int writeRemoteErrorEvent()
{
RemoteErrorEvent remoteerror;
remoteerror.setExecuteHost("<128.105.165.12:32779>");
remoteerror.setDaemonName("<write job log test>");
remoteerror.setErrorText("this is the write test error string");
remoteerror.setCriticalError(true);
if ( !logFile.writeEvent(&remoteerror) ) {
printf("Complain about bad remoteerror write\n");
exit(1);
}
return(0);
}
bool
LocalUserLog::logStarterError( const char* err_msg, bool critical )
{
if( ! is_initialized ) {
// This can happen if we hit an error talking to the shadow
// before we get the job ad. Just ignore it.
return false;
}
if( ! should_log ) {
return true;
}
RemoteErrorEvent event;
event.setErrorText( err_msg );
event.setDaemonName( "starter" );
event.setExecuteHost( daemonCore->InfoCommandSinfulString() );
event.setCriticalError( critical );
if( !u_log.writeEvent(&event) ) {
dprintf( D_ALWAYS, "Unable to log ULOG_REMOTE_ERROR event\n" );
return false;
}
return true;
}
int
pseudo_ulog( ClassAd *ad )
{
ULogEvent *event = instantiateEvent(ad);
int result = 0;
char const *critical_error = NULL;
MyString CriticalErrorBuf;
bool event_already_logged = false;
bool put_job_on_hold = false;
char const *hold_reason = NULL;
char *hold_reason_buf = NULL;
int hold_reason_code = 0;
int hold_reason_sub_code = 0;
if(!event) {
MyString add_str;
sPrintAd(add_str, *ad);
dprintf(
D_ALWAYS,
"invalid event ClassAd in pseudo_ulog: %s\n",
add_str.Value());
return -1;
}
if(ad->LookupInteger(ATTR_HOLD_REASON_CODE,hold_reason_code)) {
put_job_on_hold = true;
ad->LookupInteger(ATTR_HOLD_REASON_SUBCODE,hold_reason_sub_code);
ad->LookupString(ATTR_HOLD_REASON,&hold_reason_buf);
if(hold_reason_buf) {
hold_reason = hold_reason_buf;
}
}
if( event->eventNumber == ULOG_REMOTE_ERROR ) {
RemoteErrorEvent *err = (RemoteErrorEvent *)event;
if(!err->getExecuteHost() || !*err->getExecuteHost()) {
//Insert remote host information.
char *execute_host = NULL;
thisRemoteResource->getMachineName(execute_host);
err->setExecuteHost(execute_host);
delete[] execute_host;
}
if(err->isCriticalError()) {
CriticalErrorBuf.formatstr(
"Error from %s: %s",
err->getExecuteHost(),
err->getErrorText());
critical_error = CriticalErrorBuf.Value();
if(!hold_reason) {
hold_reason = critical_error;
}
//Temporary: the following causes critical remote errors
//to be logged as ShadowExceptionEvents, rather than
//RemoteErrorEvents. The result is ugly, but guaranteed to
//be compatible with other user-log reading tools.
BaseShadow::log_except(critical_error);
event_already_logged = true;
}
}
if( !event_already_logged && !Shadow->uLog.writeEvent( event, ad ) ) {
MyString add_str;
sPrintAd(add_str, *ad);
dprintf(
D_ALWAYS,
"unable to log event in pseudo_ulog: %s\n",
add_str.Value());
result = -1;
}
if(put_job_on_hold) {
hold_reason = critical_error;
if(!hold_reason) {
hold_reason = "Job put on hold by remote host.";
}
Shadow->holdJobAndExit(hold_reason,hold_reason_code,hold_reason_sub_code);
//should never get here, because holdJobAndExit() exits.
}
if( critical_error ) {
//Suppress ugly "Shadow exception!"
Shadow->exception_already_logged = true;
//lame: at the time of this writing, EXCEPT does not want const:
EXCEPT("%s", critical_error);
}
delete event;
return result;
}