- wrapper: fix CPU time accounting on Unix

samples/condor/curl.cpp

@@ -174,7 +174,9 @@ int upload_files (
 int create_batch(
     const char* project_url,
     const char* authenticator,
-    SUBMIT_REQ& sr
+    const char* batch_name,
+    const char* app_name,
+    int& batch_id
 ) {
     char request[1024];
     char url[1024];
@@ -187,8 +189,8 @@ int create_batch(
         "      </batch>\n"
         "</create_batch>\n",
         authenticator,
-        sr.batch_name,
-        sr.app_name
+        batch_name,
+        app_name
     );
     sprintf(url, "%ssubmit_rpc_handler.php", project_url);
     FILE* reply = tmpfile();
@@ -199,13 +201,14 @@ int create_batch(
         return retval;
     }
     char buf[256];
-    sr.batch_id = 0;
+    batch_id = 0;
     fseek(reply, 0, SEEK_SET);
     while (fgets(buf, 256, reply)) {
-        if (parse_int(buf, "<batch_id>", sr.batch_id)) break;
+        printf("reply: %s", buf);
+        if (parse_int(buf, "<batch_id>", batch_id)) break;
     }
     fclose(reply);
-    if (sr.batch_id == 0) {
+    if (batch_id == 0) {
         return -1;
     }
     return 0;

samples/wrapper/job.xml

@@ -5,11 +5,10 @@
         <stdout_filename>stdout</stdout_filename>
         <command_line>10</command_line>
     </task>
-    <unzip_input>
-        <zipfilename>in.zip</zipfilename>
-    </unzip_input>
-    <zip_output>
-        <zipfilename>foo.zip</zipfilename>
-        <filename>out</filename>
-    </zip_output>
+    <task>
+        <application>worker</application>
+        <stdin_filename>stdin</stdin_filename>
+        <stdout_filename>stdout</stdout_filename>
+        <command_line>10</command_line>
+    </task>
 </job_desc>

samples/wrapper/wrapper.cpp

@@ -66,11 +66,14 @@
 
 #include "regexp.h"
 
+//#define DEBUG
+#if 1
+#define debug_msg(x)
+#else
 inline void debug_msg(const char* x) {
-#if 0
     fprintf(stderr, "%s\n", x);
-#endif
 }
+#endif
 
 #define JOB_FILENAME "job.xml"
 #define CHECKPOINT_FILENAME "wrapper_checkpoint.txt"
@@ -116,10 +119,11 @@ struct TASK {
     HANDLE pid_handle;
     DWORD pid;
     HANDLE thread_handle;
-    struct _stat last_stat;    // mod time of checkpoint file
+    struct _stat last_stat;     // mod time of checkpoint file
 #else
     int pid;
     struct stat last_stat;
+    double start_rusage;        // getrusage() CPU time at start of task
 #endif
     bool stat_first;
 
@@ -699,6 +703,10 @@ int TASK::run(int argct, char** argvt) {
     FILE* stdin_file;
     FILE* stderr_file;
 
+    struct rusage ru;
+    getrusage(RUSAGE_CHILDREN, &ru);
+    start_rusage = (float)ru.ru_utime.tv_sec + ((float)ru.ru_utime.tv_usec)/1e+6;
+
     pid = fork();
     if (pid == -1) {
         perror("fork(): ");
@@ -790,6 +798,12 @@ bool TASK::poll(int& status) {
     if (wpid) {
         getrusage(RUSAGE_CHILDREN, &ru);
         final_cpu_time = (float)ru.ru_utime.tv_sec + ((float)ru.ru_utime.tv_usec)/1e+6;
+        final_cpu_time -= start_rusage;
+#ifdef DEBUG
+        printf("process exited; current CPU %f final CPU %f\n",
+            current_cpu_time, final_cpu_time
+        );
+#endif
         if (final_cpu_time < current_cpu_time) {
             final_cpu_time = current_cpu_time;
         }
@@ -1005,6 +1019,13 @@ int main(int argc, char** argv) {
             if (counter%10 == 0) {
                 cpu_time = task.cpu_time();
             }
+#ifdef DEBUG
+            printf("cpu time %f, checkpoint CPU time %f frac done %f\n",
+                task.starting_cpu + cpu_time,
+                checkpoint_cpu_time,
+                frac_done + delta
+            );
+#endif
             boinc_report_app_status(
                 task.starting_cpu + cpu_time,
                 checkpoint_cpu_time,
@@ -1019,6 +1040,18 @@ int main(int argc, char** argv) {
             counter++;
         }
         checkpoint_cpu_time = task.starting_cpu + task.final_cpu_time;
+#ifdef DEBUG
+        printf("cpu time %f, checkpoint CPU time %f frac done %f\n",
+            task.starting_cpu + task.final_cpu_time,
+            checkpoint_cpu_time,
+            frac_done + task.weight/total_weight
+        );
+#endif
+        boinc_report_app_status(
+            task.starting_cpu + task.final_cpu_time,
+            checkpoint_cpu_time,
+            frac_done + task.weight/total_weight
+        );
         write_checkpoint(i+1, checkpoint_cpu_time);
         weight_completed += task.weight;
     }