00001
00002
00003
00004
00005
00006
00007 #ifndef WORK_QUEUE_H
00008 #define WORK_QUEUE_H
00009
00020 #include <sys/types.h>
00021 #include "timestamp.h"
00022 #include "rmsummary.h"
00023
00024 #define WORK_QUEUE_DEFAULT_PORT 9123
00025 #define WORK_QUEUE_RANDOM_PORT 0
00026 #define WORK_QUEUE_WAITFORTASK -1
00028 #define WORK_QUEUE_SCHEDULE_UNSET 0
00029 #define WORK_QUEUE_SCHEDULE_FCFS 1
00030 #define WORK_QUEUE_SCHEDULE_FILES 2
00031 #define WORK_QUEUE_SCHEDULE_TIME 3
00032 #define WORK_QUEUE_SCHEDULE_RAND 4
00034 #define WORK_QUEUE_INPUT 0
00035 #define WORK_QUEUE_OUTPUT 1
00037 #define WORK_QUEUE_NOCACHE 0
00038 #define WORK_QUEUE_CACHE 1
00039 #define WORK_QUEUE_SYMLINK 2
00040 #define WORK_QUEUE_PREEXIST 4
00041 #define WORK_QUEUE_THIRDGET 8
00042 #define WORK_QUEUE_THIRDPUT 8
00043 #define WORK_QUEUE_WATCH 16
00045 #define WORK_QUEUE_RESET_ALL 0
00046 #define WORK_QUEUE_RESET_KEEP_TASKS 1
00048 #define WORK_QUEUE_DEFAULT_KEEPALIVE_INTERVAL 300
00049 #define WORK_QUEUE_DEFAULT_KEEPALIVE_TIMEOUT 30
00051 #define WORK_QUEUE_RESULT_SUCCESS 0
00052 #define WORK_QUEUE_RESULT_INPUT_MISSING 1
00053 #define WORK_QUEUE_RESULT_OUTPUT_MISSING 2
00054 #define WORK_QUEUE_RESULT_STDOUT_MISSING 4
00055 #define WORK_QUEUE_RESULT_SIGNAL 8
00056 #define WORK_QUEUE_RESULT_RESOURCE_EXHAUSTION 16
00057 #define WORK_QUEUE_RESULT_TASK_TIMEOUT 32
00059 extern double wq_option_fast_abort_multiplier;
00061 extern int wq_option_scheduler;
00065 struct work_queue_task {
00066 char *tag;
00067 char *command_line;
00068 int worker_selection_algorithm;
00069 char *output;
00070 struct list *input_files;
00071 struct list *output_files;
00072 int taskid;
00073 int return_status;
00074 int result;
00075 char *host;
00076 char *hostname;
00078 timestamp_t time_committed;
00080 timestamp_t time_task_submit;
00081 timestamp_t time_task_finish;
00082 timestamp_t time_send_input_start;
00083 timestamp_t time_send_input_finish;
00084 timestamp_t time_execute_cmd_start;
00085 timestamp_t time_execute_cmd_finish;
00086 timestamp_t time_receive_result_start;
00087 timestamp_t time_receive_result_finish;
00088 timestamp_t time_receive_output_start;
00089 timestamp_t time_receive_output_finish;
00091 int64_t total_bytes_received;
00092 int64_t total_bytes_sent;
00093 int64_t total_bytes_transferred;
00094 timestamp_t total_transfer_time;
00095 timestamp_t cmd_execution_time;
00096 int total_submissions;
00097 timestamp_t total_cmd_execution_time;
00099 int64_t maximum_end_time;
00100 int64_t memory;
00101 int64_t disk;
00102 int cores;
00103 int gpus;
00104 int unlabeled;
00105 struct rmsummary *resources_measured;
00107 double priority;
00109 timestamp_t time_app_delay;
00111 };
00112
00115 struct work_queue_stats {
00116 int total_workers_connected;
00117 int workers_init;
00118 int workers_idle;
00119 int workers_busy;
00120 int total_workers_joined;
00121 int total_workers_removed;
00123 int tasks_waiting;
00124 int tasks_running;
00125 int tasks_complete;
00126 int total_tasks_dispatched;
00127 int total_tasks_complete;
00128 int total_tasks_failed;
00129 int total_tasks_cancelled;
00131 timestamp_t start_time;
00132 timestamp_t total_send_time;
00133 timestamp_t total_receive_time;
00134 timestamp_t total_good_transfer_time;
00136 timestamp_t total_execute_time;
00137 timestamp_t total_good_execute_time;
00140 int64_t total_bytes_sent;
00141 int64_t total_bytes_received;
00142 double efficiency;
00143 double idle_percentage;
00144 int capacity;
00146 double bandwidth;
00147 int64_t total_cores;
00148 int64_t total_memory;
00149 int64_t total_disk;
00150 int64_t total_gpus;
00151 int64_t committed_cores;
00152 int64_t committed_memory;
00153 int64_t committed_disk;
00154 int64_t committed_gpus;
00155 int64_t min_cores;
00156 int64_t max_cores;
00157 int64_t min_memory;
00158 int64_t max_memory;
00159 int64_t min_disk;
00160 int64_t max_disk;
00161 int64_t min_gpus;
00162 int64_t max_gpus;
00163 int port;
00164 int priority;
00165 int workers_ready;
00166 int workers_full;
00167 int total_worker_slots;
00168 int avg_capacity;
00169 };
00170
00171
00175
00183 struct work_queue_task *work_queue_task_create(const char *full_command);
00184
00190 struct work_queue_task *work_queue_task_clone(const struct work_queue_task *task);
00191
00196 void work_queue_task_specify_command( struct work_queue_task *t, const char *cmd );
00197
00214 int work_queue_task_specify_file(struct work_queue_task *t, const char *local_name, const char *remote_name, int type, int flags);
00215
00230 int work_queue_task_specify_file_piece(struct work_queue_task *t, const char *local_name, const char *remote_name, off_t start_byte, off_t end_byte, int type, int flags);
00231
00242 int work_queue_task_specify_buffer(struct work_queue_task *t, const char *data, int length, const char *remote_name, int flags);
00243
00257 int work_queue_task_specify_directory(struct work_queue_task *t, const char *local_name, const char *remote_name, int type, int flags, int recursive);
00258
00264 void work_queue_task_specify_memory( struct work_queue_task *t, int64_t memory );
00265
00271 void work_queue_task_specify_disk( struct work_queue_task *t, int64_t disk );
00272
00278 void work_queue_task_specify_cores( struct work_queue_task *t, int cores );
00279
00285 void work_queue_task_specify_gpus( struct work_queue_task *t, int gpus );
00286
00292 void work_queue_task_specify_end_time( struct work_queue_task *t, int64_t seconds );
00293
00300 void work_queue_task_specify_tag(struct work_queue_task *t, const char *tag);
00301
00308 void work_queue_task_specify_priority(struct work_queue_task *t, double priority );
00309
00319 void work_queue_task_specify_algorithm(struct work_queue_task *t, int algo );
00320
00325 void work_queue_task_delete(struct work_queue_task *t);
00326
00328
00332
00349 struct work_queue *work_queue_create(int port);
00350
00358 int work_queue_enable_monitoring(struct work_queue *q, char *monitor_summary_file);
00359
00368 int work_queue_submit(struct work_queue *q, struct work_queue_task *t);
00369
00374 void work_queue_blacklist_add(struct work_queue *q, const char *hostname);
00375
00376
00381 void work_queue_blacklist_remove(struct work_queue *q, const char *hostname);
00382
00383
00387 void work_queue_blacklist_clear(struct work_queue *q);
00388
00403 struct work_queue_task *work_queue_wait(struct work_queue *q, int timeout);
00404
00416 int work_queue_hungry(struct work_queue *q);
00417
00425 int work_queue_empty(struct work_queue *q);
00426
00433 int work_queue_port(struct work_queue *q);
00434
00439 void work_queue_get_stats(struct work_queue *q, struct work_queue_stats *s);
00440
00445 void work_queue_get_stats_hierarchy(struct work_queue *q, struct work_queue_stats *s);
00446
00451 void work_queue_set_bandwidth_limit(struct work_queue *q, const char *bandwidth);
00452
00457 double work_queue_get_effective_bandwidth(struct work_queue *q);
00458
00465 char * work_queue_get_worker_summary( struct work_queue *q );
00466
00472 int work_queue_activate_fast_abort(struct work_queue *q, double multiplier);
00473
00474
00478 int work_queue_send_receive_ratio(struct work_queue *q, double ratio);
00479
00489 void work_queue_specify_algorithm(struct work_queue *q, int algo);
00490
00495 const char *work_queue_name(struct work_queue *q);
00496
00501 void work_queue_specify_name(struct work_queue *q, const char *name);
00502
00507 void work_queue_specify_priority(struct work_queue *q, int priority);
00508
00514 void work_queue_specify_catalog_server(struct work_queue *q, const char *hostname, int port);
00515
00521 struct work_queue_task *work_queue_cancel_by_taskid(struct work_queue *q, int id);
00522
00528 struct work_queue_task *work_queue_cancel_by_tasktag(struct work_queue *q, const char *tag);
00529
00534 struct list * work_queue_cancel_all_tasks(struct work_queue *q);
00535
00540 int work_queue_shut_down_workers(struct work_queue *q, int n);
00541
00546 void work_queue_delete(struct work_queue *q);
00547
00553 int work_queue_specify_log(struct work_queue *q, const char *logfile);
00554
00560 void work_queue_specify_password( struct work_queue *q, const char *password );
00561
00568 int work_queue_specify_password_file( struct work_queue *q, const char *file );
00569
00574 void work_queue_specify_keepalive_interval(struct work_queue *q, int interval);
00575
00580 void work_queue_specify_keepalive_timeout(struct work_queue *q, int timeout);
00581
00582
00598 int work_queue_tune(struct work_queue *q, const char *name, double value);
00599
00601
00605
00606 #define WORK_QUEUE_TASK_ORDER_FIFO 0
00607 #define WORK_QUEUE_TASK_ORDER_LIFO 1
00615 void work_queue_specify_task_order(struct work_queue *q, int order);
00616
00617
00618 #define WORK_QUEUE_MASTER_MODE_STANDALONE 0
00619 #define WORK_QUEUE_MASTER_MODE_CATALOG 1
00628 void work_queue_specify_master_mode(struct work_queue *q, int mode);
00629
00635 void work_queue_specify_estimate_capacity_on(struct work_queue *q, int estimate_capacity_on);
00636
00645 int work_queue_task_specify_input_buf(struct work_queue_task *t, const char *buf, int length, const char *rname);
00646
00654 int work_queue_task_specify_input_file(struct work_queue_task *t, const char *fname, const char *rname);
00655
00663 int work_queue_task_specify_input_file_do_not_cache(struct work_queue_task *t, const char *fname, const char *rname);
00664
00672 int work_queue_task_specify_output_file(struct work_queue_task *t, const char *rname, const char *fname);
00673
00681 int work_queue_task_specify_output_file_do_not_cache(struct work_queue_task *t, const char *rname, const char *fname);
00682
00684
00685
00686
00687
00688
00689
00690
00691
00692 void work_queue_activate_worker_waiting(struct work_queue *q, int resources);
00693
00694 #endif