);
if(!read_config_file(config_file)) {
fprintf(stderr,"work_queue_factory: There were errors in the configuration file: %s\n", config_file);
return 1;
}
}
if(workers_min>workers_max) {
fprintf(stderr,"work_queue_factory: min workers (%d) is greater than max workers (%d)\n",workers_min, workers_max);
return 1;
}
/*
Careful here: most of the supported batch systems expect
that jobs are submitting from a single shared filesystem.
Changing to /tmp only works in the case of Condor.
*/
if(!scratch_dir) {
if(batch_queue_type==BATCH_QUEUE_TYPE_CONDOR) {
scratch_dir = string_format("/tmp/wq-pool-%d",getuid());
} else {
scratch_dir = string_format("wq-pool-%d",getuid());
}
}
if(!create_dir(scratch_dir,0777)) {
fprintf(stderr,"work_queue_factory: couldn't create %s: %s",scratch_dir,strerror(errno));
return 1;
}
char cmd[1024];
sprintf(cmd,"cp \"$(which work_queue_worker)\" '%s'",scratch_dir);
if (system(cmd)) {
fprintf(stderr, "work_queue_factory: please add work_queue_worker to your PATH.\n");
exit(EXIT_FAILURE);
}
if(password_file) {
sprintf(cmd,"cp %s %s/pwfile",password_file,scratch_dir);
system(cmd);
}
if(chdir(scratch_dir)!=0) {
fprintf(stderr,"work_queue_factory: couldn't chdir to %s: %s",scratch_dir,strerror(errno));
return 1;
}
signal(SIGINT, handle_abort);
signal(SIGQUIT, handle_abort);
signal(SIGTERM, handle_abort);
signal(SIGHUP, ignore_signal);
queue = batch_queue_create(batch_queue_type);
if(!queue) {
fprintf(stderr,"work_queue_factory: couldn't establish queue type %s",batch_queue_type_to_string(batch_queue_type));
return 1;
}
batch_queue_set_option(queue, "batch-options", batch_submit_options);
batch_queue_set_option(queue, "autosize", autosize ? "yes" : NULL);
set_worker_resources_options( queue );
if (amazon_credentials != NULL) {
batch_queue_set_option(queue, "amazon-credentials", amazon_credentials);
}
if (amazon_ami != NULL) {
batch_queue_set_option(queue, "amazon-ami", amazon_ami);
}
if(condor_requirements != NULL && batch_queue_type != BATCH_QUEUE_TYPE_CONDOR) {
debug(D_NOTICE, "condor_requirements will be ignored as workers will not be running in condor.");
} else {
batch_queue_set_option(queue, "condor-requirements", condor_requirements);
}
mainloop( queue );
batch_queue_delete(queue);
return 0;
}
int main(int argc, char *argv[])
{
batch_queue_type_t batch_queue_type = BATCH_QUEUE_TYPE_UNKNOWN;
catalog_host = CATALOG_HOST;
catalog_port = CATALOG_PORT;
debug_config(argv[0]);
int c;
while((c = getopt_long(argc, argv, "F:N:M:T:t:w:W:E:P:S:cd:o:O:vh", long_options, NULL)) > -1) {
switch (c) {
case 'F':
foremen_regex = optarg;
break;
case 'N':
case 'M':
project_regex = optarg;
break;
case 'T':
batch_queue_type = batch_queue_type_from_string(optarg);
if(batch_queue_type == BATCH_QUEUE_TYPE_UNKNOWN) {
fprintf(stderr, "unknown batch queue type: %s\n", optarg);
return EXIT_FAILURE;
}
break;
case 't':
worker_timeout = atoi(optarg);
break;
case 'w':
workers_min = atoi(optarg);
break;
case 'W':
workers_max = atoi(optarg);
break;
case LONG_OPT_TASKS_PER_WORKER:
tasks_per_worker = atof(optarg);
break;
case 'E':
extra_worker_args = optarg;
break;
case LONG_OPT_CORES:
num_cores_option = xxstrdup(optarg);
break;
case LONG_OPT_MEMORY:
num_memory_option = xxstrdup(optarg);
break;
case LONG_OPT_DISK:
num_disk_option = xxstrdup(optarg);
break;
case LONG_OPT_GPUS:
num_gpus_option = xxstrdup(optarg);
break;
case 'P':
password_file = optarg;
break;
case 'S':
scratch_dir = optarg;
break;
case 'c':
consider_capacity = 1;
break;
case 'd':
debug_flags_set(optarg);
break;
case 'o':
debug_config_file(optarg);
break;
case 'O':
debug_config_file_size(string_metric_parse(optarg));
break;
case 'v':
cctools_version_print(stdout, argv[0]);
exit(EXIT_SUCCESS);
case 'h':
show_help(argv[0]);
exit(EXIT_SUCCESS);
default:
show_help(argv[0]);
return EXIT_FAILURE;
}
}
cctools_version_debug(D_DEBUG, argv[0]);
if(batch_queue_type == BATCH_QUEUE_TYPE_UNKNOWN) {
fprintf(stderr,"work_queue_pool: You must specify a batch type with the -T option.\n");
fprintf(stderr, "valid options:\n");
fprintf(stderr, "%s\n", batch_queue_type_string());
return 1;
}
if(!project_regex) {
fprintf(stderr,"work_queue_pool: You must give a project name with the -M option.\n");
return 1;
}
if(workers_min>workers_max) {
fprintf(stderr,"work_queue_pool: --min-workers (%d) is greater than --max-workers (%d)\n",workers_min,workers_max);
return 1;
}
if(tasks_per_worker < 1)
{
tasks_per_worker = num_cores_option ? atof(num_cores_option) : 1;
}
if(!scratch_dir) {
scratch_dir = string_format("/tmp/wq-pool-%d",getuid());
}
if(!create_dir(scratch_dir,0777)) {
fprintf(stderr,"work_queue_pool: couldn't create %s: %s",scratch_dir,strerror(errno));
return 1;
}
char cmd[1024];
sprintf(cmd,"cp \"$(which work_queue_worker)\" '%s'",scratch_dir);
if (system(cmd)) {
fprintf(stderr, "work_queue_pool: please add work_queue_worker to your PATH.\n");
exit(EXIT_FAILURE);
}
if(password_file) {
sprintf(cmd,"cp %s %s/pwfile",password_file,scratch_dir);
system(cmd);
}
if(chdir(scratch_dir)!=0) {
fprintf(stderr,"work_queue_pool: couldn't chdir to %s: %s",scratch_dir,strerror(errno));
return 1;
}
signal(SIGINT, handle_abort);
signal(SIGQUIT, handle_abort);
signal(SIGTERM, handle_abort);
signal(SIGHUP, ignore_signal);
struct batch_queue * queue = batch_queue_create(batch_queue_type);
if(!queue) {
fprintf(stderr,"work_queue_pool: couldn't establish queue type %s",batch_queue_type_to_string(batch_queue_type));
return 1;
}
set_worker_resources( queue );
mainloop( queue, project_regex, foremen_regex );
batch_queue_delete(queue);
return 0;
}
struct batch_queue *batch_queue_create(batch_queue_type_t type)
{
struct batch_queue *q;
if(type == BATCH_QUEUE_TYPE_UNKNOWN)
return 0;
q = malloc(sizeof(*q));
q->type = type;
q->options_text = 0;
q->job_table = itable_create(0);
q->output_table = itable_create(0);
if(type == BATCH_QUEUE_TYPE_CONDOR)
q->logfile = strdup("condor.logfile");
else if(type == BATCH_QUEUE_TYPE_WORK_QUEUE || type == BATCH_QUEUE_TYPE_WORK_QUEUE_SHAREDFS)
q->logfile = strdup("wq.log");
else
q->logfile = NULL;
if(type == BATCH_QUEUE_TYPE_WORK_QUEUE || type == BATCH_QUEUE_TYPE_WORK_QUEUE_SHAREDFS) {
q->work_queue = work_queue_create(0);
if(!q->work_queue) {
batch_queue_delete(q);
return 0;
}
} else {
q->work_queue = 0;
}
if(type == BATCH_QUEUE_TYPE_MPI_QUEUE) {
q->mpi_queue = mpi_queue_create(0);
if(!q->mpi_queue) {
batch_queue_delete(q);
return 0;
}
} else {
q->mpi_queue = 0;
}
if(type == BATCH_QUEUE_TYPE_SGE || type == BATCH_QUEUE_TYPE_MOAB || type == BATCH_QUEUE_TYPE_TORQUE || type == BATCH_QUEUE_TYPE_CLUSTER) {
batch_job_setup_cluster(q);
}
if(type == BATCH_QUEUE_TYPE_HADOOP) {
int fail = 0;
if(!getenv("HADOOP_HOME")) {
debug(D_NOTICE, "error: environment variable HADOOP_HOME not set\n");
fail = 1;
}
if(!getenv("HDFS_ROOT_DIR")) {
debug(D_NOTICE, "error: environment variable HDFS_ROOT_DIR not set\n");
fail = 1;
}
if(!getenv("HADOOP_USER_TMP")) {
debug(D_NOTICE, "error: environment variable HADOOP_USER_TMP not set\n");
fail = 1;
}
if(!getenv("HADOOP_PARROT_PATH")) {
/* Note: HADOOP_PARROT_PATH is the path to Parrot on the remote node, not on the local machine. */
debug(D_NOTICE, "error: environment variable HADOOP_PARROT_PATH not set\n");
fail = 1;
}
if(fail) {
batch_queue_delete(q);
return 0;
}
}
return q;
}