// Send kill signal to all app processes
// Don't wait for them to exit
//
void ACTIVE_TASK_SET::kill_tasks(PROJECT* proj) {
unsigned int i;
ACTIVE_TASK *atp;
for (i=0; i<active_tasks.size(); i++) {
atp = active_tasks[i];
if (proj && atp->wup->project != proj) continue;
if (!atp->process_exists()) continue;
atp->kill_task(false);
}
}
// a file upload has finished.
// If any running apps are waiting for it, notify them
//
void ACTIVE_TASK_SET::upload_notify_app(FILE_INFO* fip) {
for (unsigned int i=0; i<active_tasks.size(); i++) {
ACTIVE_TASK* atp = active_tasks[i];
RESULT* rp = atp->result;
FILE_REF* frp = rp->lookup_file(fip);
if (frp) {
atp->upload_notify_app(fip, frp);
}
}
}
void ACTIVE_TASK_SET::init() {
for (unsigned int i=0; i<active_tasks.size(); i++) {
ACTIVE_TASK* atp = active_tasks[i];
atp->init(atp->result);
atp->scheduler_state = CPU_SCHED_PREEMPTED;
atp->read_task_state_file();
atp->current_cpu_time = atp->checkpoint_cpu_time;
atp->elapsed_time = atp->checkpoint_elapsed_time;
}
}
void ACTIVE_TASK_SET::network_available() {
#ifndef SIM
for (unsigned int i=0; i<active_tasks.size(); i++) {
ACTIVE_TASK* atp = active_tasks[i];
if (atp->want_network) {
atp->send_network_available();
}
}
#endif
}
// Send quit message to all app processes
// This is called when the core client exits,
// or when a project is detached or reset
//
void ACTIVE_TASK_SET::request_tasks_exit(PROJECT* proj) {
LOGD("app_control: ACTIVE_TASK::request_tasks_exit");
unsigned int i;
ACTIVE_TASK *atp;
for (i=0; i<active_tasks.size(); i++) {
atp = active_tasks[i];
if (proj && atp->wup->project != proj) continue;
if (!atp->process_exists()) continue;
atp->request_exit();
}
}
// tell all running apps of a project to reread prefs
//
void ACTIVE_TASK_SET::request_reread_prefs(PROJECT* project) {
unsigned int i;
ACTIVE_TASK* atp;
for (i=0; i<active_tasks.size(); i++) {
atp = active_tasks[i];
if (atp->result->project != project) continue;
if (!atp->process_exists()) continue;
atp->request_reread_prefs();
}
}
// Check if any of the active tasks have exceeded their
// resource limits on disk, CPU time or memory
//
bool ACTIVE_TASK_SET::check_rsc_limits_exceeded() {
unsigned int i;
ACTIVE_TASK *atp;
static double last_disk_check_time = 0;
bool do_disk_check = false;
bool did_anything = false;
double ram_left = gstate.available_ram();
double max_ram = gstate.max_available_ram();
// Some slot dirs have lots of files,
// so only check every min(disk_interval, 300) secs
//
double min_interval = gstate.global_prefs.disk_interval;
if (min_interval < 300) min_interval = 300;
if (gstate.now > last_disk_check_time + min_interval) {
do_disk_check = true;
}
for (i=0; i<active_tasks.size(); i++) {
atp = active_tasks[i];
if (atp->task_state() != PROCESS_EXECUTING) continue;
if (!atp->result->project->non_cpu_intensive && (atp->elapsed_time > atp->max_elapsed_time)) {
msg_printf(atp->result->project, MSG_INFO,
"Aborting task %s: exceeded elapsed time limit %f\n",
atp->result->name, atp->max_elapsed_time
);
atp->abort_task(ERR_RSC_LIMIT_EXCEEDED, "Maximum elapsed time exceeded");
did_anything = true;
continue;
}
if (atp->procinfo.working_set_size_smoothed > max_ram) {
msg_printf(atp->result->project, MSG_INFO,
"Aborting task %s: exceeded memory limit %.2fMB > %.2fMB\n",
atp->result->name,
atp->procinfo.working_set_size_smoothed/MEGA, max_ram/MEGA
);
atp->abort_task(ERR_RSC_LIMIT_EXCEEDED, "Maximum memory exceeded");
did_anything = true;
continue;
}
if (do_disk_check && atp->check_max_disk_exceeded()) {
did_anything = true;
continue;
}
ram_left -= atp->procinfo.working_set_size_smoothed;
}
if (ram_left < 0) {
gstate.request_schedule_cpus("RAM usage limit exceeded");
}
if (do_disk_check) {
last_disk_check_time = gstate.now;
}
return did_anything;
}
// Check to see if any tasks are running
// called if benchmarking and waiting for suspends to happen
// or the system needs to suspend itself so we are suspending
// the applications
//
bool ACTIVE_TASK_SET::is_task_executing() {
unsigned int i;
ACTIVE_TASK* atp;
for (i=0; i<active_tasks.size(); i++) {
atp = active_tasks[i];
if (atp->task_state() == PROCESS_EXECUTING) {
return true;
}
}
return false;
}
void ACTIVE_TASK_SET::graphics_poll() {
unsigned int i;
ACTIVE_TASK* atp;
for (i=0; i<active_tasks.size(); i++) {
atp = active_tasks[i];
if (!atp->process_exists()) continue;
atp->graphics_request_queue.msg_queue_poll(
atp->app_client_shm.shm->graphics_request
);
atp->check_graphics_mode_ack();
}
}
// compute a per-app-version "temporary DCF" based on the elapsed time
// and fraction done of running jobs
//
void compute_temp_dcf() {
unsigned int i;
for (i=0; i<gstate.app_versions.size(); i++) {
gstate.app_versions[i]->temp_dcf = 1;
}
for (i=0; i<gstate.active_tasks.active_tasks.size(); i++) {
ACTIVE_TASK* atp = gstate.active_tasks.active_tasks[i];
double x = atp->est_dur(false) / atp->result->estimated_duration(false);
APP_VERSION* avp = atp->result->avp;
if (x < avp->temp_dcf) {
avp->temp_dcf = x;
}
}
}
// resume all currently scheduled tasks
//
void ACTIVE_TASK_SET::unsuspend_all() {
unsigned int i;
ACTIVE_TASK* atp;
for (i=0; i<active_tasks.size(); i++) {
atp = active_tasks[i];
if (atp->scheduler_state != CPU_SCHED_SCHEDULED) continue;
if (atp->task_state() == PROCESS_UNINITIALIZED) {
if (atp->start(false)) {
msg_printf(atp->wup->project, MSG_INTERNAL_ERROR,
"Couldn't restart task %s", atp->result->name
);
}
} else if (atp->task_state() == PROCESS_SUSPENDED) {
atp->unsuspend();
}
}
}
static void handle_get_screensaver_tasks(MIOFILE& fout) {
unsigned int i;
ACTIVE_TASK* atp;
fout.printf(
"<handle_get_screensaver_tasks>\n"
" <suspend_reason>%d</suspend_reason>\n",
gstate.suspend_reason
);
for (i=0; i<gstate.active_tasks.active_tasks.size(); i++) {
atp = gstate.active_tasks.active_tasks[i];
if ((atp->task_state() == PROCESS_EXECUTING) ||
((atp->task_state() == PROCESS_SUSPENDED) && (gstate.suspend_reason == SUSPEND_REASON_CPU_THROTTLE))) {
atp->result->write_gui(fout);
}
}
fout.printf("</handle_get_screensaver_tasks>\n");
}
static void handle_result_op(GUI_RPC_CONN& grc, const char* op) {
RESULT* rp;
char result_name[256];
ACTIVE_TASK* atp;
string project_url;
strcpy(result_name, "");
while (!grc.xp.get_tag()) {
if (grc.xp.parse_str("name", result_name, sizeof(result_name))) continue;
if (grc.xp.parse_string("project_url", project_url)) continue;
}
PROJECT* p = get_project(grc, project_url);
if (!p) return;
if (!strlen(result_name)) {
grc.mfout.printf("<error>Missing result name</error>\n");
return;
}
rp = gstate.lookup_result(p, result_name);
if (!rp) {
grc.mfout.printf("<error>no such result</error>\n");
return;
}
if (!strcmp(op, "abort")) {
msg_printf(p, MSG_INFO, "task %s aborted by user", result_name);
atp = gstate.lookup_active_task_by_result(rp);
if (atp) {
atp->abort_task(EXIT_ABORTED_VIA_GUI, "aborted by user");
} else {
rp->abort_inactive(EXIT_ABORTED_VIA_GUI);
}
gstate.request_work_fetch("result aborted by user");
} else if (!strcmp(op, "suspend")) {
msg_printf(p, MSG_INFO, "task %s suspended by user", result_name);
rp->suspended_via_gui = true;
gstate.request_work_fetch("result suspended by user");
} else if (!strcmp(op, "resume")) {
msg_printf(p, MSG_INFO, "task %s resumed by user", result_name);
rp->suspended_via_gui = false;
}
gstate.request_schedule_cpus("result suspended, resumed or aborted by user");
gstate.set_client_state_dirty("Result RPC");
grc.mfout.printf("<success/>\n");
}
// check for msgs from active tasks,
// and update their elapsed time and other info
//
void ACTIVE_TASK_SET::get_msgs() {
//LOGD("app_control: ACTIVE_TASK::get_msgs");
unsigned int i;
ACTIVE_TASK *atp;
double old_time;
static double last_time=0;
double delta_t;
if (last_time) {
delta_t = gstate.now - last_time;
// Normally this is called every second.
// If delta_t is > 10, we'll assume that a period of hibernation
// or suspension happened, and treat it as zero.
// If negative, must be clock reset. Ignore.
//
if (delta_t > 10 || delta_t < 0) {
delta_t = 0;
}
} else {
delta_t = 0;
}
last_time = gstate.now;
for (i=0; i<active_tasks.size(); i++) {
atp = active_tasks[i];
if (!atp->process_exists()) continue;
old_time = atp->checkpoint_cpu_time;
if (atp->task_state() == PROCESS_EXECUTING) {
atp->elapsed_time += delta_t;
}
if (atp->get_app_status_msg()) {
if (old_time != atp->checkpoint_cpu_time) {
char buf[256];
sprintf(buf, "%s checkpointed", atp->result->name);
if (atp->overdue_checkpoint) {
gstate.request_schedule_cpus(buf);
}
atp->checkpoint_wall_time = gstate.now;
atp->premature_exit_count = 0;
atp->checkpoint_elapsed_time = atp->elapsed_time;
atp->checkpoint_fraction_done = atp->fraction_done;
atp->checkpoint_fraction_done_elapsed_time = atp->fraction_done_elapsed_time;
if (log_flags.checkpoint_debug) {
msg_printf(atp->wup->project, MSG_INFO,
"[checkpoint] result %s checkpointed",
atp->result->name
);
} else if (log_flags.task_debug) {
msg_printf(atp->wup->project, MSG_INFO,
"[task] result %s checkpointed",
atp->result->name
);
}
atp->write_task_state_file();
}
}
atp->get_trickle_up_msg();
atp->get_graphics_msg();
}
}
// clean up after finished apps
//
bool CLIENT_STATE::handle_finished_apps() {
ACTIVE_TASK* atp;
bool action = false;
static double last_time = 0;
if (!clock_change && now - last_time < HANDLE_FINISHED_APPS_PERIOD) return false;
last_time = now;
vector<ACTIVE_TASK*>::iterator iter;
iter = active_tasks.active_tasks.begin();
while (iter != active_tasks.active_tasks.end()) {
atp = *iter;
switch (atp->task_state()) {
case PROCESS_EXITED:
case PROCESS_WAS_SIGNALED:
case PROCESS_EXIT_UNKNOWN:
case PROCESS_COULDNT_START:
case PROCESS_ABORTED:
if (log_flags.task) {
msg_printf(atp->wup->project, MSG_INFO,
"Computation for task %s finished", atp->result->name
);
}
app_finished(*atp);
if (!action) {
adjust_rec(); // update REC before erasing ACTIVE_TASK
}
iter = active_tasks.active_tasks.erase(iter);
delete atp;
set_client_state_dirty("handle_finished_apps");
// the following is critical; otherwise the result is
// still in the "scheduled" list and enforce_schedule()
// will try to run it again.
//
request_schedule_cpus("handle_finished_apps");
action = true;
break;
default:
++iter;
}
}
return action;
}
void ACTIVE_TASK_SET::send_trickle_downs() {
unsigned int i;
ACTIVE_TASK* atp;
bool sent;
for (i=0; i<active_tasks.size(); i++) {
atp = active_tasks[i];
if (!atp->process_exists()) continue;
if (atp->have_trickle_down) {
if (!atp->app_client_shm.shm) continue;
sent = atp->app_client_shm.shm->trickle_down.send_msg("<have_trickle_down/>\n");
if (sent) atp->have_trickle_down = false;
}
if (atp->send_upload_file_status) {
if (!atp->app_client_shm.shm) continue;
sent = atp->app_client_shm.shm->trickle_down.send_msg("<upload_file_status/>\n");
if (sent) atp->send_upload_file_status = false;
}
}
}
static void handle_result_op(char* buf, MIOFILE& fout, const char* op) {
RESULT* rp;
char result_name[256];
ACTIVE_TASK* atp;
PROJECT* p = get_project(buf, fout);
if (!p) {
fout.printf("<error>No such project</error>\n");
return;
}
if (!parse_str(buf, "<name>", result_name, sizeof(result_name))) {
fout.printf("<error>Missing result name</error>\n");
return;
}
rp = gstate.lookup_result(p, result_name);
if (!rp) {
fout.printf("<error>no such result</error>\n");
return;
}
if (!strcmp(op, "abort")) {
msg_printf(p, MSG_INFO, "task %s aborted by user", result_name);
atp = gstate.lookup_active_task_by_result(rp);
if (atp) {
atp->abort_task(ERR_ABORTED_VIA_GUI, "aborted by user");
} else {
rp->abort_inactive(ERR_ABORTED_VIA_GUI);
}
gstate.request_work_fetch("result aborted by user");
} else if (!strcmp(op, "suspend")) {
msg_printf(p, MSG_INFO, "task %s suspended by user", result_name);
rp->suspended_via_gui = true;
gstate.request_work_fetch("result suspended by user");
} else if (!strcmp(op, "resume")) {
msg_printf(p, MSG_INFO, "task %s resumed by user", result_name);
rp->suspended_via_gui = false;
}
gstate.request_schedule_cpus("result suspended, resumed or aborted by user");
gstate.set_client_state_dirty("Result RPC");
fout.printf("<success/>\n");
}
static void handle_result_show_graphics(char* buf, MIOFILE& fout) {
string result_name;
GRAPHICS_MSG gm;
ACTIVE_TASK* atp;
if (match_tag(buf, "<full_screen/>")) {
gm.mode = MODE_FULLSCREEN;
} else if (match_tag(buf, "<hide/>")) {
gm.mode = MODE_HIDE_GRAPHICS;
} else {
gm.mode = MODE_WINDOW;
}
parse_str(buf, "<window_station>", gm.window_station, sizeof(gm.window_station));
parse_str(buf, "<desktop>", gm.desktop, sizeof(gm.desktop));
parse_str(buf, "<display>", gm.display, sizeof(gm.display));
if (parse_str(buf, "<result_name>", result_name)) {
PROJECT* p = get_project(buf, fout);
if (!p) {
fout.printf("<error>No such project</error>\n");
return;
}
RESULT* rp = gstate.lookup_result(p, result_name.c_str());
if (!rp) {
fout.printf("<error>No such result</error>\n");
return;
}
atp = gstate.lookup_active_task_by_result(rp);
if (!atp) {
fout.printf("<error>no such result</error>\n");
return;
}
atp->request_graphics_mode(gm);
} else {
for (unsigned int i=0; i<gstate.active_tasks.active_tasks.size(); i++) {
atp = gstate.active_tasks.active_tasks[i];
if (atp->scheduler_state != CPU_SCHED_SCHEDULED) continue;
atp->request_graphics_mode(gm);
}
}
fout.printf("<success/>\n");
}
// suspend all currently running tasks
// called only from CLIENT_STATE::suspend_tasks(),
// e.g. because on batteries, time of day, benchmarking, CPU throttle, etc.
//
void ACTIVE_TASK_SET::suspend_all(int reason) {
for (unsigned int i=0; i<active_tasks.size(); i++) {
ACTIVE_TASK* atp = active_tasks[i];
if (atp->task_state() != PROCESS_EXECUTING) continue;
switch (reason) {
case SUSPEND_REASON_CPU_THROTTLE:
// if we're doing CPU throttling, don't bother suspending apps
// that don't use a full CPU
//
if (atp->result->dont_throttle()) continue;
if (atp->app_version->avg_ncpus < 1) continue;
atp->preempt(REMOVE_NEVER);
break;
case SUSPEND_REASON_BENCHMARKS:
atp->preempt(REMOVE_NEVER);
break;
case SUSPEND_REASON_CPU_USAGE:
// If we're suspending because of non-BOINC CPU load,
// don't remove from memory.
// Some systems do a security check when apps are launched,
// which uses a lot of CPU.
// Avoid going into a preemption loop.
//
if (atp->result->non_cpu_intensive()) break;
atp->preempt(REMOVE_NEVER);
break;
default:
atp->preempt(REMOVE_MAYBE_USER);
}
}
}
double RESULT::estimated_runtime_remaining() {
if (computing_done()) return 0;
ACTIVE_TASK* atp = gstate.lookup_active_task_by_result(this);
if (app->non_cpu_intensive) {
if (atp && atp->fraction_done>0) {
double est_dur = atp->fraction_done_elapsed_time / atp->fraction_done;
double x = est_dur - atp->elapsed_time;
if (x <= 0) x = 1;
return x;
}
return 0;
}
if (atp) {
#ifdef SIM
return sim_flops_left/avp->flops;
#else
return atp->est_dur() - atp->elapsed_time;
#endif
}
return estimated_runtime();
}
// Wait up to wait_time seconds for processes to exit
// If proj is zero, wait for all processes, else that project's
// NOTE: it's bad form to sleep, but it would be complex to avoid it here
//
int ACTIVE_TASK_SET::wait_for_exit(double wait_time, PROJECT* proj) {
bool all_exited;
unsigned int i,n;
ACTIVE_TASK *atp;
for (i=0; i<10; i++) {
all_exited = true;
for (n=0; n<active_tasks.size(); n++) {
atp = active_tasks[n];
if (proj && atp->wup->project != proj) continue;
if (!atp->has_task_exited()) {
all_exited = false;
break;
}
}
if (all_exited) return 0;
boinc_sleep(wait_time/10.0);
}
return ERR_NOT_EXITED;
}
// suspend all currently running tasks
// called only from CLIENT_STATE::suspend_tasks(),
// e.g. because on batteries, time of day, benchmarking, CPU throttle, etc.
//
void ACTIVE_TASK_SET::suspend_all(int reason) {
for (unsigned int i=0; i<active_tasks.size(); i++) {
ACTIVE_TASK* atp = active_tasks[i];
if (atp->task_state() != PROCESS_EXECUTING) continue;
switch (reason) {
case SUSPEND_REASON_CPU_THROTTLE:
// if we're doing CPU throttling, don't bother suspending apps
// that don't use a full CPU
//
if (atp->result->project->non_cpu_intensive) continue;
if (atp->app_version->avg_ncpus < 1) continue;
atp->preempt(REMOVE_NEVER);
break;
case SUSPEND_REASON_BENCHMARKS:
atp->preempt(REMOVE_NEVER);
break;
case SUSPEND_REASON_CPU_USAGE:
if (atp->result->project->non_cpu_intensive) break;
// fall through
default:
atp->preempt(REMOVE_MAYBE_USER);
}
}
}
// scan the set of all processes to
// 1) get the working-set size of active tasks
// 2) see if exclusive apps are running
// 3) get CPU time of non-BOINC processes
//
void ACTIVE_TASK_SET::get_memory_usage() {
static double last_mem_time=0;
unsigned int i;
int retval;
static bool first = true;
static double last_cpu_time;
double diff=0;
if (!first) {
diff = gstate.now - last_mem_time;
if (diff < 0 || diff > MEMORY_USAGE_PERIOD + 10) {
// user has changed system clock,
// or there has been a long system sleep
//
last_mem_time = gstate.now;
return;
}
if (diff < MEMORY_USAGE_PERIOD) return;
}
last_mem_time = gstate.now;
PROC_MAP pm;
retval = procinfo_setup(pm);
if (retval) {
if (log_flags.mem_usage_debug) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"[mem_usage] procinfo_setup() returned %d", retval
);
}
return;
}
PROCINFO boinc_total;
if (log_flags.mem_usage_debug) {
boinc_total.clear();
boinc_total.working_set_size_smoothed = 0;
}
for (i=0; i<active_tasks.size(); i++) {
ACTIVE_TASK* atp = active_tasks[i];
if (atp->task_state() == PROCESS_UNINITIALIZED) continue;
if (atp->pid ==0) continue;
// scan all active tasks with a process, even if not scheduled, because
// 1) we might have recently suspended a tasks,
// and we still need to count its time
// 2) preempted tasks might not actually suspend themselves
// (and we'd count that as non-BOINC CPU usage
// and suspend everything).
PROCINFO& pi = atp->procinfo;
unsigned long last_page_fault_count = pi.page_fault_count;
pi.clear();
pi.id = atp->pid;
vector<int>* v = NULL;
if (atp->other_pids.size()>0) {
v = &(atp->other_pids);
}
procinfo_app(pi, v, pm, atp->app_version->graphics_exec_file);
if (atp->app_version->is_vm_app) {
// the memory of virtual machine apps is not reported correctly,
// at least on Windows. Use the VM size instead.
//
pi.working_set_size_smoothed = atp->wup->rsc_memory_bound;
} else {
pi.working_set_size_smoothed = .5*(pi.working_set_size_smoothed + pi.working_set_size);
}
if (pi.working_set_size > atp->peak_working_set_size) {
atp->peak_working_set_size = pi.working_set_size;
}
if (pi.swap_size > atp->peak_swap_size) {
atp->peak_swap_size = pi.swap_size;
}
if (!first) {
int pf = pi.page_fault_count - last_page_fault_count;
pi.page_fault_rate = pf/diff;
if (log_flags.mem_usage_debug) {
msg_printf(atp->result->project, MSG_INFO,
"[mem_usage] %s%s: WS %.2fMB, smoothed %.2fMB, swap %.2fMB, %.2f page faults/sec, user CPU %.3f, kernel CPU %.3f",
atp->scheduler_state==CPU_SCHED_SCHEDULED?"":" (not running)",
atp->result->name,
pi.working_set_size/MEGA,
pi.working_set_size_smoothed/MEGA,
pi.swap_size/MEGA,
pi.page_fault_rate,
pi.user_time,
pi.kernel_time
);
boinc_total.working_set_size += pi.working_set_size;
boinc_total.working_set_size_smoothed += pi.working_set_size_smoothed;
boinc_total.swap_size += pi.swap_size;
boinc_total.page_fault_rate += pi.page_fault_rate;
}
}
}
if (!first) {
if (log_flags.mem_usage_debug) {
msg_printf(0, MSG_INFO,
"[mem_usage] BOINC totals: WS %.2fMB, smoothed %.2fMB, swap %.2fMB, %.2f page faults/sec",
boinc_total.working_set_size/MEGA,
boinc_total.working_set_size_smoothed/MEGA,
boinc_total.swap_size/MEGA,
boinc_total.page_fault_rate
);
}
}
for (i=0; i<cc_config.exclusive_apps.size(); i++) {
if (app_running(pm, cc_config.exclusive_apps[i].c_str())) {
if (log_flags.mem_usage_debug) {
msg_printf(NULL, MSG_INFO,
"[mem_usage] exclusive app %s is running", cc_config.exclusive_apps[i].c_str()
);
}
exclusive_app_running = gstate.now;
break;
}
}
for (i=0; i<cc_config.exclusive_gpu_apps.size(); i++) {
if (app_running(pm, cc_config.exclusive_gpu_apps[i].c_str())) {
if (log_flags.mem_usage_debug) {
msg_printf(NULL, MSG_INFO,
"[mem_usage] exclusive GPU app %s is running", cc_config.exclusive_gpu_apps[i].c_str()
);
}
exclusive_gpu_app_running = gstate.now;
break;
}
}
// get info on non-BOINC processes.
// mem usage info is not useful because most OSs don't
// move idle processes out of RAM, so physical memory is always full.
// Also (at least on Win) page faults are used for various things,
// not all of them generate disk I/O,
// so they're not useful for detecting paging/thrashing.
//
PROCINFO pi;
procinfo_non_boinc(pi, pm);
if (log_flags.mem_usage_debug) {
//procinfo_show(pm);
msg_printf(NULL, MSG_INFO,
"[mem_usage] All others: WS %.2fMB, swap %.2fMB, user %.3fs, kernel %.3fs",
pi.working_set_size/MEGA, pi.swap_size/MEGA,
pi.user_time, pi.kernel_time
);
}
double new_cpu_time = pi.user_time + pi.kernel_time;
if (!first) {
non_boinc_cpu_usage = (new_cpu_time - last_cpu_time)/(diff*gstate.host_info.p_ncpus);
// processes might have exited in the last 10 sec,
// causing this to be negative.
if (non_boinc_cpu_usage < 0) non_boinc_cpu_usage = 0;
if (log_flags.mem_usage_debug) {
msg_printf(NULL, MSG_INFO,
"[mem_usage] non-BOINC CPU usage: %.2f%%", non_boinc_cpu_usage*100
);
}
}
last_cpu_time = new_cpu_time;
first = false;
}
void ACTIVE_TASK_SET::handle_upload_files() {
for (unsigned int i=0; i<active_tasks.size(); i++) {
ACTIVE_TASK* atp = active_tasks[i];
atp->handle_upload_files();
}
}
// Handle the reply from a scheduler
//
int CLIENT_STATE::handle_scheduler_reply(
PROJECT* project, char* scheduler_url
) {
SCHEDULER_REPLY sr;
FILE* f;
int retval;
unsigned int i;
bool signature_valid, update_global_prefs=false, update_project_prefs=false;
char buf[1024], filename[256];
std::string old_gui_urls = project->gui_urls;
PROJECT* p2;
vector<RESULT*>new_results;
project->last_rpc_time = now;
if (requested_work()) {
had_or_requested_work = true;
}
get_sched_reply_filename(*project, filename, sizeof(filename));
f = fopen(filename, "r");
if (!f) return ERR_FOPEN;
retval = sr.parse(f, project);
fclose(f);
if (retval) return retval;
if (log_flags.sched_ops) {
if (requested_work()) {
sprintf(buf, ": got %d new tasks", (int)sr.results.size());
} else {
strcpy(buf, "");
}
msg_printf(project, MSG_INFO, "Scheduler request completed%s", buf);
}
if (log_flags.sched_op_debug) {
if (sr.scheduler_version) {
msg_printf(project, MSG_INFO,
"[sched_op] Server version %d",
sr.scheduler_version
);
}
}
// check that master URL is correct
//
if (strlen(sr.master_url)) {
canonicalize_master_url(sr.master_url);
string url1 = sr.master_url;
string url2 = project->master_url;
downcase_string(url1);
downcase_string(url2);
if (url1 != url2) {
p2 = lookup_project(sr.master_url);
if (p2) {
msg_printf(project, MSG_USER_ALERT,
"You are attached to this project twice. Please remove projects named %s, then add %s",
project->project_name,
sr.master_url
);
} else {
msg_printf(project, MSG_INFO,
_("You used the wrong URL for this project. When convenient, remove this project, then add %s"),
sr.master_url
);
}
}
}
// make sure we don't already have a project of same name
//
bool dup_name = false;
for (i=0; i<projects.size(); i++) {
p2 = projects[i];
if (project == p2) continue;
if (!strcmp(p2->project_name, project->project_name)) {
dup_name = true;
break;
}
}
if (dup_name) {
msg_printf(project, MSG_INFO,
"Already attached to a project named %s (possibly with wrong URL)",
project->project_name
);
msg_printf(project, MSG_INFO,
"Consider detaching this project, then trying again"
);
}
// show messages from server
//
for (i=0; i<sr.messages.size(); i++) {
USER_MESSAGE& um = sr.messages[i];
int prio = (!strcmp(um.priority.c_str(), "notice"))?MSG_SCHEDULER_ALERT:MSG_INFO;
string_substitute(um.message.c_str(), buf, sizeof(buf), "%", "%%");
msg_printf(project, prio, "%s", buf);
}
if (log_flags.sched_op_debug && sr.request_delay) {
msg_printf(project, MSG_INFO,
"Project requested delay of %.0f seconds", sr.request_delay
);
}
// if project is down, return error (so that we back off)
// and don't do anything else
//
if (sr.project_is_down) {
if (sr.request_delay) {
double x = now + sr.request_delay;
project->set_min_rpc_time(x, "project is down");
}
return ERR_PROJECT_DOWN;
}
// if the scheduler reply includes global preferences,
// insert extra elements, write to disk, and parse
//
if (sr.global_prefs_xml) {
// skip this if we have host-specific prefs
// and we're talking to an old scheduler
//
if (!global_prefs.host_specific || sr.scheduler_version >= 507) {
retval = save_global_prefs(
sr.global_prefs_xml, project->master_url, scheduler_url
);
if (retval) {
return retval;
}
update_global_prefs = true;
} else {
if (log_flags.sched_op_debug) {
msg_printf(project, MSG_INFO,
"ignoring prefs from old server; we have host-specific prefs"
);
}
}
}
// see if we have a new venue from this project
// (this must go AFTER the above, since otherwise
// global_prefs_source_project() is meaningless)
//
if (strcmp(project->host_venue, sr.host_venue)) {
safe_strcpy(project->host_venue, sr.host_venue);
msg_printf(project, MSG_INFO, "New computer location: %s", sr.host_venue);
update_project_prefs = true;
if (project == global_prefs_source_project()) {
strcpy(main_host_venue, sr.host_venue);
update_global_prefs = true;
}
}
if (update_global_prefs) {
read_global_prefs();
}
// deal with project preferences (should always be there)
// If they've changed, write to account file,
// then parse to get our venue, and pass to running apps
//
if (sr.project_prefs_xml) {
if (strcmp(project->project_prefs.c_str(), sr.project_prefs_xml)) {
project->project_prefs = string(sr.project_prefs_xml);
update_project_prefs = true;
}
}
// the account file has GUI URLs and project prefs.
// rewrite if either of these has changed
//
if (project->gui_urls != old_gui_urls || update_project_prefs) {
retval = project->write_account_file();
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't write account file: %s", boincerror(retval)
);
return retval;
}
}
if (update_project_prefs) {
project->parse_account_file();
if (strlen(project->host_venue)) {
project->parse_account_file_venue();
}
project->parse_preferences_for_user_files();
active_tasks.request_reread_prefs(project);
}
// if the scheduler reply includes a code-signing key,
// accept it if we don't already have one from the project.
// Otherwise verify its signature, using the key we already have.
//
if (sr.code_sign_key) {
if (!strlen(project->code_sign_key)) {
safe_strcpy(project->code_sign_key, sr.code_sign_key);
} else {
if (sr.code_sign_key_signature) {
retval = check_string_signature2(
sr.code_sign_key, sr.code_sign_key_signature,
project->code_sign_key, signature_valid
);
if (!retval && signature_valid) {
safe_strcpy(project->code_sign_key, sr.code_sign_key);
} else {
msg_printf(project, MSG_INTERNAL_ERROR,
"New code signing key doesn't validate"
);
}
} else {
msg_printf(project, MSG_INTERNAL_ERROR,
"Missing code sign key signature"
);
}
}
}
// copy new entities to client state
//
for (i=0; i<sr.apps.size(); i++) {
APP* app = lookup_app(project, sr.apps[i].name);
if (app) {
strcpy(app->user_friendly_name, sr.apps[i].user_friendly_name);
} else {
app = new APP;
*app = sr.apps[i];
retval = link_app(project, app);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't handle application %s in scheduler reply", app->name
);
delete app;
} else {
apps.push_back(app);
}
}
}
FILE_INFO* fip;
for (i=0; i<sr.file_infos.size(); i++) {
fip = lookup_file_info(project, sr.file_infos[i].name);
if (fip) {
fip->merge_info(sr.file_infos[i]);
} else {
fip = new FILE_INFO;
*fip = sr.file_infos[i];
retval = link_file_info(project, fip);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't handle file %s in scheduler reply", fip->name
);
delete fip;
} else {
file_infos.push_back(fip);
}
}
}
for (i=0; i<sr.file_deletes.size(); i++) {
fip = lookup_file_info(project, sr.file_deletes[i].c_str());
if (fip) {
if (log_flags.file_xfer_debug) {
msg_printf(project, MSG_INFO,
"[file_xfer_debug] Got server request to delete file %s",
fip->name
);
}
fip->sticky = false;
}
}
for (i=0; i<sr.app_versions.size(); i++) {
if (project->anonymous_platform) {
msg_printf(project, MSG_INTERNAL_ERROR,
"App version returned from anonymous platform project; ignoring"
);
continue;
}
APP_VERSION& avpp = sr.app_versions[i];
if (strlen(avpp.platform) == 0) {
strcpy(avpp.platform, get_primary_platform());
} else {
if (!is_supported_platform(avpp.platform)) {
msg_printf(project, MSG_INTERNAL_ERROR,
"App version has unsupported platform %s", avpp.platform
);
continue;
}
}
if (avpp.missing_coproc) {
msg_printf(project, MSG_INTERNAL_ERROR,
"App version uses non-existent %s GPU",
avpp.missing_coproc_name
);
}
APP* app = lookup_app(project, avpp.app_name);
if (!app) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Missing app %s", avpp.app_name
);
continue;
}
APP_VERSION* avp = lookup_app_version(
app, avpp.platform, avpp.version_num, avpp.plan_class
);
if (avp) {
// update performance-related info;
// generally this shouldn't change,
// but if it does it's better to use the new stuff
//
avp->avg_ncpus = avpp.avg_ncpus;
avp->max_ncpus = avpp.max_ncpus;
avp->flops = avpp.flops;
strcpy(avp->cmdline, avpp.cmdline);
avp->gpu_usage = avpp.gpu_usage;
strlcpy(avp->api_version, avpp.api_version, sizeof(avp->api_version));
avp->dont_throttle = avpp.dont_throttle;
avp->needs_network = avpp.needs_network;
// if we had download failures, clear them
//
avp->clear_errors();
continue;
}
avp = new APP_VERSION;
*avp = avpp;
retval = link_app_version(project, avp);
if (retval) {
delete avp;
continue;
}
app_versions.push_back(avp);
}
for (i=0; i<sr.workunits.size(); i++) {
if (lookup_workunit(project, sr.workunits[i].name)) continue;
WORKUNIT* wup = new WORKUNIT;
*wup = sr.workunits[i];
wup->project = project;
retval = link_workunit(project, wup);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't handle task %s in scheduler reply", wup->name
);
delete wup;
continue;
}
wup->clear_errors();
workunits.push_back(wup);
}
double est_rsc_runtime[MAX_RSC];
for (int j=0; j<coprocs.n_rsc; j++) {
est_rsc_runtime[j] = 0;
}
for (i=0; i<sr.results.size(); i++) {
if (lookup_result(project, sr.results[i].name)) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Already have task %s\n", sr.results[i].name
);
continue;
}
RESULT* rp = new RESULT;
*rp = sr.results[i];
retval = link_result(project, rp);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't handle task %s in scheduler reply", rp->name
);
delete rp;
continue;
}
if (strlen(rp->platform) == 0) {
strcpy(rp->platform, get_primary_platform());
rp->version_num = latest_version(rp->wup->app, rp->platform);
}
rp->avp = lookup_app_version(
rp->wup->app, rp->platform, rp->version_num, rp->plan_class
);
if (!rp->avp) {
msg_printf(project, MSG_INTERNAL_ERROR,
"No app version found for app %s platform %s ver %d class %s; discarding %s",
rp->wup->app->name, rp->platform, rp->version_num, rp->plan_class, rp->name
);
delete rp;
continue;
}
if (rp->avp->missing_coproc) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Missing coprocessor for task %s; aborting", rp->name
);
rp->abort_inactive(EXIT_MISSING_COPROC);
} else {
rp->set_state(RESULT_NEW, "handle_scheduler_reply");
int rt = rp->avp->gpu_usage.rsc_type;
if (rt > 0) {
est_rsc_runtime[rt] += rp->estimated_runtime();
gpus_usable = true;
// trigger a check of whether GPU is actually usable
} else {
est_rsc_runtime[0] += rp->estimated_runtime();
}
}
rp->wup->version_num = rp->version_num;
rp->received_time = now;
new_results.push_back(rp);
results.push_back(rp);
}
sort_results();
if (log_flags.sched_op_debug) {
if (sr.results.size()) {
for (int j=0; j<coprocs.n_rsc; j++) {
msg_printf(project, MSG_INFO,
"[sched_op] estimated total %s task duration: %.0f seconds",
rsc_name(j),
est_rsc_runtime[j]/time_stats.availability_frac(j)
);
}
}
}
// update records for ack'ed results
//
for (i=0; i<sr.result_acks.size(); i++) {
if (log_flags.sched_op_debug) {
msg_printf(project, MSG_INFO,
"[sched_op] handle_scheduler_reply(): got ack for task %s\n",
sr.result_acks[i].name
);
}
RESULT* rp = lookup_result(project, sr.result_acks[i].name);
if (rp) {
rp->got_server_ack = true;
} else {
msg_printf(project, MSG_INTERNAL_ERROR,
"Got ack for task %s, but can't find it", sr.result_acks[i].name
);
}
}
// handle result abort requests
//
for (i=0; i<sr.result_abort.size(); i++) {
RESULT* rp = lookup_result(project, sr.result_abort[i].name);
if (rp) {
ACTIVE_TASK* atp = lookup_active_task_by_result(rp);
if (atp) {
atp->abort_task(EXIT_ABORTED_BY_PROJECT,
"aborted by project - no longer usable"
);
} else {
rp->abort_inactive(EXIT_ABORTED_BY_PROJECT);
}
} else {
msg_printf(project, MSG_INTERNAL_ERROR,
"Server requested abort of unknown task %s",
sr.result_abort[i].name
);
}
}
for (i=0; i<sr.result_abort_if_not_started.size(); i++) {
RESULT* rp = lookup_result(project, sr.result_abort_if_not_started[i].name);
if (!rp) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Server requested conditional abort of unknown task %s",
sr.result_abort_if_not_started[i].name
);
continue;
}
if (rp->not_started) {
rp->abort_inactive(EXIT_ABORTED_BY_PROJECT);
}
}
// remove acked trickle files
//
if (sr.message_ack) {
remove_trickle_files(project);
}
if (sr.send_full_workload) {
project->send_full_workload = true;
}
project->dont_use_dcf = sr.dont_use_dcf;
project->send_time_stats_log = sr.send_time_stats_log;
project->send_job_log = sr.send_job_log;
project->trickle_up_pending = false;
// The project returns a hostid only if it has created a new host record.
// In that case reset RPC seqno
//
if (sr.hostid) {
if (project->hostid) {
// if we already have a host ID for this project,
// we must have sent it a stale seqno,
// which usually means our state file was copied from another host.
// So generate a new host CPID.
//
generate_new_host_cpid();
msg_printf(project, MSG_INFO,
"Generated new computer cross-project ID: %s",
host_info.host_cpid
);
}
//msg_printf(project, MSG_INFO, "Changing host ID from %d to %d", project->hostid, sr.hostid);
project->hostid = sr.hostid;
project->rpc_seqno = 0;
}
#ifdef ENABLE_AUTO_UPDATE
if (sr.auto_update.present) {
if (!sr.auto_update.validate_and_link(project)) {
auto_update = sr.auto_update;
}
}
#endif
project->project_files = sr.project_files;
project->link_project_files();
project->create_project_file_symlinks();
if (log_flags.state_debug) {
msg_printf(project, MSG_INFO,
"[state] handle_scheduler_reply(): State after handle_scheduler_reply():"
);
print_summary();
}
// the following must precede the backoff and request_delay checks,
// since it overrides them
//
if (sr.next_rpc_delay) {
project->next_rpc_time = now + sr.next_rpc_delay;
} else {
project->next_rpc_time = 0;
}
work_fetch.handle_reply(project, &sr, new_results);
project->nrpc_failures = 0;
project->min_rpc_time = 0;
if (sr.request_delay) {
double x = now + sr.request_delay;
project->set_min_rpc_time(x, "requested by project");
}
if (sr.got_rss_feeds) {
handle_sr_feeds(sr.sr_feeds, project);
}
update_trickle_up_urls(project, sr.trickle_up_urls);
// garbage collect in case the project sent us some irrelevant FILE_INFOs;
// avoid starting transfers for them
//
gstate.garbage_collect_always();
return 0;
}
void simulate() {
bool action;
double start = START_TIME;
gstate.now = start;
html_start();
fprintf(summary_file,
"Hardware summary\n %d CPUs, %.1f GFLOPS\n",
gstate.host_info.p_ncpus, gstate.host_info.p_fpops/1e9
);
for (int i=1; i<coprocs.n_rsc; i++) {
fprintf(summary_file,
" %d %s GPUs, %.1f GFLOPS\n",
coprocs.coprocs[i].count,
coprocs.coprocs[i].type,
coprocs.coprocs[i].peak_flops/1e9
);
}
fprintf(summary_file,
"Preferences summary\n"
" work buf min %f max %f\n"
" Scheduling period %f\n"
"Scheduling policies\n"
" Round-robin only: %s\n"
" Scheduler EDF simulation: %s\n"
" REC half-life: %f\n",
gstate.work_buf_min(), gstate.work_buf_total(),
gstate.global_prefs.cpu_scheduling_period(),
cpu_sched_rr_only?"yes":"no",
server_uses_workload?"yes":"no",
cc_config.rec_half_life
);
fprintf(summary_file, "Jobs\n");
for (unsigned int i=0; i<gstate.results.size(); i++) {
RESULT* rp = gstate.results[i];
fprintf(summary_file,
" %s %s (%s)\n time left %s deadline %s\n",
rp->project->project_name,
rp->name,
rsc_name_long(rp->avp->gpu_usage.rsc_type),
timediff_format(rp->sim_flops_left/rp->avp->flops).c_str(),
timediff_format(rp->report_deadline - START_TIME).c_str()
);
}
fprintf(summary_file,
"Simulation parameters\n"
" time step %f, duration %f\n"
"-------------------\n",
delta, duration
);
write_inputs();
while (1) {
on = on_proc.sample(delta);
if (on) {
active = active_proc.sample(delta);
if (active) {
gpu_active = gpu_active_proc.sample(delta);
} else {
gpu_active = false;
}
connected = connected_proc.sample(delta);
} else {
active = gpu_active = connected = false;
}
// do accounting for the period that just ended,
// even if we're now in an "off" state.
//
// need both of the following, else crash
//
action |= gstate.active_tasks.poll();
action |= gstate.handle_finished_apps();
if (on) {
while (1) {
action = false;
action |= gstate.schedule_cpus();
if (connected) {
action |= gstate.scheduler_rpc_poll();
// this deletes completed results
}
action |= gstate.active_tasks.poll();
action |= gstate.handle_finished_apps();
gpu_suspend_reason = gpu_active?0:1;
//msg_printf(0, MSG_INFO, action?"did action":"did no action");
if (!action) break;
}
}
//msg_printf(0, MSG_INFO, "took time step");
for (unsigned int i=0; i<gstate.active_tasks.active_tasks.size(); i++) {
ACTIVE_TASK* atp = gstate.active_tasks.active_tasks[i];
if (atp->task_state() == PROCESS_EXECUTING) {
atp->elapsed_time += delta;
}
}
html_rec();
write_recs();
gstate.now += delta;
if (gstate.now > start + duration) break;
}
html_end();
}
bool ACTIVE_TASK_SET::poll() {
unsigned int i;
char buf[256];
bool action = false;
static double last_time = START_TIME;
double diff = gstate.now - last_time;
if (diff < 1.0) return false;
last_time = gstate.now;
if (diff > delta) {
diff = 0;
}
PROJECT* p;
for (i=0; i<gstate.projects.size(); i++) {
p = gstate.projects[i];
p->idle = true;
}
// we do two kinds of FLOPs accounting:
// 1) actual FLOPS (for job completion)
// 2) peak FLOPS (for total and per-project resource usage)
//
// CPU may be overcommitted, in which case we compute
// a "cpu_scale" factor that is < 1.
// GPUs are never overcommitted.
//
// actual FLOPS is based on app_version.flops, scaled by cpu_scale for CPU jobs
// peak FLOPS is based on device peak FLOPS,
// with CPU component scaled by cpu_scale for all jobs
// get CPU usage by GPU and CPU jobs
//
double cpu_usage_cpu=0;
double cpu_usage_gpu=0;
for (i=0; i<active_tasks.size(); i++) {
ACTIVE_TASK* atp = active_tasks[i];
if (atp->task_state() != PROCESS_EXECUTING) continue;
RESULT* rp = atp->result;
if (rp->uses_gpu()) {
if (gpu_active) {
cpu_usage_gpu += rp->avp->avg_ncpus;
}
} else {
cpu_usage_cpu += rp->avp->avg_ncpus;
}
}
double cpu_usage = cpu_usage_cpu + cpu_usage_gpu;
// if CPU is overcommitted, compute cpu_scale
//
double cpu_scale = 1;
if (cpu_usage > gstate.ncpus) {
cpu_scale = (gstate.ncpus - cpu_usage_gpu) / (cpu_usage - cpu_usage_gpu);
}
double used = 0;
for (i=0; i<active_tasks.size(); i++) {
ACTIVE_TASK* atp = active_tasks[i];
if (atp->task_state() != PROCESS_EXECUTING) continue;
RESULT* rp = atp->result;
if (!gpu_active && rp->uses_gpu()) {
continue;
}
atp->elapsed_time += diff;
double flops = rp->avp->flops;
if (!rp->uses_gpu()) {
flops *= cpu_scale;
}
rp->sim_flops_left -= diff*flops;
atp->fraction_done = 1 - rp->sim_flops_left / rp->wup->rsc_fpops_est;
atp->checkpoint_wall_time = gstate.now;
if (rp->sim_flops_left <= 0) {
atp->set_task_state(PROCESS_EXITED, "poll");
rp->exit_status = 0;
rp->ready_to_report = true;
gstate.request_schedule_cpus("job finished");
gstate.request_work_fetch("job finished");
sprintf(buf, "result %s finished<br>", rp->name);
html_msg += buf;
action = true;
}
double pf = diff * app_peak_flops(rp->avp, cpu_scale);
rp->project->project_results.flops_used += pf;
rp->peak_flop_count += pf;
sim_results.flops_used += pf;
used += pf;
rp->project->idle = false;
}
for (i=0; i<gstate.projects.size(); i++) {
p = gstate.projects[i];
if (p->idle) {
p->idle_time += diff;
p->idle_time_sumsq += diff*(p->idle_time*p->idle_time);
} else {
p->idle_time = 0;
}
}
active_time += diff;
if (gpu_active) {
gpu_active_time += diff;
}
return action;
}
void show_resource(int rsc_type) {
unsigned int i;
char buf[256];
fprintf(html_out, "<td width=%d valign=top>", WIDTH2);
bool found = false;
for (i=0; i<gstate.active_tasks.active_tasks.size(); i++) {
ACTIVE_TASK* atp = gstate.active_tasks.active_tasks[i];
RESULT* rp = atp->result;
if (atp->task_state() != PROCESS_EXECUTING) continue;
double ninst=0;
if (rsc_type) {
if (rp->avp->gpu_usage.rsc_type != rsc_type) continue;
ninst = rp->avp->gpu_usage.usage;
} else {
ninst = rp->avp->avg_ncpus;
}
PROJECT* p = rp->project;
if (!found) {
found = true;
fprintf(html_out,
"<table>\n"
"<tr><th>#devs</th><th>Job name (* = high priority)</th><th>GFLOPs left</th>%s</tr>\n",
rsc_type?"<th>GPU</th>":""
);
}
if (rsc_type) {
sprintf(buf, "<td>%d</td>", rp->coproc_indices[0]);
} else {
safe_strcpy(buf, "");
}
fprintf(html_out, "<tr valign=top><td>%.2f</td><td bgcolor=%s><font color=#ffffff>%s%s</font></td><td>%.0f</td>%s</tr>\n",
ninst,
colors[p->index%NCOLORS],
rp->edf_scheduled?"*":"",
rp->name,
rp->sim_flops_left/1e9,
buf
);
}
if (found) {
fprintf(html_out, "</table>\n");
} else {
fprintf(html_out, "IDLE\n");
}
fprintf(html_out,
"<table><tr><td>Project</td><td>In progress</td><td>done</td><td>REC</td></tr>\n"
);
found = false;
for (i=0; i<gstate.projects.size(); i++) {
PROJECT* p = gstate.projects[i];
int in_progress, done;
job_count(p, rsc_type, in_progress, done);
if (in_progress || done) {
fprintf(html_out, "<td bgcolor=%s><font color=#ffffff>%s</font></td><td>%d</td><td>%d</td><td>%.3f</td></tr>\n",
colors[p->index%NCOLORS], p->project_name, in_progress, done,
p->pwf.rec
);
found = true;
}
}
//if (!found) fprintf(html_out, " ---\n");
fprintf(html_out, "</table></td>");
}
// The following runs "test_app" and sends it various messages.
// Used for testing the runtime system.
//
void run_test_app() {
WORKUNIT wu;
PROJECT project;
APP app;
APP_VERSION av;
ACTIVE_TASK at;
ACTIVE_TASK_SET ats;
RESULT result;
int retval;
char buf[256];
getcwd(buf, sizeof(buf)); // so we can see where we're running
gstate.run_test_app = true;
wu.project = &project;
wu.app = &app;
wu.command_line = string("--critical_section");
strcpy(app.name, "test app");
av.init();
av.avg_ncpus = 1;
strcpy(result.name, "test result");
result.avp = &av;
result.wup = &wu;
result.project = &project;
result.app = &app;
at.result = &result;
at.wup = &wu;
at.app_version = &av;
at.max_elapsed_time = 1e6;
at.max_disk_usage = 1e14;
at.max_mem_usage = 1e14;
strcpy(at.slot_dir, ".");
#if 1
// test file copy
//
ASYNC_COPY* ac = new ASYNC_COPY;
FILE_INFO fi;
retval = ac->init(&at, &fi, "big_file", "./big_file_copy");
if (retval) {
exit(1);
}
while (1) {
do_async_file_ops();
if (at.async_copy == NULL) {
break;
}
}
fprintf(stderr, "done\n");
exit(0);
#endif
ats.active_tasks.push_back(&at);
unlink("boinc_finish_called");
unlink("boinc_lockfile");
unlink("boinc_temporary_exit");
unlink("stderr.txt");
retval = at.start(true);
if (retval) {
fprintf(stderr, "start() failed: %s\n", boincerror(retval));
}
while (1) {
gstate.now = dtime();
at.preempt(REMOVE_NEVER);
ats.poll();
boinc_sleep(.1);
at.unsuspend();
ats.poll();
boinc_sleep(.2);
//at.request_reread_prefs();
}
}
int RESULT::write_gui(MIOFILE& out) {
out.printf(
"<result>\n"
" <name>%s</name>\n"
" <wu_name>%s</wu_name>\n"
" <version_num>%d</version_num>\n"
" <plan_class>%s</plan_class>\n"
" <project_url>%s</project_url>\n"
" <final_cpu_time>%f</final_cpu_time>\n"
" <final_elapsed_time>%f</final_elapsed_time>\n"
" <exit_status>%d</exit_status>\n"
" <state>%d</state>\n"
" <report_deadline>%f</report_deadline>\n"
" <received_time>%f</received_time>\n"
" <estimated_cpu_time_remaining>%f</estimated_cpu_time_remaining>\n",
name,
wu_name,
version_num,
plan_class,
project->master_url,
final_cpu_time,
final_elapsed_time,
exit_status,
state(),
report_deadline,
received_time,
estimated_runtime_remaining()
);
if (got_server_ack) out.printf(" <got_server_ack/>\n");
if (ready_to_report) out.printf(" <ready_to_report/>\n");
if (completed_time) out.printf(" <completed_time>%f</completed_time>\n", completed_time);
if (suspended_via_gui) out.printf(" <suspended_via_gui/>\n");
if (project->suspended_via_gui) out.printf(" <project_suspended_via_gui/>\n");
if (report_immediately) out.printf(" <report_immediately/>\n");
if (edf_scheduled) out.printf(" <edf_scheduled/>\n");
if (coproc_missing) out.printf(" <coproc_missing/>\n");
if (schedule_backoff > gstate.now) {
out.printf(" <scheduler_wait/>\n");
if (strlen(schedule_backoff_reason)) {
out.printf(
" <scheduler_wait_reason>%s</scheduler_wait_reason>\n",
schedule_backoff_reason
);
}
}
if (avp->needs_network && gstate.network_suspended) out.printf(" <network_wait/>\n");
ACTIVE_TASK* atp = gstate.active_tasks.lookup_result(this);
if (atp) {
atp->write_gui(out);
}
if (!strlen(resources)) {
// only need to compute this string once
//
if (avp->gpu_usage.rsc_type) {
if (avp->gpu_usage.usage == 1) {
sprintf(resources,
"%.3g CPUs + 1 %s",
avp->avg_ncpus,
rsc_name_long(avp->gpu_usage.rsc_type)
);
} else {
sprintf(resources,
"%.3g CPUs + %.3g %ss",
avp->avg_ncpus,
avp->gpu_usage.usage,
rsc_name_long(avp->gpu_usage.rsc_type)
);
}
} else if (avp->missing_coproc) {
sprintf(resources, "%.3g CPUs + %s GPU (missing)",
avp->avg_ncpus, avp->missing_coproc_name
);
} else if (!project->non_cpu_intensive && (avp->avg_ncpus != 1)) {
sprintf(resources, "%.3g CPUs", avp->avg_ncpus);
} else {
strcpy(resources, " ");
}
}
if (strlen(resources)>1) {
char buf[256];
strcpy(buf, "");
if (atp && atp->scheduler_state == CPU_SCHED_SCHEDULED) {
if (avp->gpu_usage.rsc_type) {
COPROC& cp = coprocs.coprocs[avp->gpu_usage.rsc_type];
if (cp.count > 1) {
// if there are multiple GPUs of this type,
// show the user which one(s) are being used
//
int n = (int)ceil(avp->gpu_usage.usage);
strcpy(buf, " (device ");
for (int i=0; i<n; i++) {
char buf2[256];
sprintf(buf2, "%d", cp.device_nums[coproc_indices[i]]);
if (i > 0) {
strcat(buf, "/");
}
strcat(buf, buf2);
}
strcat(buf, ")");
}
}
}
out.printf(
" <resources>%s%s</resources>\n", resources, buf
);
}
out.printf("</result>\n");
return 0;
}
