void HR_INFO::scan_db() {
DB_HOST host;
int retval;
int i, n=0;
double sum=0, sum_sqr=0;
while (1) {
retval = host.enumerate("where expavg_credit>1");
if (retval) break;
if (host.p_fpops > 1e7 && host.p_fpops < 1e13) {
n++;
sum += host.p_fpops;
sum_sqr += host.p_fpops*host.p_fpops;
}
//printf("host %d: %s | %s | %s\n", host.id, host.os_name, host.p_vendor, host.p_model);
for (i=1; i<HR_NTYPES; i++) {
if (hr_unknown_class(host, i)) {
//printf("type %d: unknown\n", i);
continue;
}
int hrc = hr_class(host, i);
//printf("type %d: class %d\n", i, hrc);
if (!hrc) continue;
rac_per_class[i][hrc] += host.expavg_credit;
}
}
if (retval != ERR_DB_NOT_FOUND) {
fprintf(stderr, "host enum: %d", retval);
exit(1);
}
}
int handle_trickle_init(int argc, char** argv) {
int retval;
for (int i=1; i<argc; i++) {
if (!strcmp(argv[i], "--max_runtime")) {
max_runtime = atof(argv[++i]);
} else {
log_messages.printf(MSG_CRITICAL, "unknown arg %s\n", argv[i]);
return ERR_XML_PARSE;
}
}
if (!max_runtime) {
log_messages.printf(MSG_CRITICAL, "missing --max_runtime arg\n");
return ERR_NULL;
}
DB_HOST host;
retval = host.fpops_percentile(50, flops_50_percentile);
if (retval) {
log_messages.printf(MSG_CRITICAL, "fpops_percentile failed: %d\n", retval);
return retval;
}
retval = host.fpops_percentile(95, flops_95_percentile);
if (retval) {
log_messages.printf(MSG_CRITICAL, "fpops_percentile failed: %d\n", retval);
return retval;
}
log_messages.printf(MSG_NORMAL, "default FLOPS: %f\n", flops_50_percentile);
log_messages.printf(MSG_NORMAL, "max FLOPS: %f\n", flops_95_percentile);
log_messages.printf(MSG_NORMAL, "max runtime: %f\n", max_runtime);
return 0;
}
// handle timed-out hosts
//
bool scan_hosts() {
DB_HOST h;
char buf[256];
int retval;
bool found = false;
sprintf(buf, "where cpu_efficiency < %f", dtime());
while (1) {
retval = h.enumerate(buf);
if (retval == ERR_DB_NOT_FOUND) break;
if (retval) {
log_messages.printf(MSG_CRITICAL, "host.enumerate() failed\n");
exit(1);
}
found = true;
retval = handle_host(h);
if (retval) {
log_messages.printf(MSG_CRITICAL, "handle_host() failed: %d\n", retval);
exit(1);
}
retval = h.update_field("cpu_efficiency=1e12");
if (retval) {
log_messages.printf(MSG_CRITICAL, "h.update_field() failed: %d\n", retval);
exit(1);
}
}
return found;
}
// update the DB record to the values in "xhost"
// "initial_host" stores the current DB values;
// update only those fields that have changed
//
static int update_host_record(HOST& initial_host, HOST& xhost, USER& user) {
DB_HOST host;
int retval;
char buf[1024];
host = xhost;
// hash the CPID reported by the host with the user's email address.
// This prevents one user from spoofing another one's host.
//
if (strlen(host.host_cpid)) {
sprintf(buf, "%s%s", host.host_cpid, user.email_addr);
md5_block((const unsigned char*)buf, strlen(buf), host.host_cpid);
}
char* p = getenv("REMOTE_ADDR");
if (p) {
strlcpy(host.external_ip_addr, p, sizeof(host.external_ip_addr));
}
retval = host.update_diff_sched(initial_host);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"host.update() failed: %s\n", boincerror(retval)
);
}
return 0;
}
int main() {
if ( boinc_db.open("predictor", "boinc", NULL, NULL) ) {
printf("Open failed\n");
return 0;
}
DB_WORKUNIT workunit;
char buf[256];
while (!workunit.enumerate()) {
printf("workunit %d wsn %d\n", workunit.id, workunit.workseq_next);
DB_RESULT result;
sprintf(buf, "where workunitid=%d", workunit.id);
if ( !result.enumerate(buf) ) {
DB_HOST host;
sprintf(buf, "where id=%d", result.hostid);
if ( !host.enumerate(buf) ) {
workunit.workseq_next = OS(host) + CPU(host);
if ( workunit.update() ) printf("Update failed!\n");
}
}
}
};
int update_hosts() {
DB_HOST host;
int retval;
char buf[256];
while (1) {
retval = host.enumerate("where expavg_credit>0.1");
if (retval) {
if (retval != ERR_DB_NOT_FOUND) {
log_messages.printf(MSG_CRITICAL, "lost DB conn\n");
exit(1);
}
break;
}
if (host.expavg_time > update_time_cutoff) continue;
update_average(0, 0, CREDIT_HALF_LIFE, host.expavg_credit, host.expavg_time);
sprintf(
buf,"expavg_credit=%f, expavg_time=%f",
host.expavg_credit, host.expavg_time
);
retval = host.update_field(buf);
if (retval) {
log_messages.printf(MSG_CRITICAL, "Can't update host %d\n", host.id);
return retval;
}
}
return 0;
}
int is_invalid(WORKUNIT& wu, RESULT& result) {
char buf[256];
int retval;
DB_HOST host;
retval = host.lookup_id(result.hostid);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%d] lookup of host %d failed %d\n",
result.id, result.hostid, retval
);
return retval;
}
double old_error_rate = host.error_rate;
if (!is_unreplicated(wu)) {
update_error_rate(host, false);
}
sprintf(buf, "error_rate=%f", host.error_rate);
retval = host.update_field(buf);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%d] update of host %d failed %d\n",
result.id, result.hostid, retval
);
return retval;
}
log_messages.printf(MSG_DEBUG,
"[HOST#%d] invalid result; error rate %f->%f\n",
host.id, old_error_rate, host.error_rate
);
return 0;
}
// Here when a result has been validated and its granted_credit has been set.
// Grant credit to host, user and team, and update host error rate.
//
int is_valid(RESULT& result, WORKUNIT& wu) {
DB_HOST host;
DB_CREDITED_JOB credited_job;
int retval;
char buf[256];
retval = host.lookup_id(result.hostid);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%d] lookup of host %d failed %d\n",
result.id, result.hostid, retval
);
return retval;
}
grant_credit(host, result.sent_time, result.cpu_time, result.granted_credit);
double turnaround = result.received_time - result.sent_time;
compute_avg_turnaround(host, turnaround);
double old_error_rate = host.error_rate;
if (!is_unreplicated(wu)) {
update_error_rate(host, true);
}
sprintf(
buf,
"avg_turnaround=%f, error_rate=%f",
host.avg_turnaround, host.error_rate
);
retval = host.update_field(buf);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%d] update of host %d failed %d\n",
result.id, result.hostid, retval
);
}
log_messages.printf(MSG_DEBUG,
"[HOST#%d] error rate %f->%f\n",
host.id, old_error_rate, host.error_rate
);
if (update_credited_job) {
credited_job.userid = host.userid;
credited_job.workunitid = long(wu.opaque);
retval = credited_job.insert();
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%d] Warning: credited_job insert failed (userid: %d workunit: %f err: %d)\n",
result.id, host.userid, wu.opaque, retval
);
} else {
log_messages.printf(MSG_DEBUG,
"[RESULT#%d %s] added credited_job record [WU#%d OPAQUE#%f USER#%d]\n",
result.id, result.name, wu.id, wu.opaque, host.userid
);
}
}
return 0;
}
int handle_trickle(MSG_FROM_HOST& msg) {
double runtime = 0;
MIOFILE mf;
mf.init_buf_read(msg.xml);
XML_PARSER xp(&mf);
while (!xp.get_tag()) {
if (xp.parse_double("runtime", runtime)) break;
log_messages.printf(MSG_NORMAL, "unexpected tag: %s\n", xp.parsed_tag);
}
if (runtime <= 0) {
log_messages.printf(MSG_NORMAL,
"unexpected nonpositive runtime: %f\n", runtime
);
return ERR_XML_PARSE;
}
DB_HOST host;
int retval = host.lookup_id(msg.hostid);
if (retval) return retval;
HOST old_host = host;
double flops_sec = host.p_fpops;
// sanity checks - customize as needed
//
if (runtime > max_runtime) {
log_messages.printf(MSG_NORMAL,
"Reported runtime exceeds bound: %f>%f\n", runtime, max_runtime
);
runtime = max_runtime;
}
if (flops_sec < 0) {
log_messages.printf(MSG_NORMAL,
"host CPU speed %f < 0. Using %f instead\n",
flops_sec, flops_50_percentile
);
flops_sec = flops_50_percentile;
}
if (flops_sec > flops_95_percentile) {
log_messages.printf(MSG_NORMAL,
"host CPU speed %f exceeds %f. Using %f instead\n",
flops_sec, flops_95_percentile, flops_95_percentile
);
flops_sec = flops_95_percentile;
}
double credit = cpu_time_to_credit(runtime, flops_sec);
grant_credit(host, dtime()-86400, credit);
log_messages.printf(MSG_DEBUG,
"granting %f credit to host %lu\n", credit, host.id
);
// update the host's credit fields
//
retval = host.update_diff_validator(old_host);
return 0;
}
// find the user's most recently-created host with given host CPID
//
static bool find_host_by_cpid(DB_USER& user, char* host_cpid, DB_HOST& host) {
char buf[1024], buf2[256];
sprintf(buf, "%s%s", host_cpid, user.email_addr);
md5_block((const unsigned char*)buf, strlen(buf), buf2);
sprintf(buf,
"where userid=%d and host_cpid='%s' order by id desc", user.id, buf2
);
if (!host.enumerate(buf)) {
host.end_enumerate();
return true;
}
return false;
}
int PERF_INFO::get_from_db() {
int retval, n;
DB_HOST host;
host_fpops_mean = 2.2e9;
host_fpops_stddev = .7e9;
host_fpops_50_percentile = 3.3e9;
host_fpops_95_percentile = 3.3e9;
retval = host.count(n);
if (retval) return retval;
if (n < 10) return 0;
retval = host.fpops_mean(host_fpops_mean);
retval = host.fpops_stddev(host_fpops_stddev);
retval = host.fpops_percentile(50, host_fpops_50_percentile);
retval = host.fpops_percentile(95, host_fpops_95_percentile);
return 0;
}
// find the user's most recently-created host with given various characteristics
//
static bool find_host_by_other(DB_USER& user, HOST req_host, DB_HOST& host) {
char buf[2048];
char dn[512], ip[512], os[512], pm[512];
// don't dig through hosts of these users
// prevents flooding the DB with slow queries from users with many hosts
//
for (unsigned int i=0; i < config.dont_search_host_for_userid.size(); i++) {
if (user.id == config.dont_search_host_for_userid[i]) {
return false;
}
}
// Only check if all the fields are populated
//
if (strlen(req_host.domain_name) && strlen(req_host.last_ip_addr) && strlen(req_host.os_name) && strlen(req_host.p_model)) {
safe_strcpy(dn, req_host.domain_name);
escape_string(dn, sizeof(dn));
safe_strcpy(ip, req_host.last_ip_addr);
escape_string(ip, sizeof(ip));
safe_strcpy(os, req_host.os_name);
escape_string(os, sizeof(os));
safe_strcpy(pm, req_host.p_model);
escape_string(pm, sizeof(pm));
sprintf(buf,
"where userid=%d and id>%d and domain_name='%s' and last_ip_addr = '%s' and os_name = '%s' and p_model = '%s'"
" and m_nbytes = %lf order by id desc", user.id, req_host.id, dn, ip, os, pm, req_host.m_nbytes
);
if (!host.enumerate(buf)) {
host.end_enumerate();
return true;
}
}
return false;
}
// Based on the info in the request message,
// look up the host and its user, and make sure the authenticator matches.
// Some special cases:
// 1) If no host ID is supplied, or if RPC seqno mismatch,
// create a new host record
// 2) If the host record specified by g_request->hostid is a "zombie"
// (i.e. it was merged with another host via the web site)
// then follow links to find the proper host
//
// POSTCONDITION:
// If this function returns zero, then:
// - reply.host contains a valid host record (possibly new)
// - reply.user contains a valid user record
// - if user belongs to a team, reply.team contains team record
//
int authenticate_user() {
int retval;
char buf[1024];
DB_HOST host;
DB_USER user;
DB_TEAM team;
if (g_request->hostid) {
retval = host.lookup_id(g_request->hostid);
while (!retval && host.userid==0) {
// if host record is zombie, follow link to new host
//
retval = host.lookup_id(host.rpc_seqno);
if (!retval) {
g_reply->hostid = host.id;
log_messages.printf(MSG_NORMAL,
"[HOST#%d] forwarding to new host ID %d\n",
g_request->hostid, host.id
);
}
}
if (retval) {
g_reply->insert_message("Can't find host record", "low");
log_messages.printf(MSG_NORMAL,
"[HOST#%d?] can't find host\n",
g_request->hostid
);
g_request->hostid = 0;
goto lookup_user_and_make_new_host;
}
g_reply->host = host;
// look up user based on the ID in host record,
// and see if the authenticator matches (regular or weak)
//
g_request->using_weak_auth = false;
sprintf(buf, "where id=%d", host.userid);
retval = user.lookup(buf);
if (!retval && !strcmp(user.authenticator, g_request->authenticator)) {
// req auth matches user auth - go on
} else {
if (!retval) {
// user for host.userid exists - check weak auth
//
get_weak_auth(user, buf);
if (!strcmp(buf, g_request->authenticator)) {
g_request->using_weak_auth = true;
log_messages.printf(MSG_DEBUG,
"[HOST#%d] accepting weak authenticator\n",
host.id
);
}
}
if (!g_request->using_weak_auth) {
// weak auth failed - look up user based on authenticator
//
strlcpy(
user.authenticator, g_request->authenticator, sizeof(user.authenticator)
);
escape_string(user.authenticator, sizeof(user.authenticator));
sprintf(buf, "where authenticator='%s'", user.authenticator);
retval = user.lookup(buf);
if (retval) {
g_reply->insert_message(
_("Invalid or missing account key. To fix, remove and add this project."),
"notice"
);
g_reply->set_delay(DELAY_MISSING_KEY);
g_reply->nucleus_only = true;
log_messages.printf(MSG_CRITICAL,
"[HOST#%d] [USER#%d] Bad authenticator '%s'\n",
host.id, user.id, g_request->authenticator
);
return ERR_AUTHENTICATOR;
}
}
}
g_reply->user = user;
if (host.userid != user.id) {
// If the request's host ID isn't consistent with the authenticator,
// create a new host record.
//
log_messages.printf(MSG_NORMAL,
"[HOST#%d] [USER#%d] inconsistent host ID; creating new host\n",
host.id, user.id
);
goto make_new_host;
}
// If the seqno from the host is less than what we expect,
// the user must have copied the state file to a different host.
// Make a new host record.
//
if (!batch && g_request->rpc_seqno < g_reply->host.rpc_seqno) {
g_request->hostid = 0;
log_messages.printf(MSG_NORMAL,
"[HOST#%d] [USER#%d] RPC seqno %d less than expected %d; creating new host\n",
g_reply->host.id, user.id, g_request->rpc_seqno, g_reply->host.rpc_seqno
);
goto make_new_host;
}
} else {
// Here no hostid was given, or the ID was bad.
// Look up the user, then create a new host record
//
lookup_user_and_make_new_host:
// if authenticator contains _, it's a weak auth
//
if (strchr(g_request->authenticator, '_')) {
int userid = atoi(g_request->authenticator);
retval = user.lookup_id(userid);
if (!retval) {
get_weak_auth(user, buf);
if (strcmp(buf, g_request->authenticator)) {
retval = ERR_AUTHENTICATOR;
}
}
} else {
strlcpy(
user.authenticator, g_request->authenticator,
sizeof(user.authenticator)
);
escape_string(user.authenticator, sizeof(user.authenticator));
sprintf(buf, "where authenticator='%s'", user.authenticator);
retval = user.lookup(buf);
}
if (retval) {
g_reply->insert_message(
"Invalid or missing account key. To fix, remove and add this project .",
"low"
);
g_reply->set_delay(DELAY_MISSING_KEY);
log_messages.printf(MSG_CRITICAL,
"[HOST#<none>] Bad authenticator '%s': %s\n",
g_request->authenticator, boincerror(retval)
);
return ERR_AUTHENTICATOR;
}
g_reply->user = user;
// If host CPID is present,
// scan backwards through this user's hosts,
// looking for one with the same host CPID.
// If we find one, it means the user detached and reattached.
// Use the existing host record,
// and mark in-progress results as over.
//
if (strlen(g_request->host.host_cpid)) {
if (find_host_by_cpid(user, g_request->host.host_cpid, host)) {
log_messages.printf(MSG_NORMAL,
"[HOST#%d] [USER#%d] No host ID in request, but host with matching CPID found.\n",
host.id, host.userid
);
if ((g_request->allow_multiple_clients != 1)
&& (g_request->other_results.size() == 0)
) {
mark_results_over(host);
}
goto got_host;
}
}
make_new_host:
// One final attempt to locate an existing host record:
// scan backwards through this user's hosts,
// looking for one with the same host name,
// IP address, processor and amount of RAM.
// If found, use the existing host record,
// and mark in-progress results as over.
//
// NOTE: If the client was run with --allow_multiple_clients, skip this.
//
if ((g_request->allow_multiple_clients != 1)
&& find_host_by_other(user, g_request->host, host)
) {
log_messages.printf(MSG_NORMAL,
"[HOST#%d] [USER#%d] Found similar existing host for this user - assigned.\n",
host.id, host.userid
);
mark_results_over(host);
goto got_host;
}
// either of the above cases,
// or host ID didn't match user ID,
// or RPC seqno was too low.
//
// Create a new host.
// g_reply->user is filled in and valid at this point
//
host = g_request->host;
host.id = 0;
host.create_time = time(0);
host.userid = g_reply->user.id;
host.rpc_seqno = 0;
host.expavg_time = time(0);
safe_strcpy(host.venue, g_reply->user.venue);
host.fix_nans();
retval = host.insert();
if (retval) {
g_reply->insert_message(
"Couldn't create host record in database", "low"
);
boinc_db.print_error("host.insert()");
log_messages.printf(MSG_CRITICAL, "host.insert() failed\n");
return retval;
}
host.id = boinc_db.insert_id();
got_host:
g_reply->host = host;
g_reply->hostid = g_reply->host.id;
// this tells client to updates its host ID
g_request->rpc_seqno = 0;
// this value eventually gets written to host DB record;
// for new hosts it must be zero.
// This kludge forces this.
}
// have user record in g_reply->user at this point
//
if (g_reply->user.teamid) {
retval = team.lookup_id(g_reply->user.teamid);
if (!retval) g_reply->team = team;
}
// compute email hash
//
md5_block(
(unsigned char*)g_reply->user.email_addr,
strlen(g_reply->user.email_addr),
g_reply->email_hash
);
// if new user CPID, update user record
//
if (!g_request->using_weak_auth && strlen(g_request->cross_project_id)) {
if (strcmp(g_request->cross_project_id, g_reply->user.cross_project_id)) {
user.id = g_reply->user.id;
escape_string(g_request->cross_project_id, sizeof(g_request->cross_project_id));
sprintf(buf, "cross_project_id='%s'", g_request->cross_project_id);
unescape_string(g_request->cross_project_id, sizeof(g_request->cross_project_id));
user.update_field(buf);
}
}
return 0;
}
// handle a workunit which has new results
//
int handle_wu(
DB_VALIDATOR_ITEM_SET& validator, std::vector<VALIDATOR_ITEM>& items
) {
int canonical_result_index = -1;
bool update_result, retry;
TRANSITION_TIME transition_time = NO_CHANGE;
int retval = 0, canonicalid = 0, x;
double credit = 0;
unsigned int i;
WORKUNIT& wu = items[0].wu;
g_wup = &wu;
vector<RESULT> results;
vector<DB_HOST_APP_VERSION> host_app_versions, host_app_versions_orig;
int nsuccess_results;
// Here if WU doesn't have a canonical result yet.
// Try to get one
log_messages.printf(MSG_NORMAL,
"[WU#%d %s] handle_wu(): No canonical result yet\n",
wu.id, wu.name
);
++log_messages;
// make a vector of the successful results,
// and a parallel vector of host_app_versions
//
for (i=0; i<items.size(); i++) {
RESULT& result = items[i].res;
if ((result.server_state == RESULT_SERVER_STATE_OVER) &&
(result.outcome == RESULT_OUTCOME_SUCCESS)
) {
results.push_back(result);
DB_HOST_APP_VERSION hav;
retval = hav_lookup(hav, result.hostid,
generalized_app_version_id(result.app_version_id, result.appid)
);
if (retval) {
hav.host_id=0; // flag that it's missing
}
host_app_versions.push_back(hav);
host_app_versions_orig.push_back(hav);
}
}
log_messages.printf(MSG_DEBUG,
"[WU#%d %s] Found %d successful results\n",
wu.id, wu.name, (int)results.size()
);
if (results.size() >= (unsigned int)wu.min_quorum) {
log_messages.printf(MSG_DEBUG,
"[WU#%d %s] Enough for quorum, checking set.\n",
wu.id, wu.name
);
double dummy;
retval = check_set(
results, wu, canonicalid, dummy, retry
);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[WU#%d %s] check_set returned %d, exiting\n",
wu.id, wu.name, retval
);
return retval;
}
if (retry) transition_time = DELAYED;
if (credit_from_wu) {
retval = get_credit_from_wu(wu, results, credit);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[WU#%d %s] get_credit_from_wu returned %d\n",
wu.id, wu.name, retval
);
return retval;
}
}
if (canonicalid) {
retval = assign_credit_set(
wu, results, app, app_versions, host_app_versions,
max_granted_credit, credit
);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[WU#%d %s] assign_credit_set() returned %d\n",
wu.id, wu.name, retval
);
transition_time = DELAYED;
goto leave;
}
}
if (max_granted_credit && credit>max_granted_credit) {
credit = max_granted_credit;
}
// scan results.
// update as needed, and count the # of results
// that are still outcome=SUCCESS
// (some may have changed to VALIDATE_ERROR)
//
nsuccess_results = 0;
for (i=0; i<results.size(); i++) {
RESULT& result = results[i];
DB_HOST_APP_VERSION& hav = host_app_versions[i];
DB_HOST_APP_VERSION& hav_orig = host_app_versions_orig[i];
update_result = false;
bool update_host = false;
if (result.outcome == RESULT_OUTCOME_VALIDATE_ERROR) {
transition_time = IMMEDIATE;
update_result = true;
} else {
nsuccess_results++;
}
DB_HOST host;
HOST host_initial;
switch (result.validate_state) {
case VALIDATE_STATE_VALID:
case VALIDATE_STATE_INVALID:
retval = host.lookup_id(result.hostid);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%d] lookup of host %d failed %d\n",
result.id, result.hostid, retval
);
continue;
}
host_initial = host;
}
switch (result.validate_state) {
case VALIDATE_STATE_VALID:
update_result = true;
update_host = true;
retval = is_valid(host, result, wu, host_app_versions[i]);
if (retval) {
log_messages.printf(MSG_DEBUG,
"[RESULT#%d %s] is_valid() failed: %d\n",
result.id, result.name, retval
);
}
grant_credit(
host, result.sent_time, result.cpu_time,
result.granted_credit
);
log_messages.printf(MSG_NORMAL,
"[RESULT#%d %s] Valid; granted %f credit [HOST#%d]\n",
result.id, result.name, result.granted_credit,
result.hostid
);
break;
case VALIDATE_STATE_INVALID:
update_result = true;
update_host = true;
log_messages.printf(MSG_NORMAL,
"[RESULT#%d %s] Invalid [HOST#%d]\n",
result.id, result.name, result.hostid
);
is_invalid(host_app_versions[i]);
break;
case VALIDATE_STATE_INIT:
log_messages.printf(MSG_NORMAL,
"[RESULT#%d %s] Inconclusive [HOST#%d]\n",
result.id, result.name, result.hostid
);
result.validate_state = VALIDATE_STATE_INCONCLUSIVE;
update_result = true;
break;
}
if (hav.host_id) {
retval = hav.update_validator(hav_orig);
}
if (update_host) {
retval = host.update_diff_validator(host_initial);
}
if (update_result) {
retval = validator.update_result(result);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%d %s] result.update() failed: %d\n",
result.id, result.name, retval
);
}
}
}
if (canonicalid) {
// if we found a canonical result,
// trigger the assimilator, but do NOT trigger
// the transitioner - doing so creates a race condition
//
transition_time = NEVER;
log_messages.printf(MSG_DEBUG,
"[WU#%d %s] Found a canonical result: id=%d\n",
wu.id, wu.name, canonicalid
);
wu.canonical_resultid = canonicalid;
wu.canonical_credit = credit;
wu.assimilate_state = ASSIMILATE_READY;
// don't need to send any more results
//
for (i=0; i<items.size(); i++) {
RESULT& result = items[i].res;
if (result.server_state != RESULT_SERVER_STATE_UNSENT) {
continue;
}
result.server_state = RESULT_SERVER_STATE_OVER;
result.outcome = RESULT_OUTCOME_DIDNT_NEED;
retval = validator.update_result(result);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%d %s] result.update() failed: %d\n",
result.id, result.name, retval
);
}
}
} else {
// here if no consensus.
// check if #success results is too large
//
if (nsuccess_results > wu.max_success_results) {
wu.error_mask |= WU_ERROR_TOO_MANY_SUCCESS_RESULTS;
transition_time = IMMEDIATE;
}
// if #success results >= target_nresults,
// we need more results, so bump target_nresults
// NOTE: nsuccess_results should never be > target_nresults,
// but accommodate that if it should happen
//
if (nsuccess_results >= wu.target_nresults) {
wu.target_nresults = nsuccess_results+1;
transition_time = IMMEDIATE;
}
}
}
leave:
--log_messages;
switch (transition_time) {
case IMMEDIATE:
wu.transition_time = time(0);
break;
case DELAYED:
x = time(0) + 6*3600;
if (x < wu.transition_time) wu.transition_time = x;
break;
case NEVER:
wu.transition_time = INT_MAX;
break;
case NO_CHANGE:
break;
}
wu.need_validate = 0;
retval = validator.update_workunit(wu);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[WU#%d %s] update_workunit() failed: %d; exiting\n",
wu.id, wu.name, retval
);
return retval;
}
return 0;
}
// Pick a host to send a chunk of this file to.
// The host must:
// 1) be alive (recent RPC time)
// 2) not have any chunks of this file
//
// We maintain a cache of such hosts
// The policy is:
//
// - scan the cache, removing hosts that are no longer alive;
// return if find a live host
// - pick a random starting point in host ID space,
// and enumerate 100 live hosts; wrap around if needed.
// Return one and put the rest in cache
//
int VDA_FILE_AUX::choose_host() {
int retval;
DB_HOST host;
return 467;
// replenish cache if needed
//
if (!available_hosts.size()) {
int nhosts_scanned = 0;
int rand_id;
for (int i=0; i<2; i++) {
char buf[256];
if (i == 0) {
retval = host.max_id(rand_id, "");
if (retval) {
log_messages.printf(MSG_CRITICAL, "host.max_id() failed\n");
return 0;
}
rand_id = (int)(((double)id)*drand());
sprintf(buf,
"where %s and id>=%d order by id limit 100",
host_alive_clause(), rand_id
);
} else {
sprintf(buf,
"where %s and id<%d order by id limit %d",
host_alive_clause(), rand_id, 100-nhosts_scanned
);
}
while (1) {
retval = host.enumerate(buf);
if (retval == ERR_DB_NOT_FOUND) break;
if (retval) {
log_messages.printf(MSG_CRITICAL, "host enum failed\n");
return 0;
}
nhosts_scanned++;
DB_VDA_CHUNK_HOST ch;
char buf2[256];
int count;
sprintf(buf2, "where vda_file_id=%d and host_id=%d", id, host.id);
#if 0
retval = ch.count(count, buf2);
if (retval) {
log_messages.printf(MSG_CRITICAL, "ch.count failed\n");
return 0;
}
#else
count = 0;
#endif
if (count == 0) {
available_hosts.push_back(host.id);
}
if (nhosts_scanned == 100) break;
}
if (nhosts_scanned == 100) break;
}
}
while (available_hosts.size()) {
int hostid = available_hosts.back();
available_hosts.pop_back();
retval = host.lookup_id(hostid);
if (retval || !alive(host)) {
continue;
}
return hostid;
}
log_messages.printf(MSG_CRITICAL, "No hosts available\n");
return 0;
}
// handle a workunit which has new results
//
int handle_wu(
DB_VALIDATOR_ITEM_SET& validator, std::vector<VALIDATOR_ITEM>& items
) {
int canonical_result_index = -1;
bool update_result, retry;
TRANSITION_TIME transition_time = NO_CHANGE;
int retval = 0, x;
DB_ID_TYPE canonicalid = 0;
double credit = 0;
unsigned int i;
WORKUNIT& wu = items[0].wu;
g_wup = &wu;
if (wu.canonical_resultid) {
log_messages.printf(MSG_NORMAL,
"[WU#%lu %s] Already has canonical result %lu\n",
wu.id, wu.name, wu.canonical_resultid
);
++log_messages;
// Here if WU already has a canonical result.
// Get unchecked results and see if they match the canonical result
//
for (i=0; i<items.size(); i++) {
RESULT& result = items[i].res;
if (result.id == wu.canonical_resultid) {
canonical_result_index = i;
}
}
if (canonical_result_index == -1) {
log_messages.printf(MSG_CRITICAL,
"[WU#%lu %s] Can't find canonical result %lu\n",
wu.id, wu.name, wu.canonical_resultid
);
return 0;
}
RESULT& canonical_result = items[canonical_result_index].res;
// scan this WU's results, and check the unchecked ones
//
for (i=0; i<items.size(); i++) {
RESULT& result = items[i].res;
if (result.server_state != RESULT_SERVER_STATE_OVER) continue;
if (result.outcome != RESULT_OUTCOME_SUCCESS) continue;
switch (result.validate_state) {
case VALIDATE_STATE_INIT:
case VALIDATE_STATE_INCONCLUSIVE:
break;
default:
continue;
}
log_messages.printf(MSG_NORMAL,
"[WU#%lu] handle_wu(): testing result %lu\n",
wu.id, result.id
);
check_pair(result, canonical_result, retry);
if (retry) {
// this usually means an NFS mount has failed;
// arrange to try again later.
//
transition_time = DELAYED;
goto leave;
}
update_result = false;
if (result.outcome == RESULT_OUTCOME_VALIDATE_ERROR) {
update_result = true;
}
// this might be last result, so let transitioner
// trigger file delete etc. if needed
//
transition_time = IMMEDIATE;
DB_HOST host;
retval = host.lookup_id(result.hostid);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%lu] lookup of host %lu failed: %s\n",
result.id, result.hostid, boincerror(retval)
);
continue;
}
HOST host_initial = host;
bool update_hav = false;
DB_HOST_APP_VERSION hav;
retval = hav_lookup(hav, result.hostid,
generalized_app_version_id(result.app_version_id, result.appid)
);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%lu %s] hav_lookup returned %d\n",
result.id, result.name, retval
);
hav.host_id = 0;
}
DB_HOST_APP_VERSION hav_orig = hav;
vector<DB_HOST_APP_VERSION> havv;
havv.push_back(hav);
vector<RESULT> rv;
switch (result.validate_state) {
case VALIDATE_STATE_VALID:
update_result = true;
update_hav = true;
log_messages.printf(MSG_NORMAL,
"[RESULT#%lu %s] pair_check() matched: setting result to valid\n",
result.id, result.name
);
retval = is_valid(host, result, wu, havv[0]);
if (retval) {
log_messages.printf(MSG_NORMAL,
"[RESULT#%lu %s] is_valid() error: %s\n",
result.id, result.name, boincerror(retval)
);
}
// do credit computation, but grant credit of canonical result
//
rv.push_back(result);
assign_credit_set(
wu, rv, app, app_versions, havv,
max_granted_credit, credit
);
if (!no_credit) {
result.granted_credit = canonical_result.granted_credit;
grant_credit(host, result.sent_time, result.granted_credit);
if (config.credit_by_app) {
grant_credit_by_app(result, result.granted_credit);
}
}
break;
case VALIDATE_STATE_INVALID:
update_result = true;
update_hav = true;
log_messages.printf(MSG_NORMAL,
"[RESULT#%lu %s] pair_check() didn't match: setting result to invalid\n",
result.id, result.name
);
is_invalid(havv[0]);
}
if (hav.host_id && update_hav) {
if (dry_run) {
log_messages.printf(MSG_NORMAL, "DB not updated (dry run)\n");
} else {
log_messages.printf(MSG_NORMAL,
"[HOST#%lu AV#%lu] [outlier=%d] Updating HAV in DB. pfc.n=%f->%f\n",
havv[0].host_id, havv[0].app_version_id,
result.runtime_outlier, hav_orig.pfc.n, havv[0].pfc.n
);
retval=havv[0].update_validator(hav_orig);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[HOST#%lu AV%lu] hav.update_validator() failed: %s\n",
hav.host_id, hav.app_version_id, boincerror(retval)
);
}
}
}
host.update_diff_validator(host_initial);
if (update_result) {
log_messages.printf(MSG_NORMAL,
"[RESULT#%lu %s] granted_credit %f\n",
result.id, result.name, result.granted_credit
);
if (dry_run) {
log_messages.printf(MSG_NORMAL, "DB not updated (dry run)\n");
} else {
retval = validator.update_result(result);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%lu %s] Can't update result: %s\n",
result.id, result.name, boincerror(retval)
);
}
}
}
}
} else {
// Here if WU doesn't have a canonical result yet.
// Try to get one
vector<RESULT> viable_results;
vector<DB_HOST_APP_VERSION> host_app_versions, host_app_versions_orig;
log_messages.printf(MSG_NORMAL,
"[WU#%lu %s] handle_wu(): No canonical result yet\n",
wu.id, wu.name
);
++log_messages;
// make a vector of the "viable" (i.e. possibly canonical) results,
// and a parallel vector of host_app_versions
//
for (i=0; i<items.size(); i++) {
RESULT& result = items[i].res;
if (result.server_state != RESULT_SERVER_STATE_OVER) continue;
if (result.outcome != RESULT_OUTCOME_SUCCESS) continue;
if (result.validate_state == VALIDATE_STATE_INVALID) continue;
viable_results.push_back(result);
DB_HOST_APP_VERSION hav;
retval = hav_lookup(hav, result.hostid,
generalized_app_version_id(result.app_version_id, result.appid)
);
if (retval) {
hav.host_id=0; // flag that it's missing
}
host_app_versions.push_back(hav);
host_app_versions_orig.push_back(hav);
}
log_messages.printf(MSG_DEBUG,
"[WU#%lu %s] Found %d viable results\n",
wu.id, wu.name, (int)viable_results.size()
);
if (viable_results.size() >= (unsigned int)wu.min_quorum) {
log_messages.printf(MSG_DEBUG,
"[WU#%lu %s] Enough for quorum, checking set.\n",
wu.id, wu.name
);
double dummy;
retval = check_set(viable_results, wu, canonicalid, dummy, retry);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[WU#%lu %s] check_set() error: %s\n",
wu.id, wu.name, boincerror(retval)
);
return retval;
}
if (retry) transition_time = DELAYED;
// if we found a canonical instance, decide on credit
//
if (canonicalid) {
// always do the credit calculation, to update statistics,
// even if we're granting credit a different way
//
retval = assign_credit_set(
wu, viable_results, app, app_versions, host_app_versions,
max_granted_credit, credit
);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[WU#%lu %s] assign_credit_set(): %s\n",
wu.id, wu.name, boincerror(retval)
);
transition_time = DELAYED;
goto leave;
}
if (credit_from_wu) {
retval = get_credit_from_wu(wu, viable_results, credit);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[WU#%lu %s] get_credit_from_wu(): credit not specified in WU\n",
wu.id, wu.name
);
credit = 0;
}
} else if (credit_from_runtime) {
credit = 0;
for (i=0; i<viable_results.size(); i++) {
RESULT& result = viable_results[i];
if (result.id == canonicalid) {
DB_HOST host;
retval = host.lookup_id(result.hostid);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[WU#%lu %s] host %lu lookup failed\n",
wu.id, wu.name, result.hostid
);
break;
}
double runtime = result.elapsed_time;
if (runtime <=0 || runtime > max_runtime) {
runtime = max_runtime;
}
credit = result.flops_estimate * runtime * COBBLESTONE_SCALE;
log_messages.printf(MSG_NORMAL,
"[WU#%lu][RESULT#%lu] credit_from_runtime %.2f = %.0fs * %.2fGFLOPS\n",
wu.id, result.id,
credit, runtime, result.flops_estimate/1e9
);
break;
}
}
} else if (no_credit) {
credit = 0;
}
if (max_granted_credit && credit>max_granted_credit) {
credit = max_granted_credit;
}
}
// scan the viable results.
// update as needed,
// and count the # of results that are still viable
// (some may now have outcome VALIDATE_ERROR,
// or validate_state INVALID)
//
int n_viable_results = 0;
for (i=0; i<viable_results.size(); i++) {
RESULT& result = viable_results[i];
DB_HOST_APP_VERSION& hav = host_app_versions[i];
DB_HOST_APP_VERSION& hav_orig = host_app_versions_orig[i];
update_result = false;
bool update_host = false;
if (result.outcome != RESULT_OUTCOME_SUCCESS
|| result.validate_state == VALIDATE_STATE_INVALID
) {
transition_time = IMMEDIATE;
update_result = true;
} else {
n_viable_results++;
}
DB_HOST host;
HOST host_initial;
switch (result.validate_state) {
case VALIDATE_STATE_VALID:
case VALIDATE_STATE_INVALID:
retval = host.lookup_id(result.hostid);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%lu] lookup of host %lu: %s\n",
result.id, result.hostid, boincerror(retval)
);
continue;
}
host_initial = host;
}
switch (result.validate_state) {
case VALIDATE_STATE_VALID:
update_result = true;
update_host = true;
retval = is_valid(host, result, wu, host_app_versions[i]);
if (retval) {
log_messages.printf(MSG_DEBUG,
"[RESULT#%lu %s] is_valid() failed: %s\n",
result.id, result.name, boincerror(retval)
);
}
if (!no_credit) {
result.granted_credit = credit;
grant_credit(host, result.sent_time, credit);
log_messages.printf(MSG_NORMAL,
"[RESULT#%lu %s] Valid; granted %f credit [HOST#%lu]\n",
result.id, result.name, result.granted_credit,
result.hostid
);
if (config.credit_by_app) {
grant_credit_by_app(result, credit);
}
}
break;
case VALIDATE_STATE_INVALID:
update_result = true;
update_host = true;
log_messages.printf(MSG_NORMAL,
"[RESULT#%lu %s] Invalid [HOST#%lu]\n",
result.id, result.name, result.hostid
);
is_invalid(host_app_versions[i]);
break;
case VALIDATE_STATE_INIT:
log_messages.printf(MSG_NORMAL,
"[RESULT#%lu %s] Inconclusive [HOST#%lu]\n",
result.id, result.name, result.hostid
);
result.validate_state = VALIDATE_STATE_INCONCLUSIVE;
update_result = true;
break;
}
if (dry_run) {
log_messages.printf(MSG_NORMAL, "DB not updated (dry run)\n");
} else {
if (hav.host_id) {
log_messages.printf(MSG_NORMAL,
"[HOST#%lu AV#%lu] [outlier=%d] Updating HAV in DB. pfc.n=%f->%f\n",
hav.host_id, hav.app_version_id,
result.runtime_outlier, hav_orig.pfc.n, hav.pfc.n
);
retval = hav.update_validator(hav_orig);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[HOST#%lu AV%lu] hav.update_validator() failed: %s\n",
hav.host_id, hav.app_version_id, boincerror(retval)
);
}
}
if (update_host) {
retval = host.update_diff_validator(host_initial);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[HOST#%lu] host.update_diff_validator() failed: %s\n",
host.id, boincerror(retval)
);
}
}
if (update_result) {
retval = validator.update_result(result);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%lu %s] result.update() failed: %s\n",
result.id, result.name, boincerror(retval)
);
}
}
}
}
if (canonicalid) {
// if we found a canonical result,
// trigger the assimilator, but do NOT trigger
// the transitioner - doing so creates a race condition
//
transition_time = NEVER;
log_messages.printf(MSG_DEBUG,
"[WU#%lu %s] Found a canonical result: id=%lu\n",
wu.id, wu.name, canonicalid
);
wu.canonical_resultid = canonicalid;
wu.canonical_credit = credit;
wu.assimilate_state = ASSIMILATE_READY;
// don't need to send any more results
//
for (i=0; i<items.size(); i++) {
RESULT& result = items[i].res;
if (result.server_state != RESULT_SERVER_STATE_UNSENT) {
continue;
}
result.server_state = RESULT_SERVER_STATE_OVER;
result.outcome = RESULT_OUTCOME_DIDNT_NEED;
if (dry_run) {
log_messages.printf(MSG_NORMAL, "DB not updated (dry run)\n");
} else {
retval = validator.update_result(result);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[RESULT#%lu %s] result.update() failed: %s\n",
result.id, result.name, boincerror(retval)
);
}
}
}
} else {
// here if no consensus.
// check if #viable results is too large
//
if (n_viable_results > wu.max_success_results) {
wu.error_mask |= WU_ERROR_TOO_MANY_SUCCESS_RESULTS;
transition_time = IMMEDIATE;
}
// if #viable results >= target_nresults,
// we need more results, so bump target_nresults
// NOTE: n_viable_results should never be > target_nresults,
// but accommodate that if it should happen
//
if (n_viable_results >= wu.target_nresults) {
wu.target_nresults = n_viable_results+1;
transition_time = IMMEDIATE;
}
}
}
}
leave:
--log_messages;
switch (transition_time) {
case IMMEDIATE:
wu.transition_time = time(0);
break;
case DELAYED:
x = time(0) + 6*3600;
if (x < wu.transition_time) wu.transition_time = x;
break;
case NEVER:
wu.transition_time = INT_MAX;
break;
case NO_CHANGE:
break;
}
wu.need_validate = 0;
if (dry_run) {
log_messages.printf(MSG_NORMAL, "DB not updated (dry run)\n");
} else {
retval = validator.update_workunit(wu);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"[WU#%lu %s] update_workunit() failed: %s\n",
wu.id, wu.name, boincerror(retval)
);
return retval;
}
}
return 0;
}
