// Decide what jobs to include in the simulation;
// build the "pending" lists for each (project, processor type) pair.
// NOTE: "results" is sorted by increasing arrival time.
//
void RR_SIM::init_pending_lists() {
for (unsigned int i=0; i<gstate.projects.size(); i++) {
PROJECT* p = gstate.projects[i];
for (int j=0; j<coprocs.n_rsc; j++) {
p->rsc_pwf[j].pending.clear();
p->rsc_pwf[j].queue_est = 0;
}
}
for (unsigned int i=0; i<gstate.results.size(); i++) {
RESULT* rp = gstate.results[i];
rp->rr_sim_misses_deadline = false;
rp->already_selected = false;
if (!rp->nearly_runnable()) continue;
if (rp->some_download_stalled()) continue;
if (rp->project->non_cpu_intensive) continue;
rp->rrsim_flops_left = rp->estimated_flops_remaining();
//if (rp->rrsim_flops_left <= 0) continue;
// job may have fraction_done=1 but not be done;
// if it's past its deadline, we need to mark it as such
PROJECT* p = rp->project;
p->pwf.n_runnable_jobs++;
p->rsc_pwf[0].nused_total += rp->avp->avg_ncpus;
set_rrsim_flops(rp);
int rt = rp->avp->gpu_usage.rsc_type;
if (rt) {
p->rsc_pwf[rt].nused_total += rp->avp->gpu_usage.usage;
p->rsc_pwf[rt].n_runnable_jobs++;
p->rsc_pwf[rt].queue_est += rp->rrsim_flops_left/rp->rrsim_flops;
}
p->rsc_pwf[rt].pending.push_back(rp);
rp->rrsim_done = false;
}
}
void CLIENT_STATE::compute_nuploading_results() {
unsigned int i;
for (i=0; i<projects.size(); i++) {
projects[i]->nuploading_results = 0;
projects[i]->too_many_uploading_results = false;
}
for (i=0; i<results.size(); i++) {
RESULT* rp = results[i];
if (rp->state() == RESULT_FILES_UPLOADING) {
rp->project->nuploading_results++;
}
}
int n = gstate.ncpus;
for (int j=1; j<coprocs.n_rsc; j++) {
if (coprocs.coprocs[j].count > n) {
n = coprocs.coprocs[j].count;
}
}
n *= 2;
for (i=0; i<projects.size(); i++) {
if (projects[i]->nuploading_results > n) {
projects[i]->too_many_uploading_results = true;
}
}
}
void CLIENT_STATE::compute_nuploading_results() {
unsigned int i;
for (i=0; i<projects.size(); i++) {
projects[i]->nuploading_results = 0;
projects[i]->too_many_uploading_results = false;
}
for (i=0; i<results.size(); i++) {
RESULT* rp = results[i];
if (rp->state() == RESULT_FILES_UPLOADING) {
rp->project->nuploading_results++;
}
}
int n = gstate.ncpus;
if (gstate.host_info.have_cuda() && gstate.host_info.coprocs.cuda.count > n) {
n = gstate.host_info.coprocs.cuda.count;
}
if (gstate.host_info.have_ati() && gstate.host_info.coprocs.ati.count > n) {
n = gstate.host_info.coprocs.ati.count;
}
n *= 2;
for (i=0; i<projects.size(); i++) {
if (projects[i]->nuploading_results > n) {
projects[i]->too_many_uploading_results = true;
}
}
}
void PROJECT::abort_not_started() {
for (unsigned int i=0; i<gstate.results.size(); i++) {
RESULT* rp = gstate.results[i];
if (rp->project != this) continue;
if (rp->is_not_started()) {
rp->abort_inactive(EXIT_ABORTED_VIA_GUI);
}
}
}
bool PROJECT::downloading() {
if (suspended_via_gui) return false;
for (unsigned int i=0; i<gstate.results.size(); i++) {
RESULT* rp = gstate.results[i];
if (rp->project != this) continue;
if (rp->downloading()) return true;
}
return 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);
}
}
}
// does the project have a downloading or runnable job?
//
static bool has_a_job(PROJECT* p) {
for (unsigned int j=0; j<gstate.results.size(); j++) {
RESULT* rp = gstate.results[j];
if (rp->project != p) continue;
if (rp->state() <= RESULT_FILES_DOWNLOADED) {
return true;
}
}
return false;
}
// convert results in progress to IP_RESULTs,
// and get an initial schedule for them
//
void CLIENT_STATE::get_workload(vector<IP_RESULT>& ip_results) {
for (unsigned int i=0; i<results.size(); i++) {
RESULT* rp = results[i];
double x = rp->estimated_runtime_remaining();
if (x == 0) continue;
IP_RESULT ipr(rp->name, rp->report_deadline-now, x);
ip_results.push_back(ipr);
}
//init_ip_results(work_buf_min(), ncpus, ip_results);
init_ip_results(0, ncpus, ip_results);
}
// choose a project to fetch work from,
// and set the request fields of resource objects.
//
PROJECT* WORK_FETCH::choose_project(
bool strict_hyst,
PROJECT* backoff_exempt_project
) {
PROJECT* p;
if (log_flags.work_fetch_debug) {
msg_printf(0, MSG_INFO, "[work_fetch] work fetch start");
}
p = non_cpu_intensive_project_needing_work();
if (p) return p;
gstate.compute_nuploading_results();
rr_simulation();
compute_shares();
project_priority_init(true);
// Decrement the priority of projects that have a lot of work queued.
// Specifically, subtract
// (FLOPs queued for P)/(FLOPs of max queue)
// which will generally be between 0 and 1.
// This is a little arbitrary but I can't think of anything better.
//
double max_queued_flops = gstate.work_buf_total()*total_peak_flops();
for (unsigned int i=0; i<gstate.results.size(); i++) {
RESULT* rp = gstate.results[i];
p = rp->project;
p->sched_priority -= rp->estimated_flops_remaining()/max_queued_flops;
}
p = 0;
if (gpus_usable) {
for (int i=1; i<coprocs.n_rsc; i++) {
p = rsc_work_fetch[i].choose_project_hyst(strict_hyst, backoff_exempt_project);
if (p) break;
}
}
if (!p) {
p = rsc_work_fetch[0].choose_project_hyst(strict_hyst, backoff_exempt_project);
}
if (log_flags.work_fetch_debug) {
print_state();
if (!p) {
msg_printf(0, MSG_INFO, "[work_fetch] No project chosen for work fetch");
}
}
return p;
}
void job_count(PROJECT* p, int rsc_type, int& in_progress, int& done) {
in_progress = done = 0;
unsigned int i;
for (i=0; i<gstate.results.size(); i++) {
RESULT* rp = gstate.results[i];
if (rp->project != p) continue;
if (rp->resource_type() != rsc_type) continue;
if (rp->state() < RESULT_FILES_UPLOADED) {
in_progress++;
} else {
done++;
}
}
}
bool PROJECT::runnable(int rsc_type) {
if (suspended_via_gui) return false;
for (unsigned int i=0; i<gstate.results.size(); i++) {
RESULT* rp = gstate.results[i];
if (rp->project != this) continue;
if (rsc_type != RSC_TYPE_ANY) {
if (rp->avp->gpu_usage.rsc_type != rsc_type) {
continue;
}
}
if (rp->runnable()) return true;
}
return false;
}
void PROJECT::get_task_durs(double& not_started_dur, double& in_progress_dur) {
not_started_dur = 0;
in_progress_dur = 0;
for (unsigned int i=0; i<gstate.results.size(); i++) {
RESULT* rp = gstate.results[i];
if (rp->project != this) continue;
double d = rp->estimated_runtime_remaining();
d /= gstate.time_stats.availability_frac(rp->avp->gpu_usage.rsc_type);
if (rp->is_not_started()) {
not_started_dur += d;
} else {
in_progress_dur += d;
}
}
}
int CLIENT_STATE::write_tasks_gui(MIOFILE& f, bool active_only) {
unsigned int i;
if (active_only) {
for (i=0; i<active_tasks.active_tasks.size(); i++) {
RESULT* rp = active_tasks.active_tasks[i]->result;
rp->write_gui(f);
}
} else {
for (i=0; i<results.size(); i++) {
RESULT* rp = results[i];
rp->write_gui(f);
}
}
return 0;
}
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");
}
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");
}
bool PROJECT::runnable(int rsc_type) {
if (suspended_via_gui) return false;
if (suspended_during_update) return false;
for (unsigned int i=0; i<gstate.results.size(); i++) {
RESULT* rp = gstate.results[i];
if (rp->project != this) continue;
switch (rsc_type) {
case RSC_TYPE_ANY:
break;
case RSC_TYPE_CPU:
if (rp->uses_coprocs()) continue;
break;
case RSC_TYPE_CUDA:
if (rp->avp->ncudas == 0) continue;
break;
case RSC_TYPE_ATI:
if (rp->avp->natis == 0) continue;
break;
}
if (rp->runnable()) return true;
}
return false;
}
// Compute an "overall long-term debt" for each project.
// This is a sum of per-resource terms, scaled by the relative speed of the resource.
// The term for a resource is its LTD plus an estimate of queued work.
//
void WORK_FETCH::set_overall_debts() {
unsigned int i;
PROJECT* p;
RESULT* rp;
APP_VERSION* avp;
for (i=0; i<gstate.projects.size(); i++) {
p = gstate.projects[i];
p->cpu_pwf.queue_est = 0;
p->cuda_pwf.queue_est = 0;
p->ati_pwf.queue_est = 0;
}
for (i=0; i<gstate.results.size(); i++) {
rp = gstate.results[i];
p = rp->project;
if (!rp->nearly_runnable()) continue;
if (p->non_cpu_intensive) continue;
double dt = rp->estimated_time_remaining();
avp = rp->avp;
p->cpu_pwf.queue_est += dt*avp->avg_ncpus;
p->cuda_pwf.queue_est += dt*avp->ncudas;
p->ati_pwf.queue_est += dt*avp->natis;
}
for (i=0; i<gstate.projects.size(); i++) {
p = gstate.projects[i];
double queue_debt = p->cpu_pwf.queue_est/gstate.ncpus;
p->pwf.overall_debt = p->cpu_pwf.long_term_debt - queue_debt;
if (gstate.host_info.have_cuda()) {
p->pwf.overall_debt += cuda_work_fetch.relative_speed*
(p->cuda_pwf.long_term_debt - p->cuda_pwf.queue_est/gstate.host_info.coprocs.cuda.count);
}
if (gstate.host_info.have_ati()) {
p->pwf.overall_debt += ati_work_fetch.relative_speed*
(p->ati_pwf.long_term_debt - p->ati_pwf.queue_est/gstate.host_info.coprocs.ati.count);
}
}
}
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;
}
// Write a scheduler request to a disk file,
// to be sent to a scheduling server
//
int CLIENT_STATE::make_scheduler_request(PROJECT* p) {
char buf[1024];
MIOFILE mf;
unsigned int i;
RESULT* rp;
get_sched_request_filename(*p, buf, sizeof(buf));
FILE* f = boinc_fopen(buf, "wb");
if (!f) return ERR_FOPEN;
double trs = total_resource_share();
double rrs = runnable_resource_share(RSC_TYPE_ANY);
double prrs = potentially_runnable_resource_share();
double resource_share_fraction, rrs_fraction, prrs_fraction;
if (trs) {
resource_share_fraction = p->resource_share / trs;
} else {
resource_share_fraction = 1;
}
if (rrs) {
rrs_fraction = p->resource_share / rrs;
} else {
rrs_fraction = 1;
}
if (prrs) {
prrs_fraction = p->resource_share / prrs;
} else {
prrs_fraction = 1;
}
// if hostid is zero, rpc_seqno better be also
//
if (!p->hostid) {
p->rpc_seqno = 0;
}
mf.init_file(f);
fprintf(f,
"<scheduler_request>\n"
" <authenticator>%s</authenticator>\n"
" <hostid>%d</hostid>\n"
" <rpc_seqno>%d</rpc_seqno>\n"
" <core_client_major_version>%d</core_client_major_version>\n"
" <core_client_minor_version>%d</core_client_minor_version>\n"
" <core_client_release>%d</core_client_release>\n"
" <resource_share_fraction>%f</resource_share_fraction>\n"
" <rrs_fraction>%f</rrs_fraction>\n"
" <prrs_fraction>%f</prrs_fraction>\n"
" <duration_correction_factor>%f</duration_correction_factor>\n"
" <allow_multiple_clients>%d</allow_multiple_clients>\n"
" <sandbox>%d</sandbox>\n",
p->authenticator,
p->hostid,
p->rpc_seqno,
core_client_version.major,
core_client_version.minor,
core_client_version.release,
resource_share_fraction,
rrs_fraction,
prrs_fraction,
p->duration_correction_factor,
config.allow_multiple_clients?1:0,
g_use_sandbox?1:0
);
work_fetch.write_request(f, p);
// write client capabilities
//
fprintf(f,
" <client_cap_plan_class>1</client_cap_plan_class>\n"
);
write_platforms(p, mf);
if (strlen(p->code_sign_key)) {
fprintf(f, " <code_sign_key>\n%s\n</code_sign_key>\n", p->code_sign_key);
}
// send working prefs
//
fprintf(f, "<working_global_preferences>\n");
global_prefs.write(mf);
fprintf(f, "</working_global_preferences>\n");
// send master global preferences if present and not host-specific
//
if (!global_prefs.host_specific && boinc_file_exists(GLOBAL_PREFS_FILE_NAME)) {
FILE* fprefs = fopen(GLOBAL_PREFS_FILE_NAME, "r");
if (fprefs) {
copy_stream(fprefs, f);
fclose(fprefs);
}
PROJECT* pp = lookup_project(global_prefs.source_project);
if (pp && strlen(pp->email_hash)) {
fprintf(f,
"<global_prefs_source_email_hash>%s</global_prefs_source_email_hash>\n",
pp->email_hash
);
}
}
// Of the projects with same email hash as this one,
// send the oldest cross-project ID.
// Use project URL as tie-breaker.
//
PROJECT* winner = p;
for (i=0; i<projects.size(); i++ ) {
PROJECT* project = projects[i];
if (project == p) continue;
if (strcmp(project->email_hash, p->email_hash)) continue;
if (project->cpid_time < winner->cpid_time) {
winner = project;
} else if (project->cpid_time == winner->cpid_time) {
if (strcmp(project->master_url, winner->master_url) < 0) {
winner = project;
}
}
}
fprintf(f,
"<cross_project_id>%s</cross_project_id>\n",
winner->cross_project_id
);
time_stats.write(mf, true);
net_stats.write(mf);
if (global_prefs.daily_xfer_period_days) {
daily_xfer_history.write_scheduler_request(
mf, global_prefs.daily_xfer_period_days
);
}
// update hardware info, and write host info
//
host_info.get_host_info();
set_ncpus();
host_info.write(mf, !config.suppress_net_info, false);
// get and write disk usage
//
get_disk_usages();
get_disk_shares();
fprintf(f,
" <disk_usage>\n"
" <d_boinc_used_total>%f</d_boinc_used_total>\n"
" <d_boinc_used_project>%f</d_boinc_used_project>\n"
" <d_project_share>%f</d_project_share>\n"
" </disk_usage>\n",
total_disk_usage, p->disk_usage, p->disk_share
);
// copy request values from RSC_WORK_FETCH to COPROC
//
int j = rsc_index(GPU_TYPE_NVIDIA);
if (j > 0) {
coprocs.nvidia.req_secs = rsc_work_fetch[j].req_secs;
coprocs.nvidia.req_instances = rsc_work_fetch[j].req_instances;
coprocs.nvidia.estimated_delay = rsc_work_fetch[j].req_secs?rsc_work_fetch[j].busy_time_estimator.get_busy_time():0;
}
j = rsc_index(GPU_TYPE_ATI);
if (j > 0) {
coprocs.ati.req_secs = rsc_work_fetch[j].req_secs;
coprocs.ati.req_instances = rsc_work_fetch[j].req_instances;
coprocs.ati.estimated_delay = rsc_work_fetch[j].req_secs?rsc_work_fetch[j].busy_time_estimator.get_busy_time():0;
}
if (coprocs.n_rsc > 1) {
coprocs.write_xml(mf, true);
}
// report completed jobs
//
unsigned int last_reported_index = 0;
p->nresults_returned = 0;
for (i=0; i<results.size(); i++) {
rp = results[i];
if (rp->project == p && rp->ready_to_report) {
p->nresults_returned++;
rp->write(mf, true);
}
if (config.max_tasks_reported
&& (p->nresults_returned >= config.max_tasks_reported)
) {
last_reported_index = i;
break;
}
}
read_trickle_files(p, f);
// report sticky files as needed
//
for (i=0; i<file_infos.size(); i++) {
FILE_INFO* fip = file_infos[i];
if (fip->project != p) continue;
if (!fip->sticky) continue;
fprintf(f,
" <file_info>\n"
" <name>%s</name>\n"
" <nbytes>%f</nbytes>\n"
" <status>%d</status>\n"
" </file_info>\n",
fip->name, fip->nbytes, fip->status
);
}
if (p->send_time_stats_log) {
fprintf(f, "<time_stats_log>\n");
time_stats.get_log_after(p->send_time_stats_log, mf);
fprintf(f, "</time_stats_log>\n");
}
if (p->send_job_log) {
fprintf(f, "<job_log>\n");
job_log_filename(*p, buf, sizeof(buf));
send_log_after(buf, p->send_job_log, mf);
fprintf(f, "</job_log>\n");
}
// send descriptions of app versions
//
fprintf(f, "<app_versions>\n");
j=0;
for (i=0; i<app_versions.size(); i++) {
APP_VERSION* avp = app_versions[i];
if (avp->project != p) continue;
avp->write(mf, false);
avp->index = j++;
}
fprintf(f, "</app_versions>\n");
// send descriptions of jobs in progress for this project
//
fprintf(f, "<other_results>\n");
for (i=0; i<results.size(); i++) {
rp = results[i];
if (rp->project != p) continue;
if ((last_reported_index && (i > last_reported_index)) || !rp->ready_to_report) {
fprintf(f,
" <other_result>\n"
" <name>%s</name>\n"
" <app_version>%d</app_version>\n",
rp->name,
rp->avp->index
);
// the following is for backwards compatibility w/ old schedulers
//
if (strlen(rp->avp->plan_class)) {
fprintf(f,
" <plan_class>%s</plan_class>\n",
rp->avp->plan_class
);
}
fprintf(f,
" </other_result>\n"
);
}
}
fprintf(f, "</other_results>\n");
// if requested by project, send summary of all in-progress results
// (for EDF simulation by scheduler)
//
if (p->send_full_workload) {
fprintf(f, "<in_progress_results>\n");
for (i=0; i<results.size(); i++) {
rp = results[i];
double x = rp->estimated_runtime_remaining();
if (x == 0) continue;
strcpy(buf, "");
int rt = rp->avp->gpu_usage.rsc_type;
if (rt) {
if (rt == rsc_index(GPU_TYPE_NVIDIA)) {
sprintf(buf, " <ncudas>%f</ncudas>\n", rp->avp->gpu_usage.usage);
} else if (rt == rsc_index(GPU_TYPE_ATI)) {
sprintf(buf, " <natis>%f</natis>\n", rp->avp->gpu_usage.usage);
}
}
fprintf(f,
" <ip_result>\n"
" <name>%s</name>\n"
" <report_deadline>%.0f</report_deadline>\n"
" <time_remaining>%.2f</time_remaining>\n"
" <avg_ncpus>%f</avg_ncpus>\n"
"%s"
" </ip_result>\n",
rp->name,
rp->report_deadline,
x,
rp->avp->avg_ncpus,
buf
);
}
fprintf(f, "</in_progress_results>\n");
}
FILE* cof = boinc_fopen(CLIENT_OPAQUE_FILENAME, "r");
if (cof) {
fprintf(f, "<client_opaque>\n<![CDATA[\n");
copy_stream(cof, f);
fprintf(f, "\n]]>\n</client_opaque>\n");
fclose(cof);
}
fprintf(f, "</scheduler_request>\n");
fclose(f);
return 0;
}
// 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;
}
// parse a scheduler reply.
// Some of the items go into the SCHEDULER_REPLY object.
// Others are copied straight to the PROJECT
//
int SCHEDULER_REPLY::parse(FILE* in, PROJECT* project) {
char buf[256], msg_buf[1024], pri_buf[256], attr_buf[256];
int retval;
MIOFILE mf;
XML_PARSER xp(&mf);
std::string delete_file_name;
mf.init_file(in);
bool found_start_tag = false, btemp;
double cpid_time = 0;
clear();
safe_strcpy(host_venue, project->host_venue);
// the project won't send us a venue if it's doing maintenance
// or doesn't check the DB because no work.
// Don't overwrite the host venue in that case.
sr_feeds.clear();
trickle_up_urls.clear();
if (!project->anonymous_platform) {
for (int i=0; i<MAX_RSC; i++) {
project->no_rsc_apps[i] = false;
}
}
// First line should either be tag (HTTP 1.0) or
// hex length of response (HTTP 1.1)
//
while (!xp.get_tag(attr_buf, sizeof(attr_buf))) {
if (!found_start_tag) {
if (xp.match_tag("scheduler_reply")) {
found_start_tag = true;
}
continue;
}
if (xp.match_tag("/scheduler_reply")) {
// update statistics after parsing the scheduler reply
// add new record if vector is empty or we have a new day
//
if (project->statistics.empty() || project->statistics.back().day!=dday()) {
project->trim_statistics();
DAILY_STATS nds;
project->statistics.push_back(nds);
}
DAILY_STATS& ds = project->statistics.back();
ds.day=dday();
ds.user_total_credit=project->user_total_credit;
ds.user_expavg_credit=project->user_expavg_credit;
ds.host_total_credit=project->host_total_credit;
ds.host_expavg_credit=project->host_expavg_credit;
project->write_statistics_file();
if (cpid_time) {
project->cpid_time = cpid_time;
} else {
project->cpid_time = project->user_create_time;
}
if (project->dont_use_dcf) {
project->duration_correction_factor = 1;
}
return 0;
}
else if (xp.parse_str("project_name", project->project_name, sizeof(project->project_name))) {
continue;
}
else if (xp.parse_str("master_url", master_url, sizeof(master_url))) {
continue;
}
else if (xp.parse_str("symstore", project->symstore, sizeof(project->symstore))) continue;
else if (xp.parse_str("user_name", project->user_name, sizeof(project->user_name))) continue;
else if (xp.parse_double("user_total_credit", project->user_total_credit)) continue;
else if (xp.parse_double("user_expavg_credit", project->user_expavg_credit)) continue;
else if (xp.parse_double("user_create_time", project->user_create_time)) continue;
else if (xp.parse_double("cpid_time", cpid_time)) continue;
else if (xp.parse_str("team_name", project->team_name, sizeof(project->team_name))) continue;
else if (xp.parse_int("hostid", hostid)) continue;
else if (xp.parse_double("host_total_credit", project->host_total_credit)) continue;
else if (xp.parse_double("host_expavg_credit", project->host_expavg_credit)) continue;
else if (xp.parse_str("host_venue", host_venue, sizeof(host_venue))) continue;
else if (xp.parse_double("host_create_time", project->host_create_time)) continue;
else if (xp.parse_double("request_delay", request_delay)) continue;
else if (xp.parse_double("next_rpc_delay", next_rpc_delay)) continue;
else if (xp.match_tag("global_preferences")) {
retval = dup_element_contents(
xp.f->f,
"</global_preferences>",
&global_prefs_xml
);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse global prefs in scheduler reply: %s",
boincerror(retval)
);
return retval;
}
} else if (xp.match_tag("project_preferences")) {
retval = dup_element_contents(
xp.f->f,
"</project_preferences>",
&project_prefs_xml
);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse project prefs in scheduler reply: %s",
boincerror(retval)
);
return retval;
}
} else if (xp.match_tag("gui_urls")) {
std::string foo;
retval = copy_element_contents(xp.f->f, "</gui_urls>", foo);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse GUI URLs in scheduler reply: %s",
boincerror(retval)
);
return retval;
}
project->gui_urls = "<gui_urls>\n"+foo+"</gui_urls>\n";
} else if (xp.match_tag("code_sign_key")) {
retval = dup_element_contents(
xp.f->f,
"</code_sign_key>",
&code_sign_key
);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse code sign key in scheduler reply: %s",
boincerror(retval)
);
return ERR_XML_PARSE;
}
strip_whitespace(code_sign_key);
} else if (xp.match_tag("code_sign_key_signature")) {
retval = dup_element_contents(
xp.f->f,
"</code_sign_key_signature>",
&code_sign_key_signature
);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse code sign key signature in scheduler reply: %s",
boincerror(retval)
);
return ERR_XML_PARSE;
}
} else if (xp.match_tag("app")) {
APP app;
retval = app.parse(xp);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse application in scheduler reply: %s",
boincerror(retval)
);
} else {
apps.push_back(app);
}
} else if (xp.match_tag("file_info")) {
FILE_INFO file_info;
retval = file_info.parse(xp);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse file info in scheduler reply: %s",
boincerror(retval)
);
} else {
file_infos.push_back(file_info);
}
} else if (xp.match_tag("app_version")) {
APP_VERSION av;
retval = av.parse(xp);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse application version in scheduler reply: %s",
boincerror(retval)
);
} else {
app_versions.push_back(av);
}
} else if (xp.match_tag("workunit")) {
WORKUNIT wu;
retval = wu.parse(xp);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse workunit in scheduler reply: %s",
boincerror(retval)
);
} else {
workunits.push_back(wu);
}
} else if (xp.match_tag("result")) {
RESULT result; // make sure this is here so constructor
// gets called each time
retval = result.parse_server(xp);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse task in scheduler reply: %s",
boincerror(retval)
);
} else {
results.push_back(result);
}
} else if (xp.match_tag("result_ack")) {
RESULT result;
retval = result.parse_name(xp, "/result_ack");
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse ack in scheduler reply: %s",
boincerror(retval)
);
} else {
result_acks.push_back(result);
}
} else if (xp.match_tag("result_abort")) {
RESULT result;
retval = result.parse_name(xp, "/result_abort");
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse result abort in scheduler reply: %s",
boincerror(retval)
);
} else {
result_abort.push_back(result);
}
} else if (xp.match_tag("result_abort_if_not_started")) {
RESULT result;
retval = result.parse_name(xp, "/result_abort_if_not_started");
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't parse result abort-if-not-started in scheduler reply: %s",
boincerror(retval)
);
} else {
result_abort_if_not_started.push_back(result);
}
} else if (xp.parse_string("delete_file_info", delete_file_name)) {
file_deletes.push_back(delete_file_name);
} else if (xp.parse_str("message", msg_buf, sizeof(msg_buf))) {
parse_attr(attr_buf, "priority", pri_buf, sizeof(pri_buf));
USER_MESSAGE um(msg_buf, pri_buf);
messages.push_back(um);
continue;
} else if (xp.parse_bool("message_ack", message_ack)) {
continue;
} else if (xp.parse_bool("project_is_down", project_is_down)) {
continue;
} else if (xp.parse_str("email_hash", project->email_hash, sizeof(project->email_hash))) {
continue;
} else if (xp.parse_str("cross_project_id", project->cross_project_id, sizeof(project->cross_project_id))) {
continue;
} else if (xp.parse_str("external_cpid", project->external_cpid, sizeof(project->external_cpid))) {
continue;
} else if (xp.match_tag("trickle_down")) {
retval = gstate.handle_trickle_down(project, in);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"handle_trickle_down failed: %s", boincerror(retval)
);
}
continue;
} else if (xp.parse_bool("non_cpu_intensive", project->non_cpu_intensive)) {
continue;
} else if (xp.parse_bool("ended", project->ended)) {
continue;
} else if (xp.parse_bool("no_cpu_apps", btemp)) {
if (!project->anonymous_platform) {
handle_no_rsc_apps("CPU", project, btemp);
}
continue;
// deprecated syntax
} else if (xp.parse_bool("no_cuda_apps", btemp)) {
if (!project->anonymous_platform) {
handle_no_rsc_apps(GPU_TYPE_NVIDIA, project, btemp);
}
continue;
} else if (xp.parse_bool("no_ati_apps", btemp)) {
if (!project->anonymous_platform) {
handle_no_rsc_apps(GPU_TYPE_ATI, project, btemp);
}
continue;
} else if (xp.parse_str("no_rsc_apps", buf, sizeof(buf))) {
if (!project->anonymous_platform) {
handle_no_rsc_apps(buf, project, true);
}
continue;
} else if (xp.parse_bool("verify_files_on_app_start", project->verify_files_on_app_start)) {
continue;
} else if (xp.parse_bool("send_full_workload", send_full_workload)) {
continue;
} else if (xp.parse_bool("dont_use_dcf", dont_use_dcf)) {
continue;
} else if (xp.parse_int("send_time_stats_log", send_time_stats_log)){
continue;
} else if (xp.parse_int("send_job_log", send_job_log)) {
continue;
} else if (xp.parse_int("scheduler_version", scheduler_version)) {
continue;
} else if (xp.match_tag("project_files")) {
retval = parse_project_files(xp, project_files);
#ifdef ENABLE_AUTO_UPDATE
} else if (xp.match_tag("auto_update")) {
retval = auto_update.parse(xp);
if (!retval) auto_update.present = true;
#endif
} else if (xp.match_tag("rss_feeds")) {
got_rss_feeds = true;
parse_rss_feed_descs(xp, sr_feeds);
continue;
} else if (xp.match_tag("trickle_up_urls")) {
parse_trickle_up_urls(xp, trickle_up_urls);
continue;
} else if (xp.parse_int("userid", project->userid)) {
continue;
} else if (xp.parse_int("teamid", project->teamid)) {
continue;
} else if (xp.parse_double("desired_disk_usage", project->desired_disk_usage)) {
continue;
} else {
if (log_flags.unparsed_xml) {
msg_printf(project, MSG_INFO,
"[unparsed_xml] SCHEDULER_REPLY::parse(): unrecognized %s\n",
xp.parsed_tag
);
}
}
}
if (found_start_tag) {
msg_printf(project, MSG_INTERNAL_ERROR, "No close tag in scheduler reply");
} else {
msg_printf(project, MSG_INTERNAL_ERROR, "No start tag in scheduler reply");
}
return ERR_XML_PARSE;
}
// Handle a task that has finished.
// Mark its output files as present, and delete scratch files.
// Don't delete input files because they might be shared with other WUs.
// Update state of result record.
//
int CLIENT_STATE::app_finished(ACTIVE_TASK& at) {
RESULT* rp = at.result;
bool had_error = false;
#ifndef SIM
FILE_INFO* fip;
unsigned int i;
char path[MAXPATHLEN];
int retval;
double size;
// scan the output files, check if missing or too big.
// Don't bother doing this if result was aborted via GUI or by project
//
switch (rp->exit_status) {
case EXIT_ABORTED_VIA_GUI:
case EXIT_ABORTED_BY_PROJECT:
break;
default:
for (i=0; i<rp->output_files.size(); i++) {
FILE_REF& fref = rp->output_files[i];
fip = fref.file_info;
if (fip->uploaded) continue;
get_pathname(fip, path, sizeof(path));
retval = file_size(path, size);
if (retval) {
if (fref.optional) {
fip->upload_urls.clear();
continue;
}
// an output file is unexpectedly absent.
//
fip->status = retval;
had_error = true;
msg_printf(
rp->project, MSG_INFO,
"Output file %s for task %s absent",
fip->name, rp->name
);
} else if (size > fip->max_nbytes) {
// Note: this is only checked when the application finishes.
// The total disk space is checked while the application is running.
//
msg_printf(
rp->project, MSG_INFO,
"Output file %s for task %s exceeds size limit.",
fip->name, rp->name
);
msg_printf(
rp->project, MSG_INFO,
"File size: %f bytes. Limit: %f bytes",
size, fip->max_nbytes
);
fip->delete_file();
fip->status = ERR_FILE_TOO_BIG;
had_error = true;
} else {
if (!fip->uploadable() && !fip->sticky) {
fip->delete_file(); // sets status to NOT_PRESENT
} else {
retval = 0;
if (fip->gzip_when_done) {
retval = fip->gzip();
}
if (!retval) {
retval = md5_file(path, fip->md5_cksum, fip->nbytes);
}
if (retval) {
fip->status = retval;
had_error = true;
} else {
fip->status = FILE_PRESENT;
}
}
}
}
}
#endif
if (rp->exit_status != 0) {
had_error = true;
}
if (had_error) {
switch (rp->exit_status) {
case EXIT_ABORTED_VIA_GUI:
case EXIT_ABORTED_BY_PROJECT:
rp->set_state(RESULT_ABORTED, "CS::app_finished");
break;
default:
rp->set_state(RESULT_COMPUTE_ERROR, "CS::app_finished");
}
rp->project->njobs_error++;
} else {
#ifdef SIM
rp->set_state(RESULT_FILES_UPLOADED, "CS::app_finished");
rp->set_ready_to_report();
rp->completed_time = now;
#else
rp->set_state(RESULT_FILES_UPLOADING, "CS::app_finished");
rp->append_log_record();
#endif
rp->project->update_duration_correction_factor(&at);
rp->project->njobs_success++;
}
double elapsed_time = now - rec_interval_start;
work_fetch.accumulate_inst_sec(&at, elapsed_time);
rp->project->pwf.request_if_idle_and_uploading = true;
// set this to allow work fetch if idle instance,
// even before upload finishes
return 0;
}
int CLIENT_STATE::parse_state_file_aux(const char* fname) {
PROJECT *project=NULL;
int retval=0;
string stemp;
FILE* f = fopen(fname, "r");
if (!f) return ERR_FOPEN;
MIOFILE mf;
XML_PARSER xp(&mf);
mf.init_file(f);
while (!xp.get_tag()) {
if (xp.match_tag("/client_state")) {
break;
}
if (xp.match_tag("client_state")) {
continue;
}
if (xp.match_tag("project")) {
PROJECT temp_project;
retval = temp_project.parse_state(xp);
if (retval) {
msg_printf(NULL, MSG_INTERNAL_ERROR, "Can't parse project in state file");
} else {
#ifdef SIM
project = new PROJECT;
*project = temp_project;
projects.push_back(project);
#else
project = lookup_project(temp_project.master_url);
if (project) {
project->copy_state_fields(temp_project);
} else {
msg_printf(&temp_project, MSG_INTERNAL_ERROR,
"Project %s is in state file but no account file found",
temp_project.get_project_name()
);
}
#endif
}
continue;
}
if (xp.match_tag("app")) {
APP* app = new APP;
retval = app->parse(xp);
if (!project) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Application %s outside project in state file",
app->name
);
delete app;
continue;
}
if (project->anonymous_platform) {
delete app;
continue;
}
if (retval) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Can't parse application in state file"
);
delete app;
continue;
}
retval = link_app(project, app);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't handle application %s in state file",
app->name
);
delete app;
continue;
}
apps.push_back(app);
continue;
}
if (xp.match_tag("file_info") || xp.match_tag("file")) {
FILE_INFO* fip = new FILE_INFO;
retval = fip->parse(xp);
if (!project) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"File info outside project in state file"
);
delete fip;
continue;
}
if (retval) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Can't handle file info in state file"
);
delete fip;
continue;
}
retval = link_file_info(project, fip);
if (project->anonymous_platform && retval == ERR_NOT_UNIQUE) {
delete fip;
continue;
}
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't handle file info %s in state file",
fip->name
);
delete fip;
continue;
}
file_infos.push_back(fip);
#ifndef SIM
// If the file had a failure before,
// don't start another file transfer
//
int failnum;
if (fip->had_failure(failnum)) {
if (fip->pers_file_xfer) {
delete fip->pers_file_xfer;
fip->pers_file_xfer = NULL;
}
}
if (fip->pers_file_xfer) {
retval = fip->pers_file_xfer->init(fip, fip->pers_file_xfer->is_upload);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't initialize file transfer for %s",
fip->name
);
}
retval = pers_file_xfers->insert(fip->pers_file_xfer);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't start persistent file transfer for %s",
fip->name
);
}
}
#endif
continue;
}
if (xp.match_tag("app_version")) {
APP_VERSION* avp = new APP_VERSION;
retval = avp->parse(xp);
if (!project) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Application version outside project in state file"
);
delete avp;
continue;
}
if (project->anonymous_platform) {
delete avp;
continue;
}
if (retval) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Can't parse application version in state file"
);
delete avp;
continue;
}
if (strlen(avp->platform) == 0) {
safe_strcpy(avp->platform, get_primary_platform());
} else {
if (!is_supported_platform(avp->platform)) {
// if it's a platform we haven't heard of,
// must be that the user tried out a 64 bit client
// and then reverted to a 32-bit client.
// Let's not throw away the app version and its WUs
//
#ifndef SIM
msg_printf(project, MSG_INTERNAL_ERROR,
"App version has unsupported platform %s; changing to %s",
avp->platform, get_primary_platform()
);
#endif
safe_strcpy(avp->platform, get_primary_platform());
}
}
if (avp->missing_coproc) {
msg_printf(project, MSG_INFO,
"Application uses missing %s GPU",
avp->missing_coproc_name
);
}
retval = link_app_version(project, avp);
if (retval) {
delete avp;
continue;
}
app_versions.push_back(avp);
continue;
}
if (xp.match_tag("workunit")) {
WORKUNIT* wup = new WORKUNIT;
retval = wup->parse(xp);
if (!project) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Workunit outside project in state file"
);
delete wup;
continue;
}
if (retval) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Can't parse workunit in state file"
);
delete wup;
continue;
}
retval = link_workunit(project, wup);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't handle workunit in state file"
);
delete wup;
continue;
}
workunits.push_back(wup);
continue;
}
if (xp.match_tag("result")) {
RESULT* rp = new RESULT;
retval = rp->parse_state(xp);
if (!project) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Task %s outside project in state file",
rp->name
);
delete rp;
continue;
}
if (retval) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Can't parse task in state file"
);
delete rp;
continue;
}
retval = link_result(project, rp);
if (retval) {
msg_printf(project, MSG_INTERNAL_ERROR,
"Can't link task %s in state file",
rp->name
);
delete rp;
continue;
}
// handle transition from old clients which didn't store result.platform;
// skip for anon platform
if (!project->anonymous_platform) {
if (!strlen(rp->platform) || !is_supported_platform(rp->platform)) {
safe_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 application found for task: %s %d %s; discarding",
rp->platform, rp->version_num, rp->plan_class
);
delete rp;
continue;
}
if (rp->avp->missing_coproc) {
msg_printf(project, MSG_INFO,
"Missing coprocessor for task %s", rp->name
);
rp->coproc_missing = true;
}
rp->wup->version_num = rp->version_num;
results.push_back(rp);
continue;
}
if (xp.match_tag("project_files")) {
if (!project) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Project files outside project in state file"
);
xp.skip_unexpected();
continue;
}
parse_project_files(xp, project->project_files);
project->link_project_files();
continue;
}
if (xp.match_tag("host_info")) {
#ifdef SIM
retval = host_info.parse(xp, false);
coprocs = host_info.coprocs;
coprocs.bound_counts();
#else
retval = host_info.parse(xp, true);
#endif
if (retval) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Can't parse host info in state file"
);
}
continue;
}
if (xp.match_tag("time_stats")) {
retval = time_stats.parse(xp);
if (retval) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Can't parse time stats in state file"
);
}
continue;
}
if (xp.match_tag("net_stats")) {
retval = net_stats.parse(xp);
if (retval) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Can't parse network stats in state file"
);
}
continue;
}
if (xp.match_tag("active_task_set")) {
retval = active_tasks.parse(xp);
if (retval) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Can't parse active tasks in state file"
);
}
continue;
}
if (xp.parse_string("platform_name", statefile_platform_name)) {
continue;
}
if (xp.parse_string("alt_platform", stemp)) {
continue;
}
if (xp.parse_int("user_run_request", retval)) {
cpu_run_mode.set(retval, 0);
continue;
}
if (xp.parse_int("user_run_prev_request", retval)) {
cpu_run_mode.set_prev(retval);
continue;
}
if (xp.parse_int("user_gpu_request", retval)) {
gpu_run_mode.set(retval, 0);
continue;
}
if (xp.parse_int("user_gpu_prev_request", retval)) {
gpu_run_mode.set_prev(retval);
continue;
}
if (xp.parse_int("user_network_request", retval)) {
network_run_mode.set(retval, 0);
continue;
}
if (xp.parse_int("core_client_major_version", old_major_version)) {
continue;
}
if (xp.parse_int("core_client_minor_version", old_minor_version)) {
continue;
}
if (xp.parse_int("core_client_release", old_release)) {
continue;
}
if (xp.parse_str("language", language, sizeof(language))) {
continue;
}
if (xp.match_tag("proxy_info")) {
retval = gui_proxy_info.parse(xp);
if (retval) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"Can't parse proxy info in state file"
);
}
continue;
}
if (xp.parse_str("host_venue", main_host_venue, sizeof(main_host_venue))) {
continue;
}
if (xp.parse_double("new_version_check_time", new_version_check_time)) {
continue;
}
if (xp.parse_double("all_projects_list_check_time", all_projects_list_check_time)) {
continue;
}
if (xp.parse_string("newer_version", newer_version)) {
continue;
}
#ifdef ENABLE_AUTO_UPDATE
if (xp.match_tag("auto_update")) {
if (!project) {
msg_printf(NULL, MSG_INTERNAL_ERROR,
"auto update outside project in state file"
);
xp.skip_unexpected();
continue;
}
if (!auto_update.parse(xp) && !auto_update.validate_and_link(project)) {
auto_update.present = true;
}
continue;
}
#endif
if (log_flags.unparsed_xml) {
msg_printf(0, MSG_INFO,
"[unparsed_xml] state_file: unrecognized: %s",
xp.parsed_tag
);
}
xp.skip_unexpected();
}
sort_results();
fclose(f);
// if total resource share is zero, set all shares to 1
//
if (projects.size()) {
unsigned int i;
double x=0;
for (i=0; i<projects.size(); i++) {
x += projects[i]->resource_share;
}
if (!x) {
msg_printf(NULL, MSG_INFO,
"All projects have zero resource share; setting to 100"
);
for (i=0; i<projects.size(); i++) {
projects[i]->resource_share = 100;
}
}
}
return 0;
}
void do_client_simulation() {
char buf[256], buf2[256];
int retval;
FILE* f;
sprintf(buf, "%s%s", infile_prefix, CONFIG_FILE);
cc_config.defaults();
read_config_file(true, buf);
log_flags.init();
sprintf(buf, "%s%s", outfile_prefix, "log_flags.xml");
f = fopen(buf, "r");
if (f) {
MIOFILE mf;
mf.init_file(f);
XML_PARSER xp(&mf);
xp.get_tag(); // skip open tag
log_flags.parse(xp);
fclose(f);
}
gstate.add_platform("client simulator");
sprintf(buf, "%s%s", infile_prefix, STATE_FILE_NAME);
if (!boinc_file_exists(buf)) {
fprintf(stderr, "No client state file\n");
exit(1);
}
retval = gstate.parse_state_file_aux(buf);
if (retval) {
fprintf(stderr, "state file parse error %d\n", retval);
exit(1);
}
// if tasks have pending transfers, mark as completed
//
for (unsigned int i=0; i<gstate.results.size(); i++) {
RESULT* rp = gstate.results[i];
if (rp->state() < RESULT_FILES_DOWNLOADED) {
rp->set_state(RESULT_FILES_DOWNLOADED, "init");
} else if (rp->state() == RESULT_FILES_UPLOADING) {
rp->set_state(RESULT_FILES_UPLOADED, "init");
}
}
check_app_config(infile_prefix);
show_app_config();
cc_config.show();
log_flags.show();
sprintf(buf, "%s%s", infile_prefix, GLOBAL_PREFS_FILE_NAME);
sprintf(buf2, "%s%s", infile_prefix, GLOBAL_PREFS_OVERRIDE_FILE);
gstate.read_global_prefs(buf, buf2);
fprintf(index_file,
"<h3>Output files</h3>\n"
"<a href=%s>Summary</a>\n"
"<br><a href=%s>Log file</a>\n",
SUMMARY_FNAME, LOG_FNAME
);
// fill in GPU device nums and OpenCL flags
//
for (int i=0; i<coprocs.n_rsc; i++) {
COPROC& cp = coprocs.coprocs[i];
for (int j=0; j<cp.count; j++) {
cp.device_nums[j] = j;
if (cp.have_opencl) {
cp.instance_has_opencl[j] = true;
}
}
}
set_no_rsc_config();
process_gpu_exclusions();
get_app_params();
if (!include_empty_projects) {
cull_projects();
}
fprintf(summary_file, "--------------------------\n");
int j=0;
for (unsigned int i=0; i<gstate.projects.size(); i++) {
gstate.projects[i]->index = j++;
}
clear_backoff();
gstate.log_show_projects();
gstate.set_ncpus();
work_fetch.init();
//set_initial_rec();
rec_adjust_period = delta;
gstate.request_work_fetch("init");
simulate();
sim_results.compute_figures_of_merit();
sprintf(buf, "%s%s", outfile_prefix, RESULTS_DAT_FNAME);
f = fopen(buf, "w");
sim_results.print(f);
fclose(f);
sprintf(buf, "%s%s", outfile_prefix, RESULTS_TXT_FNAME);
f = fopen(buf, "w");
sim_results.print(f, true);
fclose(f);
fprintf(summary_file,
"Simulation done.\n"
"-------------------------\n"
"Figures of merit:\n"
);
sim_results.print(summary_file, true);
double cpu_time;
boinc_calling_thread_cpu_time(cpu_time);
fprintf(summary_file,
"-------------------------\n"
"Simulator CPU time: %f secs\n"
"-------------------------\n"
"Peak FLOPS: CPU %.2fG GPU %.2fG\n",
cpu_time,
cpu_peak_flops()/1e9,
gpu_peak_flops()/1e9
);
print_project_results(summary_file);
fclose(rec_file);
make_graph("REC", "rec", 0);
}
// simulate trying to do an RPC;
// return true if we actually did one
//
bool CLIENT_STATE::simulate_rpc(PROJECT* p) {
char buf[256], buf2[256];
vector<IP_RESULT> ip_results;
vector<RESULT*> new_results;
bool avail;
if (p->last_rpc_time) {
double delta = now - p->last_rpc_time;
avail = p->available.sample(delta);
} else {
avail = p->available.sample(0);
}
p->last_rpc_time = now;
if (!avail) {
sprintf(buf, "RPC to %s skipped - project down<br>", p->project_name);
html_msg += buf;
msg_printf(p, MSG_INFO, "RPC skipped: project down");
gstate.scheduler_op->project_rpc_backoff(p, "project down");
p->master_url_fetch_pending = false;
return false;
}
// save request params for WORK_FETCH::handle_reply
//
double save_cpu_req_secs = rsc_work_fetch[0].req_secs;
for (int i=1; i<coprocs.n_rsc; i++) {
COPROC& cp = coprocs.coprocs[i];
if (!strcmp(cp.type, "NVIDIA")) {
coprocs.nvidia.req_secs = rsc_work_fetch[i].req_secs;
}
if (!strcmp(cp.type, "ATI")) {
coprocs.ati.req_secs = rsc_work_fetch[i].req_secs;
}
if (!strcmp(cp.type, "intel_gpu")) {
coprocs.intel_gpu.req_secs = rsc_work_fetch[i].req_secs;
}
}
if (!server_uses_workload) {
for (int i=0; i<coprocs.n_rsc; i++) {
rsc_work_fetch[i].estimated_delay = rsc_work_fetch[i].busy_time_estimator.get_busy_time();
}
}
for (unsigned int i=0; i<app_versions.size(); i++) {
app_versions[i]->dont_use = false;
}
work_fetch.request_string(buf2, sizeof(buf2));
sprintf(buf, "RPC to %s: %s<br>", p->project_name, buf2);
html_msg += buf;
msg_printf(p, MSG_INFO, "RPC: %s", buf2);
handle_completed_results(p);
if (server_uses_workload) {
get_workload(ip_results);
}
bool sent_something = false;
while (!existing_jobs_only) {
vector<APP*> apps;
get_apps_needing_work(p, apps);
if (apps.empty()) break;
RESULT* rp = new RESULT;
WORKUNIT* wup = new WORKUNIT;
make_job(p, wup, rp, apps);
double et = wup->rsc_fpops_est / rp->avp->flops;
if (server_uses_workload) {
IP_RESULT c(rp->name, rp->report_deadline-now, et);
if (check_candidate(c, ncpus, ip_results)) {
ip_results.push_back(c);
} else {
msg_printf(p, MSG_INFO, "job for %s misses deadline sim\n", rp->app->name);
APP_VERSION* avp = rp->avp;
delete rp;
delete wup;
avp->dont_use = true;
continue;
}
} else {
double est_delay = get_estimated_delay(rp);
if (est_delay + et > wup->app->latency_bound) {
msg_printf(p, MSG_INFO,
"job for %s misses deadline approx: del %f + et %f > %f\n",
rp->app->name,
est_delay, et, wup->app->latency_bound
);
APP_VERSION* avp = rp->avp;
delete rp;
delete wup;
avp->dont_use = true;
continue;
}
}
sent_something = true;
rp->set_state(RESULT_FILES_DOWNLOADED, "simulate_rpc");
results.push_back(rp);
new_results.push_back(rp);
#if 0
sprintf(buf, "got job %s: CPU time %.2f, deadline %s<br>",
rp->name, rp->final_cpu_time, time_to_string(rp->report_deadline)
);
html_msg += buf;
#endif
decrement_request(rp);
}
njobs += (int)new_results.size();
msg_printf(0, MSG_INFO, "Got %lu tasks", new_results.size());
sprintf(buf, "got %lu tasks<br>", new_results.size());
html_msg += buf;
SCHEDULER_REPLY sr;
rsc_work_fetch[0].req_secs = save_cpu_req_secs;
work_fetch.handle_reply(p, &sr, new_results);
p->nrpc_failures = 0;
p->sched_rpc_pending = 0;
//p->min_rpc_time = now + 900;
p->min_rpc_time = now;
if (sent_something) {
request_schedule_cpus("simulate_rpc");
request_work_fetch("simulate_rpc");
}
sim_results.nrpcs++;
return true;
}
// return an error message or NULL
//
const char* SCHEDULER_REQUEST::parse(FILE* fin) {
char buf[256];
RESULT result;
int retval;
strcpy(authenticator, "");
strcpy(platform.name, "");
strcpy(cross_project_id, "");
hostid = 0;
core_client_major_version = 0;
core_client_minor_version = 0;
core_client_release = 0;
rpc_seqno = 0;
work_req_seconds = 0;
cpu_req_secs = 0;
cpu_req_instances = 0;
resource_share_fraction = 1.0;
rrs_fraction = 1.0;
prrs_fraction = 1.0;
cpu_estimated_delay = 0;
strcpy(global_prefs_xml, "");
strcpy(working_global_prefs_xml, "");
strcpy(code_sign_key, "");
memset(&global_prefs, 0, sizeof(global_prefs));
memset(&host, 0, sizeof(host));
have_other_results_list = false;
have_ip_results_list = false;
have_time_stats_log = false;
client_cap_plan_class = false;
sandbox = -1;
coproc_cuda = 0;
coproc_ati = 0;
fgets(buf, sizeof(buf), fin);
if (!match_tag(buf, "<scheduler_request>")) return "no start tag";
while (fgets(buf, sizeof(buf), fin)) {
// If a line is too long, ignore it.
// This can happen e.g. if the client has bad global_prefs.xml
// This won't be necessary if we rewrite this using XML_PARSER
//
if (!strchr(buf, '\n')) {
while (fgets(buf, sizeof(buf), fin)) {
if (strchr(buf, '\n')) break;
}
continue;
}
if (match_tag(buf, "</scheduler_request>")) {
core_client_version = 10000*core_client_major_version + 100*core_client_minor_version + core_client_release;
return NULL;
}
if (parse_str(buf, "<authenticator>", authenticator, sizeof(authenticator))) {
remove_quotes(authenticator);
continue;
}
if (parse_str(buf, "<cross_project_id>", cross_project_id, sizeof(cross_project_id))) continue;
if (parse_int(buf, "<hostid>", hostid)) continue;
if (parse_int(buf, "<rpc_seqno>", rpc_seqno)) continue;
if (parse_str(buf, "<platform_name>", platform.name, sizeof(platform.name))) continue;
if (match_tag(buf, "<alt_platform>")) {
CLIENT_PLATFORM cp;
retval = cp.parse(fin);
if (!retval) {
alt_platforms.push_back(cp);
}
continue;
}
if (match_tag(buf, "<app_versions>")) {
while (fgets(buf, sizeof(buf), fin)) {
if (match_tag(buf, "</app_versions>")) break;
if (match_tag(buf, "<app_version>")) {
CLIENT_APP_VERSION cav;
retval = cav.parse(fin);
if (retval) {
g_reply->insert_message(
"Invalid app version description", "high"
);
} else {
client_app_versions.push_back(cav);
}
}
}
continue;
}
if (parse_int(buf, "<core_client_major_version>", core_client_major_version)) continue;
if (parse_int(buf, "<core_client_minor_version>", core_client_minor_version)) continue;
if (parse_int(buf, "<core_client_release>", core_client_release)) continue;
if (parse_double(buf, "<work_req_seconds>", work_req_seconds)) continue;
if (parse_double(buf, "<cpu_req_secs>", cpu_req_secs)) continue;
if (parse_double(buf, "<cpu_req_instances>", cpu_req_instances)) continue;
if (parse_double(buf, "<resource_share_fraction>", resource_share_fraction)) continue;
if (parse_double(buf, "<rrs_fraction>", rrs_fraction)) continue;
if (parse_double(buf, "<prrs_fraction>", prrs_fraction)) continue;
if (parse_double(buf, "<estimated_delay>", cpu_estimated_delay)) continue;
if (parse_double(buf, "<duration_correction_factor>", host.duration_correction_factor)) continue;
if (match_tag(buf, "<global_preferences>")) {
strcpy(global_prefs_xml, "<global_preferences>\n");
while (fgets(buf, sizeof(buf), fin)) {
if (strstr(buf, "</global_preferences>")) break;
safe_strcat(global_prefs_xml, buf);
}
safe_strcat(global_prefs_xml, "</global_preferences>\n");
continue;
}
if (match_tag(buf, "<working_global_preferences>")) {
while (fgets(buf, sizeof(buf), fin)) {
if (strstr(buf, "</working_global_preferences>")) break;
safe_strcat(working_global_prefs_xml, buf);
}
continue;
}
if (parse_str(buf, "<global_prefs_source_email_hash>", global_prefs_source_email_hash, sizeof(global_prefs_source_email_hash))) continue;
if (match_tag(buf, "<host_info>")) {
host.parse(fin);
continue;
}
if (match_tag(buf, "<time_stats>")) {
host.parse_time_stats(fin);
continue;
}
if (match_tag(buf, "<time_stats_log>")) {
handle_time_stats_log(fin);
have_time_stats_log = true;
continue;
}
if (match_tag(buf, "<net_stats>")) {
host.parse_net_stats(fin);
continue;
}
if (match_tag(buf, "<disk_usage>")) {
host.parse_disk_usage(fin);
continue;
}
if (match_tag(buf, "<result>")) {
result.parse_from_client(fin);
static int max_results = 200;
--max_results;
if (max_results >= 0)
results.push_back(result);
continue;
}
if (match_tag(buf, "<code_sign_key>")) {
copy_element_contents(fin, "</code_sign_key>", code_sign_key, sizeof(code_sign_key));
continue;
}
if (match_tag(buf, "<msg_from_host>")) {
MSG_FROM_HOST_DESC md;
retval = md.parse(fin);
if (!retval) {
msgs_from_host.push_back(md);
}
continue;
}
if (match_tag(buf, "<file_info>")) {
FILE_INFO fi;
retval = fi.parse(fin);
if (!retval) {
file_infos.push_back(fi);
}
continue;
}
if (match_tag(buf, "<host_venue>")) {
continue;
}
if (match_tag(buf, "<other_results>")) {
have_other_results_list = true;
while (fgets(buf, sizeof(buf), fin)) {
if (match_tag(buf, "</other_results>")) break;
if (match_tag(buf, "<other_result>")) {
OTHER_RESULT o_r;
retval = o_r.parse(fin);
if (!retval) {
other_results.push_back(o_r);
}
}
}
continue;
}
if (match_tag(buf, "<in_progress_results>")) {
have_ip_results_list = true;
int i = 0;
double now = time(0);
while (fgets(buf, sizeof(buf), fin)) {
if (match_tag(buf, "</in_progress_results>")) break;
if (match_tag(buf, "<ip_result>")) {
IP_RESULT ir;
retval = ir.parse(fin);
if (!retval) {
if (!strlen(ir.name)) {
sprintf(ir.name, "ip%d", i++);
}
ir.report_deadline -= now;
ip_results.push_back(ir);
}
}
}
continue;
}
if (match_tag(buf, "coprocs")) {
MIOFILE mf;
mf.init_file(fin);
coprocs.parse(mf);
coproc_cuda = (COPROC_CUDA*)coprocs.lookup("CUDA");
coproc_ati = (COPROC_ATI*)coprocs.lookup("ATI");
continue;
}
if (parse_bool(buf, "client_cap_plan_class", client_cap_plan_class)) continue;
if (parse_int(buf, "<sandbox>", sandbox)) continue;
if (match_tag(buf, "<active_task_set>")) continue;
if (match_tag(buf, "<app>")) continue;
if (match_tag(buf, "<app_version>")) continue;
if (match_tag(buf, "<duration_variability>")) continue;
if (match_tag(buf, "<new_version_check_time>")) continue;
if (match_tag(buf, "<newer_version>")) continue;
if (match_tag(buf, "<project>")) continue;
if (match_tag(buf, "<project_files>")) continue;
if (match_tag(buf, "<proxy_info>")) continue;
if (match_tag(buf, "<user_network_request>")) continue;
if (match_tag(buf, "<user_run_request>")) continue;
if (match_tag(buf, "<master_url>")) continue;
if (match_tag(buf, "<project_name>")) continue;
if (match_tag(buf, "<user_name>")) continue;
if (match_tag(buf, "<team_name>")) continue;
if (match_tag(buf, "<email_hash>")) continue;
if (match_tag(buf, "<user_total_credit>")) continue;
if (match_tag(buf, "<user_expavg_credit>")) continue;
if (match_tag(buf, "<user_create_time>")) continue;
if (match_tag(buf, "<host_total_credit>")) continue;
if (match_tag(buf, "<host_expavg_credit>")) continue;
if (match_tag(buf, "<host_create_time>")) continue;
if (match_tag(buf, "<nrpc_failures>")) continue;
if (match_tag(buf, "<master_fetch_failures>")) continue;
if (match_tag(buf, "<min_rpc_time>")) continue;
if (match_tag(buf, "<short_term_debt>")) continue;
if (match_tag(buf, "<long_term_debt>")) continue;
if (match_tag(buf, "<resource_share>")) continue;
if (match_tag(buf, "<scheduler_url>")) continue;
if (match_tag(buf, "</project>")) continue;
if (match_tag(buf, "<?xml")) continue;
strip_whitespace(buf);
if (!strlen(buf)) continue;
log_messages.printf(MSG_NORMAL,
"SCHEDULER_REQUEST::parse(): unrecognized: %s\n", buf
);
MIOFILE mf;
mf.init_file(fin);
retval = skip_unrecognized(buf, mf);
if (retval) return "unterminated unrecognized XML";
}
return "no end tag";
}