// do fast checks on this job, i.e. ones that don't require DB access
// if any check fails, return false
//
static bool quick_check(
WU_RESULT& wu_result,
WORKUNIT& wu, // a mutable copy of wu_result.workunit.
// We may modify its delay_bound, rsc_fpops_est, and rsc_fpops_bound
BEST_APP_VERSION* &bavp,
APP* app,
int& last_retval
) {
int retval;
// If we're looking for beta jobs and this isn't one, skip it
//
if (g_wreq->beta_only) {
if (!app->beta) {
if (config.debug_send_job) {
log_messages.printf(MSG_NORMAL,
"[send_job] job is not from beta app; skipping\n"
);
}
return false;
}
if (config.debug_send_job) {
log_messages.printf(MSG_NORMAL,
"[send_job] [HOST#%lu] beta work found: [RESULT#%lu]\n",
g_reply->host.id, wu_result.resultid
);
}
} else {
if (app->beta) {
if (config.debug_send_job) {
log_messages.printf(MSG_NORMAL,
"[send_job] job is from beta app; skipping\n"
);
}
return false;
}
}
// Are we scanning for need_reliable results?
// skip this check the app is beta
// (beta apps don't use the reliable mechanism)
//
if (!app->beta) {
if (g_wreq->reliable_only && (!wu_result.need_reliable)) {
if (config.debug_send_job) {
log_messages.printf(MSG_NORMAL,
"[send_job] job doesn't need reliable host; skipping\n"
);
}
return false;
} else if (!g_wreq->reliable_only && wu_result.need_reliable) {
if (config.debug_send_job) {
log_messages.printf(MSG_NORMAL,
"[send_job] job needs reliable host; skipping\n"
);
}
return false;
}
}
// don't send if we are looking for infeasible results
// and the result is not infeasible
//
if (g_wreq->infeasible_only && (wu_result.infeasible_count==0)) {
if (config.debug_send_job) {
log_messages.printf(MSG_NORMAL,
"[send_job] job is not infeasible; skipping\n"
);
}
return false;
}
// locality sched lite check.
// Allow non-LSL jobs; otherwise we could starve them
// NOTE: THIS NEGATES THE OTHER SCHED POLICIES (reliable, etc.).
// Need to think of some way of combining them.
//
if (g_wreq->locality_sched_lite) {
// skip this job if host has sticky files
// but none of them is used by this job.
// TODO: it should really be "host has sticky files for this app".
// However, we don't have a way of making that association.
// Could add something based on filename
//
if (app->locality_scheduling == LOCALITY_SCHED_LITE
&& g_request->file_infos.size()
) {
int n = nfiles_on_host(wu_result.workunit);
if (config.debug_locality_lite) {
log_messages.printf(MSG_NORMAL,
"[loc_lite] job %s has %d files on this host\n",
wu_result.workunit.name, n
);
}
if (n == 0) {
return false;
}
}
}
// Find the best app_version for this host.
//
bavp = get_app_version(wu, true, g_wreq->reliable_only);
if (!bavp) {
if (config.debug_send_job) {
log_messages.printf(MSG_NORMAL,
"[send_job] No app version for job; skipping\n"
);
}
return false;
}
// Check app filter if needed.
// Do this AFTER get_app_version(), otherwise we could send
// a misleading message to user
//
if (g_wreq->user_apps_only &&
(!g_wreq->beta_only || config.distinct_beta_apps)
) {
if (app_not_selected(app->id)) {
g_wreq->no_allowed_apps_available = true;
if (config.debug_send_job) {
log_messages.printf(MSG_NORMAL,
"[send_job] [USER#%lu] [WU#%lu] user doesn't want work for app %s\n",
g_reply->user.id, wu.id, app->name
);
}
return false;
}
}
// Check whether we can send this job.
// This may modify wu.delay_bound and wu.rsc_fpops_est
//
retval = wu_is_infeasible_fast(
wu,
wu_result.res_server_state, wu_result.res_priority,
wu_result.res_report_deadline,
*app, *bavp
);
if (retval) {
if (retval != last_retval && config.debug_send_job) {
log_messages.printf(MSG_NORMAL,
"[send_job] [HOST#%lu] [WU#%lu %s] WU is infeasible: %s\n",
g_reply->host.id, wu.id, wu.name, infeasible_string(retval)
);
}
last_retval = retval;
if (config.debug_send_job) {
log_messages.printf(MSG_NORMAL,
"[send_job] is_infeasible_fast() failed; skipping\n"
);
}
return false;
}
return true;
}
// Assign a score to this job,
// representing the value of sending the job to this host.
// Also do some initial screening,
// and return false if can't send the job to host
//
bool JOB::get_score(WU_RESULT& wu_result) {
score = 0;
if (!app->beta && wu_result.need_reliable) {
if (!bavp->reliable) {
return false;
}
}
if (app->beta) {
if (g_wreq->allow_beta_work) {
score += 1;
} else {
return false;
}
}
if (app_not_selected(wu_result.workunit)) {
if (g_wreq->allow_non_preferred_apps) {
score -= 1;
} else {
return false;
}
}
if (wu_result.infeasible_count) {
score += 1;
}
if (app->locality_scheduling == LOCALITY_SCHED_LITE
&& g_request->file_infos.size()
) {
int n = nfiles_on_host(wu_result.workunit);
if (config.debug_locality_lite) {
log_messages.printf(MSG_NORMAL,
"[loc_lite] job %s has %d files on this host\n",
wu_result.workunit.name, n
);
}
if (n > 0) {
score += 10;
}
}
if (app->n_size_classes > 1) {
double effective_speed = bavp->host_usage.projected_flops * g_reply->host.on_frac * g_reply->host.active_frac;
int target_size = get_size_class(*app, effective_speed);
if (config.debug_send) {
log_messages.printf(MSG_NORMAL,
"[send] size: host %d job %d speed %f\n",
target_size, wu_result.workunit.size_class, effective_speed
);
}
if (target_size == wu_result.workunit.size_class) {
score += 5;
} else if (target_size < wu_result.workunit.size_class) {
score -= 2;
} else {
score -= 1;
}
}
if (config.debug_send) {
log_messages.printf(MSG_NORMAL,
"[send]: job score %f\n", score
);
}
return true;
}
