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;
}
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[] )
{
signed char c;
const char *progname = "wavefront";
debug_config(progname);
progress_log_file = stdout;
struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"version", no_argument, 0, 'v'},
{"debug", required_argument, 0, 'd'},
{"jobs", required_argument, 0, 'n'},
{"block-size", required_argument, 0, 'b'},
{"debug-file", required_argument, 0, 'o'},
{"log-file", required_argument, 0, 'l'},
{"bitmap", required_argument, 0, 'B'},
{"bitmap-interval", required_argument, 0, 'i'},
{"auto", no_argument, 0, 'A'},
{"local", no_argument, 0, 'L'},
{"batch-type", required_argument, 0, 'T'},
{"verify", no_argument, 0, 'V'},
{0,0,0,0}
};
while((c=getopt_long(argc,argv,"n:b:d:o:l:B:i:qALDT:VX:Y:vh", long_options, NULL)) > -1) {
switch(c) {
case 'n':
manual_max_jobs_running = atoi(optarg);
break;
case 'b':
manual_block_size = atoi(optarg);
break;
case 'd':
debug_flags_set(optarg);
break;
case 'o':
debug_config_file(optarg);
break;
case 'B':
progress_bitmap_file = optarg;
break;
case 'i':
progress_bitmap_interval = atoi(optarg);
break;
case 'l':
progress_log_file = fopen(optarg,"w");
if(!progress_log_file) {
fprintf(stderr,"couldn't open %s: %s\n",optarg,strerror(errno));
return 1;
}
break;
case 'A':
wavefront_mode = WAVEFRONT_MODE_AUTO;
break;
case 'L':
wavefront_mode = WAVEFRONT_MODE_MULTICORE;
break;
case 'T':
wavefront_mode = WAVEFRONT_MODE_DISTRIBUTED;
batch_system_type = batch_queue_type_from_string(optarg);
if(batch_system_type==BATCH_QUEUE_TYPE_UNKNOWN) {
fprintf(stderr,"unknown batch system type: %s\n",optarg);
exit(1);
}
break;
case 'V':
verify_mode = 1;
break;
case 'X':
xstart = atoi(optarg);
break;
case 'Y':
ystart = atoi(optarg);
break;
case 'v':
cctools_version_print(stdout, progname);
exit(0);
break;
case 'h':
show_help(progname);
exit(0);
break;
}
}
cctools_version_debug(D_DEBUG, argv[0]);
if( (argc-optind<3) ) {
show_help(progname);
exit(1);
}
function = argv[optind];
xsize=atoi(argv[optind+1]);
ysize=atoi(argv[optind+2]);
total_cells = xsize*ysize;
if(!verify_mode && !check_configuration(function,xsize,ysize)) exit(1);
int ncpus = load_average_get_cpus();
if(wavefront_mode!=WAVEFRONT_MODE_MULTICORE) {
double task_time = measure_task_time();
printf("Each function takes %.02lfs to run.\n",task_time);
block_size = find_best_block_size(xsize,1000,2,task_time,average_dispatch_time);
double distributed_time = wavefront_distributed_model(xsize,1000,2,task_time,block_size,average_dispatch_time);
double multicore_time = wavefront_multicore_model(xsize,ncpus,task_time);
double ideal_multicore_time = wavefront_multicore_model(xsize,xsize,task_time);
double sequential_time = wavefront_multicore_model(xsize,1,task_time);
printf("---------------------------------\n");
printf("This workload would take:\n");
printf("%.02lfs sequentially\n",sequential_time);
printf("%.02lfs on this %d-core machine\n",multicore_time,ncpus);
printf("%.02lfs on a %d-core machine\n",ideal_multicore_time,xsize);
printf("%.02lfs on a 1000-node distributed system with block size %d\n",distributed_time,block_size);
printf("---------------------------------\n");
if(wavefront_mode==WAVEFRONT_MODE_AUTO) {
if(multicore_time < distributed_time*2) {
wavefront_mode = WAVEFRONT_MODE_MULTICORE;
} else {
wavefront_mode = WAVEFRONT_MODE_DISTRIBUTED;
}
}
}
if(wavefront_mode==WAVEFRONT_MODE_MULTICORE) {
batch_system_type = BATCH_QUEUE_TYPE_LOCAL;
max_jobs_running = ncpus;
} else {
max_jobs_running = 1000;
}
if(manual_block_size!=0) {
block_size = manual_block_size;
}
if(manual_max_jobs_running!=0) {
max_jobs_running = manual_max_jobs_running;
}
if(wavefront_mode==WAVEFRONT_MODE_MULTICORE) {
printf("Running in multicore mode with %d CPUs.\n",max_jobs_running);
} else {
printf("Running in distributed mode with block size %d on up to %d CPUs\n",block_size,max_jobs_running);
}
batch_q = batch_queue_create(batch_system_type);
if(verify_mode) exit(0);
struct bitmap * b = bitmap_create(xsize+1,ysize+1);
struct list *ready_list = list_create();
struct itable *running_table = itable_create(0);
struct batch_job_info info;
UINT64_T jobid;
struct wavefront_task *task;
wavefront_task_initialize(b,ready_list);
printf("Starting workload...\n");
fprintf(progress_log_file,"# elapsed time : waiting jobs / running jobs / cells complete (percent complete)\n");
while(1) {
if(abort_mode) {
while((task=list_pop_tail(ready_list))) {
wavefront_task_delete(task);
}
itable_firstkey(running_table);
while(itable_nextkey(running_table,&jobid,(void**)&task)) {
batch_job_remove(batch_q,jobid);
}
}
if(list_size(ready_list)==0 && itable_size(running_table)==0) break;
while(1) {
if(itable_size(running_table)>=max_jobs_running) break;
task = list_pop_tail(ready_list);
if(!task) break;
jobid = wavefront_task_submit(task);
if(jobid>0) {
itable_insert(running_table,jobid,task);
wavefront_task_mark_range(task,b,WAVEFRONT_TASK_STATE_RUNNING);
} else {
abort();
sleep(1);
list_push_head(ready_list,task);
}
}
save_status(b,ready_list,running_table);
jobid = batch_job_wait(batch_q,&info);
if(jobid>0) {
task = itable_remove(running_table,jobid);
if(task) {
if(info.exited_normally && info.exit_code==0) {
total_dispatch_time += info.started-info.submitted;
total_execute_time += MAX(info.finished-info.started,1);
total_cells_complete+=task->width*task->height;
total_jobs_complete++;
average_dispatch_time = 1.0*total_dispatch_time / total_jobs_complete;
average_task_time = 1.0*total_execute_time / total_cells_complete;
wavefront_task_complete(b,ready_list,task);
} else {
printf("job %" PRIu64 " failed, aborting this workload\n",jobid);
abort_mode = 1;
}
}
}
}
save_status(b,ready_list,running_table);
if(abort_mode) {
printf("Workload was aborted.\n");
} else {
printf("Workload complete.\n");
}
return 0;
}
