Rooba
/
boinc
mirror of https://github.com/BOINC/boinc.git
1
0
Fork 0
Code Issues 1 Packages Projects Releases Wiki Activity
boinc/client/cs_scheduler.C

1331 lines
41 KiB
C
Raw Blame History

// Berkeley Open Infrastructure for Network Computing
// http://boinc.berkeley.edu
// Copyright (C) 2005 University of California
//
// This is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation;
// either version 2.1 of the License, or (at your option) any later version.
//
// This software is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
// See the GNU Lesser General Public License for more details.
//
// To view the GNU Lesser General Public License visit
// http://www.gnu.org/copyleft/lesser.html
// or write to the Free Software Foundation, Inc.,
// 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
// High-level logic for communicating with scheduling servers,
// and for merging the result of a scheduler RPC into the client state
// The scheduler RPC mechanism is in scheduler_op.C
#include "cpp.h"
#ifdef _WIN32
#include "boinc_win.h"
#endif
#ifndef _WIN32
#include "config.h"
#include <stdio.h>
#include <math.h>
#include <time.h>
#include <strings.h>
#include <map>
#include <set>
#endif
#include "crypt.h"
#include "error_numbers.h"
#include "file_names.h"
#include "filesys.h"
#include "parse.h"
#include "util.h"
#include "client_msgs.h"
#include "scheduler_op.h"
#include "client_state.h"
using std::max;
using std::vector;
using std::string;
// quantities like avg CPU time decay by a factor of e every week
//
#define EXP_DECAY_RATE (1./(SECONDS_PER_DAY*7))
// try to report results this much before their deadline
//
#define REPORT_DEADLINE_CUSHION ((double)SECONDS_PER_DAY)
static const char* urgency_name(int urgency) {
switch(urgency) {
case WORK_FETCH_DONT_NEED: return "Don't need";
case WORK_FETCH_OK: return "OK";
case WORK_FETCH_NEED: return "Need";
case WORK_FETCH_NEED_IMMEDIATELY: return "Need immediately";
}
return "Unknown";
}
// how many CPUs should this project occupy on average,
// based on its resource share relative to a given set
//
int CLIENT_STATE::proj_min_results(PROJECT* p, double subset_resource_share) {
if (p->non_cpu_intensive) {
return 1;
}
if (!subset_resource_share) return 1; // TODO - fix
return (int)(ceil(ncpus*p->resource_share/subset_resource_share));
}
void CLIENT_STATE::check_project_timeout() {
unsigned int i;
for (i=0; i<projects.size(); i++) {
PROJECT* p = projects[i];
if (p->possibly_backed_off && now > p->min_rpc_time) {
p->possibly_backed_off = false;
request_work_fetch("Project backoff ended");
}
}
}
void PROJECT::set_min_rpc_time(double future_time, const char* reason) {
if (future_time > min_rpc_time) {
min_rpc_time = future_time;
possibly_backed_off = true;
msg_printf(this, MSG_INFO,
"Deferring communication for %s",
timediff_format(min_rpc_time - gstate.now).c_str()
);
msg_printf(this, MSG_INFO, "Reason: %s\n", reason);
}
}
// Return true if we should not contact the project yet.
//
bool PROJECT::waiting_until_min_rpc_time() {
return (min_rpc_time > gstate.now);
}
// find a project that needs to have its master file fetched
//
PROJECT* CLIENT_STATE::next_project_master_pending() {
unsigned int i;
PROJECT* p;
for (i=0; i<projects.size(); i++) {
p = projects[i];
if (p->waiting_until_min_rpc_time()) continue;
if (p->suspended_via_gui) continue;
if (p->master_url_fetch_pending) {
return p;
}
}
return 0;
}
// find a project for which a scheduler RPC is pending
// and we're not backed off
//
PROJECT* CLIENT_STATE::next_project_sched_rpc_pending() {
unsigned int i;
PROJECT* p;
for (i=0; i<projects.size(); i++) {
p = projects[i];
if (p->waiting_until_min_rpc_time()) continue;
if (p->next_rpc_time && p->next_rpc_time<now) {
p->sched_rpc_pending = RPC_REASON_PROJECT_REQ;
p->next_rpc_time = 0;
}
//if (p->suspended_via_gui) continue;
// do the RPC even if suspended.
// This is critical for acct mgrs, to propagate new host CPIDs
//
if (p->sched_rpc_pending) {
return p;
}
}
return 0;
}
PROJECT* CLIENT_STATE::next_project_trickle_up_pending() {
unsigned int i;
PROJECT* p;
for (i=0; i<projects.size(); i++) {
p = projects[i];
if (p->waiting_until_min_rpc_time()) continue;
if (p->suspended_via_gui) continue;
if (p->trickle_up_pending) {
return p;
}
}
return 0;
}
// Return the best project to fetch work from, NULL if none
//
// Pick the one with largest (long term debt - amount of current work)
//
// PRECONDITIONS:
// - work_request_urgency and work_request set for all projects
// - CLIENT_STATE::overall_work_fetch_urgency is set
// (by previous call to compute_work_requests())
//
PROJECT* CLIENT_STATE::next_project_need_work() {
PROJECT *p, *p_prospect = NULL;
unsigned int i;
for (i=0; i<projects.size(); i++) {
p = projects[i];
if (p->work_request_urgency == WORK_FETCH_DONT_NEED) continue;
if (p->work_request == 0) continue;
if (!p->contactable()) continue;
// if we don't need work, only get work from non-cpu intensive projects.
//
if (overall_work_fetch_urgency == WORK_FETCH_DONT_NEED && !p->non_cpu_intensive) continue;
// if we don't really need work,
// and we don't really need work from this project, pass.
//
if (overall_work_fetch_urgency == WORK_FETCH_OK) {
if (p->work_request_urgency <= WORK_FETCH_OK) {
continue;
}
}
if (p_prospect) {
if (p->work_request_urgency == WORK_FETCH_OK &&
p_prospect->work_request_urgency > WORK_FETCH_OK
) {
continue;
}
if (p->long_term_debt + p->cpu_shortfall < p_prospect->long_term_debt + p_prospect->cpu_shortfall
&& !p->non_cpu_intensive
) {
continue;
}
}
p_prospect = p;
}
if (p_prospect && (p_prospect->work_request <= 0)) {
p_prospect->work_request = 1.0;
if (log_flags.work_fetch_debug) {
msg_printf(0, MSG_INFO,
"[work_fetch_debug] next_project_need_work: project picked %s",
p_prospect->project_name
);
}
}
return p_prospect;
}
// 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;
int retval;
double disk_total, disk_project;
get_sched_request_filename(*p, buf);
FILE* f = boinc_fopen(buf, "wb");
double trs = total_resource_share();
double rrs = runnable_resource_share();
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;
}
if (!f) return ERR_FOPEN;
mf.init_file(f);
fprintf(f,
"<scheduler_request>\n"
" <authenticator>%s</authenticator>\n"
" <hostid>%d</hostid>\n"
" <rpc_seqno>%d</rpc_seqno>\n"
" <platform_name>%s</platform_name>\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"
" <work_req_seconds>%f</work_req_seconds>\n"
" <resource_share_fraction>%f</resource_share_fraction>\n"
" <rrs_fraction>%f</rrs_fraction>\n"
" <prrs_fraction>%f</prrs_fraction>\n"
" <estimated_delay>%f</estimated_delay>\n"
" <duration_correction_factor>%f</duration_correction_factor>\n",
p->authenticator,
p->hostid,
p->rpc_seqno,
p->anonymous_platform?"anonymous":platform_name,
core_client_version.major,
core_client_version.minor,
core_client_version.release,
p->work_request,
resource_share_fraction,
rrs_fraction,
prrs_fraction,
time_until_work_done(p, proj_min_results(p, prrs)-1, prrs),
p->duration_correction_factor
);
if (p->anonymous_platform) {
fprintf(f, " <app_versions>\n");
for (i=0; i<app_versions.size(); i++) {
APP_VERSION* avp = app_versions[i];
if (avp->project != p) continue;
avp->write(mf);
}
fprintf(f, " </app_versions>\n");
}
if (strlen(p->code_sign_key)) {
fprintf(f, " <code_sign_key>\n%s</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->user_create_time < winner->user_create_time) {
winner = project;
} else if (project->user_create_time == winner->user_create_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
);
retval = time_stats.write(mf, true);
if (retval) return retval;
retval = net_stats.write(mf);
if (retval) return retval;
// update hardware info, and write host info
//
host_info.get_host_info();
retval = host_info.write(mf);
if (retval) return retval;
// get and write disk usage
//
total_disk_usage(disk_total);
project_disk_usage(p, disk_project);
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"
" </disk_usage>\n",
disk_total, disk_project
);
// report results
//
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);
}
}
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->report_on_rpc) continue;
if (fip->marked_for_delete) continue;
fprintf(f,
" <file_info>\n"
" <name>%s</name>\n"
" <nbytes>%f</nbytes>\n"
" <status>%d</status>\n"
" <report_on_rpc/>\n"
" </file_info>\n",
fip->name, fip->nbytes, fip->status
);
}
// send names of results in progress for this project
//
fprintf(f, "<other_results>\n");
for (i=0; i<results.size(); i++) {
rp = results[i];
if (rp->project == p && !rp->ready_to_report) {
fprintf(f,
" <other_result>\n"
" <name>%s</name>\n"
" </other_result>\n",
rp->name
);
}
}
fprintf(f, "</other_results>\n");
// send summary of in-progress results
// to give scheduler info on our CPU commitment
//
fprintf(f, "<in_progress_results>\n");
for (i=0; i<results.size(); i++) {
rp = results[i];
double x = rp->estimated_cpu_time_remaining();
if (x == 0) continue;
fprintf(f,
" <ip_result>\n"
" <report_deadline>%f</report_deadline>\n"
" <cpu_time_remaining>%f</cpu_time_remaining>\n"
" </ip_result>\n",
rp->report_deadline,
x
);
}
fprintf(f, "</in_progress_results>\n");
fprintf(f, "</scheduler_request>\n");
fclose(f);
return 0;
}
// find a project with finished results that should be reported.
// This means:
// - we're not backing off contacting the project
// - the result is ready_to_report (compute done; files uploaded)
// - we're either within a day of the report deadline,
// or at least work_buf_min_days time has elapsed since
// result was completed,
// or we have a sporadic connection
//
PROJECT* CLIENT_STATE::find_project_with_overdue_results() {
unsigned int i;
RESULT* r;
for (i=0; i<results.size(); i++) {
r = results[i];
// return the project for this result to report if:
//
PROJECT* p = r->project;
if (p->waiting_until_min_rpc_time()) continue;
if (p->suspended_via_gui) continue;
if (!r->ready_to_report) continue;
if (net_status.have_sporadic_connection) {
return p;
}
double cushion = std::max(REPORT_DEADLINE_CUSHION, work_buf_min());
if (gstate.now > r->report_deadline - cushion) {
return p;
}
if (gstate.now > r->completed_time + work_buf_min()) {
return p;
}
if (gstate.now > r->report_deadline - work_buf_min()) {
return p; // Handles the case where the report is due before the next reconnect is
// likely.
}
}
return 0;
}
// the fraction of time a given CPU is working for BOINC
//
double CLIENT_STATE::overall_cpu_frac() {
double running_frac = time_stats.on_frac * time_stats.active_frac * time_stats.cpu_efficiency;
if (running_frac < 0.01) running_frac = 0.01;
if (running_frac > 1) running_frac = 1;
return running_frac;
}
// the expected number of CPU seconds completed by the client
// in a second of wall-clock time.
// May be > 1 on a multiprocessor.
//
double CLIENT_STATE::avg_proc_rate() {
return ncpus*overall_cpu_frac();
}
// estimate wall-clock time until the number of uncompleted results
// for project p will reach k,
// given the total resource share of a set of competing projects
//
double CLIENT_STATE::time_until_work_done(
PROJECT *p, int k, double subset_resource_share
) {
int num_results_to_skip = k;
double est = 0;
// total up the estimated time for this project's unstarted
// and partially completed results,
// omitting the last k
//
for (vector<RESULT*>::reverse_iterator iter = results.rbegin();
iter != results.rend(); iter++
) {
RESULT *rp = *iter;
if (rp->project != p
|| rp->state > RESULT_FILES_DOWNLOADED
|| rp->ready_to_report
) continue;
if (num_results_to_skip > 0) {
--num_results_to_skip;
continue;
}
if (rp->project->non_cpu_intensive) {
// if it is a non_cpu intensive project,
// it needs only one at a time.
//
est = max(rp->estimated_cpu_time_remaining(), work_buf_min());
} else {
est += rp->estimated_cpu_time_remaining();
}
}
if (log_flags.work_fetch_debug) {
msg_printf(NULL, MSG_INFO,
"[work_fetch_debug] time_until_work_done(): est %f ssr %f apr %f prs %f",
est, subset_resource_share, avg_proc_rate(), p->resource_share
);
}
if (subset_resource_share) {
double apr = avg_proc_rate()*p->resource_share/subset_resource_share;
return est/apr;
} else {
return est/avg_proc_rate(); // TODO - fix
}
}
// Top-level function for work fetch policy.
// Outputs:
// - overall_work_fetch_urgency
// - for each contactable project:
// - work_request and work_request_urgency
//
// Notes:
// - at most 1 CPU-intensive project will have a nonzero work_request
// and a work_request_urgency higher than DONT_NEED.
// This prevents projects with low LTD from getting work
// even though there was a higher LTD project that should get work.
// - all non-CPU-intensive projects that need work
// and are contactable will have a work request of 1.
//
// return false
//
bool CLIENT_STATE::compute_work_requests() {
unsigned int i;
static double last_time = 0;
if (gstate.now - last_time >= 60) {
gstate.request_work_fetch("timer");
}
if (!must_check_work_fetch) return 0;
if (log_flags.work_fetch_debug) {
msg_printf(0, MSG_INFO, "[work_fetch_debug] compute_work_requests(): start");
}
last_time = gstate.now;
must_check_work_fetch = false;
adjust_debts();
rr_simulation();
// compute per-project and overall urgency
//
bool possible_deadline_miss = false;
bool project_shortfall = false;
bool non_cpu_intensive_needs_work = false;
for (i=0; i< projects.size(); i++) {
PROJECT* p = projects[i];
if (p->non_cpu_intensive) {
if (p->runnable() || !p->contactable()) {
p->work_request = 0;
p->work_request_urgency = WORK_FETCH_DONT_NEED;
} else {
p->work_request = 1.0;
p->work_request_urgency = WORK_FETCH_NEED_IMMEDIATELY;
non_cpu_intensive_needs_work = true;
if (log_flags.work_fetch_debug) {
msg_printf(p, MSG_INFO,
"[work_fetch_debug] non-CPU-intensive project needs work"
);
}
return false;
}
} else {
p->work_request_urgency = WORK_FETCH_DONT_NEED;
p->work_request = 0;
if (p->rr_sim_deadlines_missed) {
possible_deadline_miss = true;
}
if (p->cpu_shortfall && p->long_term_debt > -global_prefs.cpu_scheduling_period_minutes * 60) {
project_shortfall = true;
}
}
}
if (cpu_shortfall <= 0.0 && (possible_deadline_miss || !project_shortfall)) {
overall_work_fetch_urgency = WORK_FETCH_DONT_NEED;
} else if (no_work_for_a_cpu()) {
overall_work_fetch_urgency = WORK_FETCH_NEED_IMMEDIATELY;
} else if (cpu_shortfall > 0) {
overall_work_fetch_urgency = WORK_FETCH_NEED;
} else {
overall_work_fetch_urgency = WORK_FETCH_OK;
}
if (log_flags.work_fetch_debug) {
msg_printf(0, MSG_INFO,
"[work_fetch_debug] compute_work_requests(): cpu_shortfall %f, overall urgency %s",
cpu_shortfall, urgency_name(overall_work_fetch_urgency)
);
}
if (overall_work_fetch_urgency == WORK_FETCH_DONT_NEED) {
if (non_cpu_intensive_needs_work) {
overall_work_fetch_urgency = WORK_FETCH_NEED_IMMEDIATELY;
}
return false;
}
// loop over projects, and pick one to get work from
//
double prrs = potentially_runnable_resource_share();
PROJECT *pbest = NULL;
for (i=0; i<projects.size(); i++) {
PROJECT *p = projects[i];
// see if this project can be ruled out completely
//
if (p->non_cpu_intensive) continue;
if (!p->contactable()) {
if (log_flags.work_fetch_debug) {
msg_printf(p, MSG_INFO, "[work_fetch_debug] work fetch: project not contactable");
}
continue;
}
if (p->deadlines_missed
&& overall_work_fetch_urgency != WORK_FETCH_NEED_IMMEDIATELY
) {
if (log_flags.work_fetch_debug) {
msg_printf(p, MSG_INFO,
"[work_fetch_debug] project has %d deadline misses",
p->deadlines_missed
);
}
continue;
}
if (p->some_download_stalled()) {
if (log_flags.work_fetch_debug) {
msg_printf(p, MSG_INFO,
"[work_fetch_debug] project has stalled download"
);
}
continue;
}
if (p->some_result_suspended()) {
if (log_flags.work_fetch_debug) {
msg_printf(p, MSG_INFO, "[work_fetch_debug] project has suspended result");
}
continue;
}
if (p->overworked() && overall_work_fetch_urgency < WORK_FETCH_NEED) {
if (log_flags.work_fetch_debug) {
msg_printf(p, MSG_INFO, "[work_fetch_debug] project is overworked");
}
continue;
}
if (p->cpu_shortfall == 0.0 && overall_work_fetch_urgency < WORK_FETCH_NEED) {
if (log_flags.work_fetch_debug) {
msg_printf(p, MSG_INFO, "[work_fetch_debug] project has no shortfall");
}
continue;
}
// see if this project is better than our current best
//
if (pbest) {
// avoid getting work from a project in deadline trouble
//
if (p->deadlines_missed && !pbest->deadlines_missed) {
if (log_flags.work_fetch_debug) {
msg_printf(p, MSG_INFO,
"[work_fetch_debug] project has deadline misses, %s doesn't",
pbest->get_project_name()
);
}
continue;
}
// avoid getting work from an overworked project
//
if (p->overworked() && !pbest->overworked()) {
if (log_flags.work_fetch_debug) {
msg_printf(p, MSG_INFO,
"[work_fetch_debug] project is overworked, %s isn't",
pbest->get_project_name()
);
}
continue;
}
// get work from project with highest LTD
//
if (pbest->long_term_debt + pbest->cpu_shortfall > p->long_term_debt + p->cpu_shortfall) {
if (log_flags.work_fetch_debug) {
msg_printf(p, MSG_INFO,
"[work_fetch_debug] project has less LTD than %s",
pbest->get_project_name()
);
}
continue;
}
}
pbest = p;
if (log_flags.work_fetch_debug) {
msg_printf(pbest, MSG_INFO, "[work_fetch_debug] best project so far");
}
}
if (pbest) {
pbest->work_request = max(
pbest->cpu_shortfall,
cpu_shortfall * (prrs ? pbest->resource_share/prrs : 1)
);
if (!pbest->nearly_runnable()) {
pbest->work_request_urgency = WORK_FETCH_NEED_IMMEDIATELY;
} else if (pbest->cpu_shortfall) {
pbest->work_request_urgency = WORK_FETCH_NEED;
} else {
pbest->work_request_urgency = WORK_FETCH_OK;
}
if (log_flags.work_fetch_debug) {
msg_printf(pbest, MSG_INFO,
"[work_fetch_debug] compute_work_requests(): work req %f, shortfall %f, urgency %s\n",
pbest->work_request, pbest->cpu_shortfall,
urgency_name(pbest->work_request_urgency)
);
}
} else if (non_cpu_intensive_needs_work) {
overall_work_fetch_urgency = WORK_FETCH_NEED_IMMEDIATELY;
}
return false;
}
// called from the client's polling loop.
// initiate scheduler RPC activity if needed and possible
//
bool CLIENT_STATE::scheduler_rpc_poll() {
PROJECT *p;
bool action=false;
static double last_time=0;
// check only every 5 sec, unless there's a tentative (new) project
//
if (!have_tentative_project() && gstate.now - last_time < 5.0) return false;
last_time = gstate.now;
switch(scheduler_op->state) {
case SCHEDULER_OP_STATE_IDLE:
if (scheduler_op->check_master_fetch_start()) {
action = true;
break;
}
p = next_project_sched_rpc_pending();
if (p) {
scheduler_op->init_op_project(p, p->sched_rpc_pending);
action = true;
break;
}
if (network_suspended) break;
p = next_project_trickle_up_pending();
if (p) {
scheduler_op->init_op_project(p, RPC_REASON_TRICKLE_UP);
action = true;
break;
}
// report overdue results
//
p = find_project_with_overdue_results();
if (p) {
scheduler_op->init_op_project(p, RPC_REASON_RESULTS_DUE);
action = true;
break;
}
if (!(exit_when_idle && contacted_sched_server)) {
scheduler_op->init_get_work();
if (scheduler_op->state != SCHEDULER_OP_STATE_IDLE) {
break;
}
}
break;
default:
scheduler_op->poll();
if (scheduler_op->state == SCHEDULER_OP_STATE_IDLE) {
action = true;
}
break;
}
return action;
}
// Handle the reply from a scheduler
//
int CLIENT_STATE::handle_scheduler_reply(
PROJECT* project, char* scheduler_url, int& nresults
) {
SCHEDULER_REPLY sr;
FILE* f;
int retval;
unsigned int i;
bool signature_valid, update_global_prefs=false, update_project_prefs=false;
char buf[256], filename[256];
std::string old_gui_urls = project->gui_urls;
PROJECT* p2;
nresults = 0;
contacted_sched_server = true;
project->last_rpc_time = now;
get_sched_reply_filename(*project, 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 (sr.scheduler_version) {
msg_printf(project, MSG_INFO,
"Scheduler RPC succeeded [server version %d]",
sr.scheduler_version
);
} else {
msg_printf(project, MSG_INFO, "Scheduler RPC succeeded");
}
}
// check that master URL is correct
//
if (strlen(sr.master_url)) {
canonicalize_master_url(sr.master_url);
if (strcmp(sr.master_url, project->master_url)) {
msg_printf(project, MSG_ERROR,
"You used the wrong URL for this project"
);
msg_printf(project, MSG_ERROR,
"The correct URL is %s", sr.master_url
);
p2 = gstate.lookup_project(sr.master_url);
if (p2) {
msg_printf(project, MSG_INFO,
"You seem to be attached to this project twice"
);
msg_printf(project, MSG_INFO,
"We suggest that you detach projects named %s,",
project->project_name
);
msg_printf(project, MSG_INFO,
"then reattach to %s", sr.master_url
);
} else {
msg_printf(project, MSG_INFO,
"Using the wrong URL can cause problems in some cases."
);
msg_printf(project, MSG_INFO,
"When convenient, detach this project, then reattach to %s",
sr.master_url
);
}
}
}
// make sure we don't already have a project of same name
//
if (project->tentative) {
bool dup_name = false;
for (i=0; i<gstate.projects.size(); i++) {
p2 = gstate.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_ERROR,
"Already attached to a project named %s (possibly with wrong URL)",
project->project_name
);
msg_printf(project, MSG_ERROR,
"Consider detaching this project, then trying again"
);
}
}
// on the off chance that this is the initial RPC for a project
// being attached, copy messages to a safe place
//
for (i=0; i<sr.messages.size(); i++) {
USER_MESSAGE& um = sr.messages[i];
sprintf(buf, "Message from server: %s", um.message.c_str());
int prio = (!strcmp(um.priority.c_str(), "high"))?MSG_ERROR:MSG_INFO;
show_message(project, buf, prio);
gstate.project_attach.messages.push_back(um.message);
}
if (log_flags.sched_op_debug && sr.request_delay) {
msg_printf(project, MSG_INFO,
"Project requested delay of %f 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 = gstate.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 host venue: %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_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 = verify_string2(
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_ERROR,
"New code signing key doesn't validate"
);
}
} else {
msg_printf(project, MSG_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) {
app = new APP;
*app = sr.apps[i];
retval = link_app(project, app);
if (retval) {
msg_printf(project, MSG_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_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) {
msg_printf(project, MSG_INFO,
"Got server request to delete file %s", fip->name
);
fip->marked_for_delete = true;
}
}
for (i=0; i<sr.app_versions.size(); i++) {
APP* app = lookup_app(project, sr.app_versions[i].app_name);
APP_VERSION* avp = lookup_app_version(app, sr.app_versions[i].version_num);
if (avp) {
// if we had download failures, clear them
//
avp->clear_errors();
continue;
}
avp = new APP_VERSION;
*avp = sr.app_versions[i];
retval = link_app_version(project, avp);
if (retval) {
msg_printf(project, MSG_ERROR,
"Can't handle application version %s %d in scheduler reply",
avp->app_name, avp->version_num
);
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;
int vnum = choose_version_num(wup, sr);
if (vnum < 0) {
msg_printf(project, MSG_ERROR,
"Can't find application version for task %s", wup->name
);
delete wup;
continue;
}
wup->version_num = vnum;
retval = link_workunit(project, wup);
if (retval) {
msg_printf(project, MSG_ERROR,
"Can't handle task %s in scheduler reply", wup->name
);
delete wup;
continue;
}
wup->clear_errors();
workunits.push_back(wup);
}
for (i=0; i<sr.results.size(); i++) {
if (lookup_result(project, sr.results[i].name)) {
msg_printf(project, MSG_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_ERROR,
"Can't handle task %s in scheduler reply", rp->name
);
delete rp;
continue;
}
results.push_back(rp);
rp->state = RESULT_NEW;
nresults++;
}
// update records for ack'ed results
//
for (i=0; i<sr.result_acks.size(); i++) {
if (log_flags.sched_op_debug) {
msg_printf(0, MSG_INFO,
"[sched_op_debug] handle_scheduler_reply(): got ack for result %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_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(ERR_ABORTED_BY_PROJECT, "aborted by project");
} else {
rp->abort_inactive(ERR_ABORTED_BY_PROJECT);
}
}
}
for (i=0; i<sr.result_abort_if_unstarted.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) {
rp->abort_inactive(ERR_ABORTED_BY_PROJECT);
}
}
}
// remove acked trickle files
//
if (sr.message_ack) {
remove_trickle_files(project);
}
if (sr.send_file_list) {
project->send_file_list = true;
}
project->sched_rpc_pending = 0;
project->trickle_up_pending = false;
// handle delay request
//
if (sr.request_delay) {
double x = gstate.now + sr.request_delay;
project->set_min_rpc_time(x, "requested by project");
} else {
project->min_rpc_time = 0;
}
if (sr.next_rpc_delay) {
project->next_rpc_time = gstate.now + sr.next_rpc_delay;
} else {
project->next_rpc_time = 0;
}
// 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 host CPID: %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;
}
if (sr.auto_update.present) {
if (!sr.auto_update.validate_and_link(project)) {
auto_update = sr.auto_update;
}
}
project->link_project_files(true);
set_client_state_dirty("handle_scheduler_reply");
if (log_flags.state_debug) {
msg_printf(0, MSG_INFO,
"[state_debug] handle_scheduler_reply(): State after handle_scheduler_reply():"
);
print_summary();
}
return 0;
}
double CLIENT_STATE::work_needed_secs() {
double total_work = 0;
for(unsigned int i=0; i<results.size(); i++) {
if (results[i]->project->non_cpu_intensive) continue;
total_work += results[i]->estimated_cpu_time_remaining();
}
double x = work_buf_min() * avg_proc_rate() - total_work;
if (x < 0) {
return 0;
}
return x;
}
void CLIENT_STATE::scale_duration_correction_factors(double factor) {
if (factor <= 0) return;
for (unsigned int i=0; i<projects.size(); i++) {
PROJECT* p = projects[i];
p->duration_correction_factor *= factor;
}
if (log_flags.cpu_sched_debug) {
msg_printf(NULL, MSG_INFO, "[cpu_sched_debug] scaling duration correction factors by %f", factor);
}
}
// Choose a new host CPID.
// If using account manager, do scheduler RPCs
// to all acct-mgr-attached projects to propagate the CPID
//
void CLIENT_STATE::generate_new_host_cpid() {
host_info.generate_host_cpid();
for (unsigned int i=0; i<projects.size(); i++) {
if (projects[i]->attached_via_acct_mgr) {
projects[i]->sched_rpc_pending = RPC_REASON_ACCT_MGR_REQ;
projects[i]->set_min_rpc_time(now + 15, "Sending new host CPID");
}
}
}
const char *BOINC_RCSID_d35a4a7711 = "$Id$";
Reference in New Issue View Git Blame Copy Permalink