// 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;
}
// A result timed out; penalize the corresponding host_app_version
//
static int result_timed_out(
TRANSITIONER_ITEM res_item, TRANSITIONER_ITEM& wu_item
) {
DB_HOST_APP_VERSION hav;
char query[512], clause[512];
int gavid = generalized_app_version_id(
res_item.res_app_version_id, wu_item.appid
);
int retval = hav_lookup(hav, res_item.res_hostid, gavid);
if (retval) {
log_messages.printf(MSG_NORMAL,
"result_timed_out(): hav_lookup failed: %s\n", boincerror(retval)
);
return 0;
}
hav.turnaround.update_var(
(double)wu_item.delay_bound,
HAV_AVG_THRESH, HAV_AVG_WEIGHT, HAV_AVG_LIMIT
);
int n = hav.max_jobs_per_day;
if (n == 0) {
n = config.daily_result_quota;
}
if (n > config.daily_result_quota) {
n = config.daily_result_quota;
}
n -= 1;
if (n < 1) {
n = 1;
}
if (config.debug_quota) {
log_messages.printf(MSG_NORMAL,
"[quota] max_jobs_per_day for %d; %d->%d\n",
gavid, hav.max_jobs_per_day, n
);
}
hav.max_jobs_per_day = n;
hav.consecutive_valid = 0;
sprintf(query,
"turnaround_n=%.15e, turnaround_avg=%.15e, turnaround_var=%.15e, turnaround_q=%.15e, max_jobs_per_day=%d, consecutive_valid=%d",
hav.turnaround.n,
hav.turnaround.avg,
hav.turnaround.var,
hav.turnaround.q,
hav.max_jobs_per_day,
hav.consecutive_valid
);
sprintf(clause,
"host_id=%d and app_version_id=%d",
hav.host_id, hav.app_version_id
);
retval = hav.update_fields_noid(query, clause);
if (retval) {
log_messages.printf(MSG_CRITICAL,
"CRITICAL result_timed_out(): hav updated failed: %s\n",
boincerror(retval)
);
}
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;
}