/**
* @brief Handle a PSP_ACCOUNT_CHILD message.
*
* This message is sent when a new child is started.
*
* @param msg The message to handle.
*
* @return No return value.
*/
static void handleAccountChild(DDTypedBufferMsg_t *msg)
{
Job_t *job;
Client_t *client;
PStask_ID_t logger;
uid_t uid;
gid_t gid;
int32_t rank;
size_t used = 0;
/* logger's task ID */
PSP_getTypedMsgBuf(msg, &used, __func__, "logger", &logger, sizeof(logger));
/* get job information */
job = findJobByLogger(logger);
if (!job) job = addJob(logger);
PSP_getTypedMsgBuf(msg, &used, __func__, "rank", &rank, sizeof(rank));
PSP_getTypedMsgBuf(msg, &used, __func__, "uid", &uid, sizeof(uid));
PSP_getTypedMsgBuf(msg, &used, __func__, "gid", &gid, sizeof(gid));
client = addClient(msg->header.sender, ACC_CHILD_PSIDCHILD);
bool triggerMonitor = !job->latestChildStart;
job->latestChildStart = time(NULL);
if (triggerMonitor) triggerJobStartMonitor();
if (!findHist(logger)) saveHist(logger);
job->nrOfChilds++;
client->logger = logger;
client->uid = uid;
client->gid = gid;
client->job = job;
client->rank = rank;
}
void setHist(const char* name, const int lc=0, const int ls=0, const int lw=0, const int msz=0, const int mst=0, const double norm = 1, const char* xtitle = "", const char* ytitle = "")
{
TH1F * h;
if (h=dynamic_cast<TH1F*>(findHist(name))) setHist(h,lc,ls,lw,msz,mst,norm,xtitle,ytitle);
}
/**
* @brief Handle a PSP_ACCOUNT_END message.
*
* This function will add extended accounting information to a
* account end message.
*
* @param msg The message to handle.
*
* @return No return value.
*/
static void handleAccountEnd(DDTypedBufferMsg_t *msg)
{
PStask_ID_t sender = msg->header.sender, logger, childTID;
PSnodes_ID_t childNode;
Client_t *client;
Job_t *job;
pid_t child;
uint64_t avgRss, avgVsize, avgThrds, dummy;
size_t used = 0;
mdbg(PSACC_LOG_ACC_MSG, "%s(%s)\n", __func__, PSC_printTID(sender));
PSP_getTypedMsgBuf(msg, &used, __func__, "logger", &logger, sizeof(logger));
/* end msg from logger */
if (sender == logger) {
/* find the job */
job = findJobByLogger(logger);
if (!job) {
mlog("%s: job for logger %s not found\n", __func__,
PSC_printTID(logger));
} else {
job->endTime = time(NULL);
job->complete = true;
if (job->childsExit < job->nrOfChilds) {
mdbg(PSACC_LOG_VERBOSE, "%s: logger %s exited, but %i"
" children are still alive\n", __func__,
PSC_printTID(logger), job->nrOfChilds - job->childsExit);
}
}
return;
}
PSP_getTypedMsgBuf(msg, &used, __func__, "rank(skipped)", &dummy,
sizeof(int32_t));
PSP_getTypedMsgBuf(msg, &used, __func__, "uid(skipped)", &dummy,
sizeof(uid_t));
PSP_getTypedMsgBuf(msg, &used, __func__, "gid(skipped)", &dummy,
sizeof(gid_t));
PSP_getTypedMsgBuf(msg, &used, __func__, "pid", &child, sizeof(child));
/* calculate childs TaskID */
childNode = PSC_getID(sender);
childTID = PSC_getTID(childNode, child);
/* find the child exiting */
client = findClientByTID(childTID);
if (!client) {
if (!findHist(logger)) {
mlog("%s: end msg for unknown client %s from %s\n", __func__,
PSC_printTID(childTID), PSC_printTID(sender));
}
return;
}
if (client->type != ACC_CHILD_JOBSCRIPT && client->logger != logger) {
mlog("%s: logger mismatch (%s/", __func__, PSC_printTID(logger));
mlog("%s)\n", PSC_printTID(client->logger));
}
/* stop accounting of dead child */
client->doAccounting = false;
client->endTime = time(NULL);
PSP_getTypedMsgBuf(msg, &used, __func__, "rusage",
&client->data.rusage, sizeof(client->data.rusage));
PSP_getTypedMsgBuf(msg, &used, __func__, "pageSize",
&client->data.pageSize, sizeof(client->data.pageSize));
PSP_getTypedMsgBuf(msg, &used, __func__, "walltime",
&client->walltime, sizeof(client->walltime));
PSP_getTypedMsgBuf(msg, &used, __func__, "status",
&client->status, sizeof(client->status));
mdbg(PSACC_LOG_VERBOSE, "%s: child rank %i pid %i logger %s uid %i"
" gid %i msg type %s finished\n", __func__, client->rank, child,
PSC_printTID(client->logger), client->uid, client->gid,
getAccountMsgType(msg->type));
if (client->type == ACC_CHILD_JOBSCRIPT) return; /* drop message */
/* Now add further information to the message */
msg->header.len = offsetof(DDTypedBufferMsg_t, buf) + used;
uint32_t one = 1;
PSP_putTypedMsgBuf(msg, __func__, "extended info", &one, sizeof(one));
PSP_putTypedMsgBuf(msg, __func__, "maxRss", &client->data.maxRss,
sizeof(client->data.maxRss));
PSP_putTypedMsgBuf(msg, __func__, "maxVsize", &client->data.maxVsize,
sizeof(client->data.maxVsize));
uint32_t myMaxThreads = client->data.maxThreads;
PSP_putTypedMsgBuf(msg, __func__, "maxThreads", &myMaxThreads,
sizeof(myMaxThreads));
PSP_putTypedMsgBuf(msg, __func__, "session", &client->data.session,
sizeof(client->data.session));
/* add size of average used mem */
if (client->data.avgRssTotal < 1 || client->data.avgRssCount < 1) {
avgRss = 0;
} else {
avgRss = client->data.avgRssTotal / client->data.avgRssCount;
}
PSP_putTypedMsgBuf(msg, __func__, "avgRss", &avgRss, sizeof(avgRss));
/* add size of average used vmem */
if (client->data.avgVsizeTotal < 1 || client->data.avgVsizeCount < 1) {
avgVsize = 0;
} else {
avgVsize = client->data.avgVsizeTotal / client->data.avgVsizeCount;
}
PSP_putTypedMsgBuf(msg, __func__, "avgVsize", &avgVsize, sizeof(avgVsize));
/* add number of average threads */
if (client->data.avgThreadsTotal < 1 || client->data.avgThreadsCount < 1) {
avgThrds = 0;
} else {
avgThrds = client->data.avgThreadsTotal / client->data.avgThreadsCount;
}
PSP_putTypedMsgBuf(msg, __func__, "avgThrds", &avgThrds, sizeof(avgThrds));
/* find the job */
job = findJobByLogger(client->logger);
if (!job) {
mlog("%s: job for child %i not found\n", __func__, child);
} else {
job->childsExit++;
if (job->childsExit >= job->nrOfChilds) {
/* all children exited */
if (globalCollectMode && PSC_getID(logger) != PSC_getMyID()) {
forwardJobData(job, true);
sendAggDataFinish(logger);
}
job->complete = true;
job->endTime = time(NULL);
mdbg(PSACC_LOG_VERBOSE, "%s: job complete [%i:%i]\n", __func__,
job->childsExit, job->nrOfChilds);
if (PSC_getID(job->logger) != PSC_getMyID()) {
deleteJob(job->logger);
}
}
}
}