Merge pull request #5870 from BOINC/dpa_docker_rsc

client/API/docker_wrapper: get CPU and mem usage of Docker apps

api/boinc_api.cpp

@@ -1052,7 +1052,8 @@ int boinc_report_app_status_aux(
     double _fraction_done,
     int other_pid,
     double _bytes_sent,
-    double _bytes_received
+    double _bytes_received,
+    double wss
 ) {
     char msg_buf[MSG_CHANNEL_SIZE], buf[1024];
     if (standalone) return 0;
@@ -1081,6 +1082,10 @@ int boinc_report_app_status_aux(
         sprintf(buf, "<sporadic_ac>%d</sporadic_ac>\n", ac_state);
         strlcat(msg_buf, buf, sizeof(msg_buf));
     }
+    if (wss) {
+        sprintf(buf, "<wss>%f</wss>\n", wss);
+        strlcat(msg_buf, buf, sizeof(msg_buf));
+    }
 #ifdef MSGS_FROM_FILE
     if (fout) {
         fputs(msg_buf, fout);
@@ -1100,7 +1105,7 @@ int boinc_report_app_status(
     double _fraction_done
 ){
     return boinc_report_app_status_aux(
-        cpu_time, checkpoint_cpu_time, _fraction_done, 0, 0, 0
+        cpu_time, checkpoint_cpu_time, _fraction_done, 0, 0, 0, 0
     );
 }
 

api/boinc_api.h

@@ -138,7 +138,8 @@ extern int boinc_upload_status(std::string& name);
 extern char* boinc_msg_prefix(char*, int);
 extern int boinc_report_app_status_aux(
     double cpu_time, double checkpoint_cpu_time, double _fraction_done,
-    int other_pid, double bytes_sent, double bytes_received
+    int other_pid, double bytes_sent, double bytes_received,
+    double wss
 );
 extern int boinc_temporary_exit(
     int delay, const char* reason=NULL, bool is_notice=false

client/app.cpp

@@ -110,6 +110,7 @@ ACTIVE_TASK::ACTIVE_TASK() {
     peak_disk_usage = 0;
     once_ran_edf = false;
 
+    wss_from_app = 0;
     fraction_done = 0;
     fraction_done_elapsed_time = 0;
     first_fraction_done = 0;
@@ -420,6 +421,8 @@ void ACTIVE_TASK_SET::get_memory_usage() {
             // at least on Windows.  Use the VM size instead.
             //
             pi.working_set_size_smoothed = atp->wup->rsc_memory_bound;
+        } else if (atp->wss_from_app > 0) {
+            pi.working_set_size_smoothed = .5*(pi.working_set_size_smoothed + atp->wss_from_app);
         } else {
             pi.working_set_size_smoothed = .5*(pi.working_set_size_smoothed + pi.working_set_size);
         }

client/app.h

@@ -53,12 +53,14 @@ typedef int PROCESS_ID;
 
 // Represents a job in progress.
 
-// When an active task is created, it is assigned a "slot"
+// When a job is started, it is assigned a "slot"
 // which determines the directory it runs in.
-// This doesn't change over the life of the active task;
-// thus the task can use the slot directory for temp files
+// This doesn't change over the life of the job;
+// so it can use the slot directory for temp files
 // that BOINC doesn't know about.
 
+// If you add anything, initialize it in the constructor
+//
 struct ACTIVE_TASK {
 #ifdef _WIN32
     HANDLE process_handle, shm_handle;
@@ -100,8 +102,12 @@ struct ACTIVE_TASK {
         // most recent CPU time reported by app
     bool once_ran_edf;
 
-    // END OF ITEMS SAVED IN STATE FILE
+    // END OF ITEMS SAVED IN STATE FILES
 
+    double wss_from_app;
+        // work set size reported by the app
+        // (e.g. docker_wrapper does this).
+        // If nonzero, use this instead of procinfo data
     double fraction_done;
         // App's estimate of how much of the work unit is done.
         // Passed from the application via an API call;

client/app_control.cpp

@@ -1439,8 +1439,23 @@ bool ACTIVE_TASK::get_app_status_msg() {
             }
         }
     }
-    parse_double(msg_buf, "<current_cpu_time>", current_cpu_time);
-    parse_double(msg_buf, "<checkpoint_cpu_time>", checkpoint_cpu_time);
+    if (parse_double(msg_buf, "<current_cpu_time>", current_cpu_time)) {
+        if (current_cpu_time < 0) {
+            msg_printf(result->project, MSG_INFO,
+                "app reporting negative CPU: %f", current_cpu_time
+            );
+            current_cpu_time = 0;
+        }
+    }
+    if (parse_double(msg_buf, "<checkpoint_cpu_time>", checkpoint_cpu_time)) {
+        if (checkpoint_cpu_time < 0) {
+            msg_printf(result->project, MSG_INFO,
+                "app reporting negative checkpoint CPU: %f", checkpoint_cpu_time
+            );
+            checkpoint_cpu_time = 0;
+        }
+    }
+    parse_double(msg_buf, "<wss>", wss_from_app);
     parse_double(msg_buf, "<fpops_per_cpu_sec>", result->fpops_per_cpu_sec);
     parse_double(msg_buf, "<fpops_cumulative>", result->fpops_cumulative);
     parse_double(msg_buf, "<intops_per_cpu_sec>", result->intops_per_cpu_sec);
@@ -1470,18 +1485,6 @@ bool ACTIVE_TASK::get_app_status_msg() {
     if (parse_int(msg_buf, "<sporadic_ac>", i)) {
         sporadic_ac_state = (SPORADIC_AC_STATE)i;
     }
-    if (current_cpu_time < 0) {
-        msg_printf(result->project, MSG_INFO,
-            "app reporting negative CPU: %f", current_cpu_time
-        );
-        current_cpu_time = 0;
-    }
-    if (checkpoint_cpu_time < 0) {
-        msg_printf(result->project, MSG_INFO,
-            "app reporting negative checkpoint CPU: %f", checkpoint_cpu_time
-        );
-        checkpoint_cpu_time = 0;
-    }
     return true;
 }
 

client/app_test.cpp

@@ -26,7 +26,7 @@
 //      input/output files, attributes, etc.
 //      It currently has several test cases, selected with #ifdef
 // - build the BOINC client with these changes
-// - make a BOINC data directory, say 'test'
+// - Linux: make a BOINC data directory, say 'test'
 //      (or you can use an existing BOINC data directory,
 //      in which case the client will also run existing jobs)
 // - make a directory test/slots/app_test
@@ -224,6 +224,20 @@ void CLIENT_STATE::app_test_init() {
         *make_file(app->project, "Dockerfile_copy", "Dockerfile", INPUT_FILE, true)
     );
 #endif
+#ifdef APP_DOCKER_WRAPPER_MOUNT
+    av->app_files.push_back(
+        *make_file(app->project, "docker_wrapper.exe", NULL, MAIN_PROG, false)
+    );
+    av->app_files.push_back(
+        *make_file(app->project, "worker", NULL, INPUT_FILE, false)
+    );
+    av->app_files.push_back(
+        *make_file(app->project, "job_copy.toml", "job.toml", INPUT_FILE, true)
+    );
+    av->app_files.push_back(
+        *make_file(app->project, "Dockerfile_copy", "Dockerfile", INPUT_FILE, true)
+    );
+#endif
 
     // can put other stuff here like
 #if 0
@@ -243,11 +257,17 @@ void CLIENT_STATE::app_test_init() {
     );
 #endif
 #ifdef APP_DOCKER_WRAPPER_COPY
+    wu->command_line = "--verbose";
     wu->input_files.push_back(
         *make_file(proj, "infile", "in", INPUT_FILE, true)
     );
 #endif
-
+#ifdef APP_DOCKER_WRAPPER_MOUNT
+    wu->command_line = "--verbose";
+    wu->input_files.push_back(
+        *make_file(proj, "infile", "in", INPUT_FILE, false)
+    );
+#endif
     RESULT *result = make_result(av, wu);
 
 ////////////// OUTPUT FILES /////////////////
@@ -262,6 +282,11 @@ void CLIENT_STATE::app_test_init() {
         *make_file(proj, "outfile", "out", OUTPUT_FILE, true)
     );
 #endif
+#ifdef APP_DOCKER_WRAPPER_MOUNT
+    result->output_files.push_back(
+        *make_file(proj, "outfile", "out", OUTPUT_FILE, false)
+    );
+#endif
 
     // tell the client not to get work or run benchmarks
     //