Merge pull request #6403 from BOINC/dpa_docker_cpu

client: avoid erroneous 'suspended - non-BOINC CPU' when running Docker apps

Author: AenBleidd

client/app.cpp

@@ -415,7 +415,7 @@ void ACTIVE_TASK_SET::get_memory_usage() {
             v = &(atp->other_pids);
         }
         procinfo_app(pi, v, pm, atp->app_version->graphics_exec_file);
-        if (atp->app_version->is_vm_app) {
+        if (atp->app_version->is_vbox_app) {
             vbox_app_running = true;
             // the memory of virtual machine apps is not reported correctly,
             // at least on Windows.  Use the VM size instead.
@@ -538,12 +538,29 @@ void ACTIVE_TASK_SET::get_memory_usage() {
 
 #if defined(__linux__) || defined(_WIN32) || defined(__APPLE__)
     // compute non_boinc_cpu_usage
-    // Improved version for systems where we can get total CPU (Win, Linux, Mac)
+    // Improved version for systems where we can get total CPU
+    // (Win, Linux, Mac)
     //
     static double last_nbrc=0;
     double total_cpu_time_now = total_cpu_time();
-    if (total_cpu_time_now != 0.0) {    // total_cpu_time() returns 0.0 on error
-        double nbrc = total_cpu_time_now - boinc_related_cpu_time(pm, vbox_app_running);
+
+    // total_cpu_time() returns 0.0 on error
+    //
+    if (total_cpu_time_now != 0.0) {
+        double brc = boinc_related_cpu_time(pm, vbox_app_running);
+#ifndef _WIN32
+        // on Win, boinc_related_cpu_time() includes CPU time of Docker jobs.
+        // On other platforms we need to do it by looking at the
+        // reported CPU times of the jobs
+        // (which may be less reliable/accurate)
+        //
+        for (ACTIVE_TASK* atp: active_tasks) {
+            if (atp->app_version->is_docker_app) {
+                brc += atp->current_cpu_time;
+            }
+        }
+#endif
+        double nbrc = total_cpu_time_now - brc;
         double delta_nbrc = nbrc - last_nbrc;
         if (delta_nbrc < 0) delta_nbrc = 0;
         last_nbrc = nbrc;

client/client_types.cpp

@@ -882,7 +882,8 @@ void APP_VERSION::init() {
     graphics_exec_path[0] = 0;
     graphics_exec_file[0] = 0;
     max_working_set_size = 0;
-    is_vm_app = false;
+    is_vbox_app = false;
+    is_docker_app = false;
     is_wrapper = false;
     index = 0;
 #ifdef SIM
@@ -902,7 +903,10 @@ int APP_VERSION::parse(XML_PARSER& xp) {
                 dont_throttle = true;
             }
             if (strstr(plan_class, "vbox")) {
-                is_vm_app = true;
+                is_vbox_app = true;
+            }
+            if (strstr(plan_class, "docker")) {
+                is_docker_app = true;
             }
             return 0;
         }
@@ -915,9 +919,6 @@ int APP_VERSION::parse(XML_PARSER& xp) {
                 );
                 return retval;
             }
-            if (strstr(file_ref.file_name, "vboxwrapper")) {
-                is_vm_app = true;
-            }
             app_files.push_back(file_ref);
             continue;
         }

client/client_types.h

@@ -372,8 +372,10 @@ struct APP_VERSION {
         // to use this much RAM,
         // so that we don't run a long sequence of jobs,
         // each of which turns out not to fit in available RAM
-    bool is_vm_app;
-        // currently this set if plan class includes "vbox" (kludge)
+    bool is_vbox_app;
+        // set if plan class includes "vbox"
+    bool is_docker_app;
+        // set if plan class includes "docker"
     bool is_wrapper;
         // the main program is a wrapper; run it above idle priority
 

lib/procinfo.cpp

@@ -123,9 +123,11 @@ void find_children(PROC_MAP& pm) {
     }
 }
 
-// get resource usage of non-BOINC apps
+// get resource usage of non-BOINC apps running above background priority.
 // NOTE: this is flawed because it doesn't account for short-lived processes.
-// It's not used on Win, Mac, or Linux, which have better ways of getting total CPU usage.
+// It's not used on Win, Mac, or Linux,
+// which have ways of getting total CPU usage.
+// See client/app.cpp
 //
 void procinfo_non_boinc(PROCINFO& procinfo, PROC_MAP& pm) {
     procinfo.clear();
@@ -162,8 +164,20 @@ void procinfo_non_boinc(PROCINFO& procinfo, PROC_MAP& pm) {
 #endif
 }
 
-// get CPU time of BOINC-related processes, low-priority processes,
-// and (if we're using Vbox) the Vbox daemon.
+// get CPU time of things we don't want to count as non-BOINC-related
+// - BOINC apps
+// - low-priority processes
+// - (if Vbox apps are running) the Vbox daemon
+// - Windows: WSL daemon ('vmmem')
+// - Linux/Mac:
+//      we don't account Docker/podman CPU time here,
+//      since we don't know what the processes are.
+//      Instead we do it in the client (by looking at ACTIVE_TASKS)
+//
+//      processes named 'podman'
+//
+// This is subtracted from total CPU time to get
+// the 'non-BOINC CPU time' used in computing preferences
 //
 double boinc_related_cpu_time(PROC_MAP& pm, bool vbox_app_running) {
     double sum = 0;
@@ -180,8 +194,11 @@ double boinc_related_cpu_time(PROC_MAP& pm, bool vbox_app_running) {
                 // if a VBox app is running,
                 // count VBox processes as BOINC-related
                 // e.g. VBoxHeadless.exe and VBoxSVC.exe on Win
+#ifdef _WIN32
+            || strstr(p.command, "vmmem")
+#endif
         ) {
-            sum += p.user_time;
+            sum += (p.user_time + p.kernel_time);
         }
     }
     return sum;

lib/procinfo_unix.cpp

@@ -109,6 +109,9 @@ struct PROC_STAT {
     int parse(char*);
 };
 
+// parse a /proc/<pid>/stat files (1 line)
+// see https://man7.org/linux/man-pages/man5/proc_pid_stat.5.html
+//
 int PROC_STAT::parse(char* buf) {
     int n = sscanf(buf,
         "%d (%[^)]) %c %d %d %d %d %d "