added option to run task based on dependency exit code

justanhduc · Feb 4, 2021 · 1916f1c · 1916f1c
1 parent 1488f71
commit 1916f1c
Show file tree

Hide file tree

Showing 4 changed files with 49 additions and 26 deletions.
diff --git a/client.c b/client.c
@@ -8,9 +8,7 @@
 #include <string.h>
 #include <unistd.h>
 #include <stdlib.h>
-#include <sys/types.h>
 #include <sys/socket.h>
-#include <sys/time.h>
 #include <signal.h>
 #include <time.h>
 
@@ -136,33 +134,43 @@ int c_wait_server_commands() {
         if (m.type == RUNJOB) {
             struct Result result;
             result.skipped = 0;
-            if (command_line.gpus) {
-                int numFree;
-                int *freeList = getFreeGpuList(&numFree);
-                if (command_line.gpus > numFree) {
-                    result.errorlevel = -1;
-                    result.user_ms = 0.;
-                    result.system_ms = 0.;
-                    result.real_ms = 0.;
-                    result.skipped = 1;
-                    c_send_runjob_ok(0, -1);
-                } else {
-                    char tmp[50];
-                    strcpy(tmp, "CUDA_VISIBLE_DEVICES=");
-                    shuffle(freeList, numFree);
-                    for (int i = 0; i < command_line.gpus; i++) {
-                        char tmp2[5];
-                        sprintf(tmp2, "%d", freeList[i]);
-                        strcat(tmp, tmp2);
-                        if (i < command_line.gpus - 1)
-                            strcat(tmp, ",");
+            if (command_line.do_depend && command_line.require_elevel && m.u.last_errorlevel != 0) {
+                result.errorlevel = -1;
+                result.user_ms = 0.;
+                result.system_ms = 0.;
+                result.real_ms = 0.;
+                result.skipped = 1;
+                c_send_runjob_ok(0, -1);
+            } else {
+                if (command_line.gpus) {
+                    int numFree;
+                    int *freeList = getFreeGpuList(&numFree);
+                    if ((command_line.gpus > numFree)) {
+                        result.errorlevel = -1;
+                        result.user_ms = 0.;
+                        result.system_ms = 0.;
+                        result.real_ms = 0.;
+                        result.skipped = 1;
+                        c_send_runjob_ok(0, -1);
+                    } else {
+                        char tmp[50];
+                        strcpy(tmp, "CUDA_VISIBLE_DEVICES=");
+                        shuffle(freeList, numFree);
+                        for (int i = 0; i < command_line.gpus; i++) {
+                            char tmp2[5];
+                            sprintf(tmp2, "%d", freeList[i]);
+                            strcat(tmp, tmp2);
+                            if (i < command_line.gpus - 1)
+                                strcat(tmp, ",");
+                        }
+                        putenv(tmp);
                     }
-                    putenv(tmp);
+                    free(freeList);
                 }
-                free(freeList);
+
+                run_job(&result);
             }
 
-            run_job(&result);
             c_end_of_job(&result);
             return result.errorlevel;
         } else if (m.type == REMINDER) {

diff --git a/jobs.c b/jobs.c
@@ -854,6 +854,13 @@ void s_send_runjob(int s, int jobid) {
         error("Job %i was expected to run", jobid);
 
     m.type = RUNJOB;
+
+    /* TODO
+     * We should make the dependencies update the jobids they're do_depending on.
+     * Then, on finish, these could set the errorlevel to send to its dependency childs.
+     * We cannot consider that the jobs will leave traces in the finished job list (-nf?) . */
+
+    m.u.last_errorlevel = p->dependency_errorlevel;
     send_msg(s, &m);
 }
 

diff --git a/main.c b/main.c
@@ -44,6 +44,7 @@ static void default_command_line() {
     command_line.wait_enqueuing = 1;
     command_line.stderr_apart = 0;
     command_line.num_slots = 1;
+    command_line.require_elevel = 0;
     command_line.gpus = 0;
 }
 
@@ -93,7 +94,7 @@ void parse_opts(int argc, char **argv) {
 
     /* Parse options */
     while (1) {
-        c = getopt_long(argc, argv, ":RTVhKgClnfmBEr:a:F:t:c:o:p:w:k:u:s:U:qi:N:L:dS:D:G:",
+        c = getopt_long(argc, argv, ":RTVhKgClnfmBEr:a:F:t:c:o:p:w:k:u:s:U:qi:N:L:dS:D:G:W:",
                         longOptions, &optionIdx);
 
         if (c == -1)
@@ -225,6 +226,11 @@ void parse_opts(int argc, char **argv) {
                 command_line.do_depend = 1;
                 command_line.depend_on = atoi(optarg);
                 break;
+            case 'W':
+                command_line.do_depend = 1;
+                command_line.depend_on = atoi(optarg);
+                command_line.require_elevel = 1;
+                break;
             case 'U':
                 command_line.request = c_SWAP_JOBS;
                 res = get_two_jobs(optarg, &command_line.jobid,
@@ -426,6 +432,7 @@ static void print_help(const char *cmd) {
     printf("  -m       send the output by e-mail (uses sendmail).\n");
     printf("  -d       the job will be run after the last job ends.\n");
     printf("  -D <id>  the job will be run after the job of given id ends.\n");
+    printf("  -W <id>  the job will be run after the job of given id ends well (exit code 0).\n");
     printf("  -L <lab> name this task with a label, to be distinguished on listing.\n");
     printf("  -N <num> number of slots required by the job (1 default).\n");
 }

diff --git a/main.h b/main.h
@@ -104,6 +104,7 @@ struct CommandLine {
     } command;
     char *label;
     int num_slots; /* Slots for the job to use. Default 1 */
+    int require_elevel;  /* whether requires error level of dependencies or not */
     int gpus;
     int gpu_wait_time;
 };