void HandleMPIDisconnect( int procID, const char *pReason )
{
int nLiveWorkers = VMPI_GetCurrentNumberOfConnections() - g_nDisconnects - 1;
// We ran into the size limit before and it wasn't readily apparent that the size limit had
// been breached, so make sure to show errors about invalid packet sizes..
bool bOldSuppress = g_bSuppressPrintfOutput;
g_bSuppressPrintfOutput = ( Q_stristr( pReason, "invalid packet size" ) == 0 );
Warning( "\n\n--- WARNING: lost connection to '%s' (%s).\n", VMPI_GetMachineName( procID ), pReason );
if ( g_bMPIMaster )
{
Warning( "%d workers remain.\n\n", nLiveWorkers );
++g_nDisconnects;
/*
if ( VMPI_GetCurrentNumberOfConnections() - g_nDisconnects <= 1 )
{
Error( "All machines disconnected!" );
}
*/
}
else
{
VMPI_HandleAutoRestart();
Error( "Worker quitting." );
}
g_bSuppressPrintfOutput = bOldSuppress;
}
//--------------------------------------------------
// UnSerialize face data
//
void UnSerializeFace( MessageBuffer * pmb, int facenum, int iSource )
{
int i, n;
dface_t * f = &g_pFaces[facenum];
facelight_t * fl = &facelight[facenum];
if (pmb->read(f, sizeof(dface_t)) < 0)
Error("UnSerializeFace - invalid dface_t from %s (mb len: %d, offset: %d)", VMPI_GetMachineName( iSource ), pmb->getLen(), pmb->getOffset() );
if (pmb->read(fl, sizeof(facelight_t)) < 0)
Error("UnSerializeFace - invalid facelight_t from %s (mb len: %d, offset: %d)", VMPI_GetMachineName( iSource ), pmb->getLen(), pmb->getOffset() );
fl->sample = (sample_t *) calloc(fl->numsamples, sizeof(sample_t));
if (pmb->read(fl->sample, sizeof(sample_t) * fl->numsamples) < 0)
Error("UnSerializeFace - invalid sample_t from %s (mb len: %d, offset: %d, fl->numsamples: %d)", VMPI_GetMachineName( iSource ), pmb->getLen(), pmb->getOffset(), fl->numsamples );
//
// Read the light information
//
for (i=0; i<MAXLIGHTMAPS; ++i) {
for (n=0; n<NUM_BUMP_VECTS+1; ++n) {
if (fl->light[i][n])
{
fl->light[i][n] = (LightingValue_t *) calloc( fl->numsamples, sizeof(LightingValue_t ) );
if ( ReadValues( pmb, fl->light[i][n], fl->numsamples) < 0)
Error("UnSerializeFace - invalid fl->light from %s (mb len: %d, offset: %d)", VMPI_GetMachineName( iSource ), pmb->getLen(), pmb->getOffset() );
}
}
}
if (fl->luxel) {
fl->luxel = (Vector *) calloc(fl->numluxels, sizeof(Vector));
if (ReadValues( pmb, fl->luxel, fl->numluxels) < 0)
Error("UnSerializeFace - invalid fl->luxel from %s (mb len: %d, offset: %d)", VMPI_GetMachineName( iSource ), pmb->getLen(), pmb->getOffset() );
}
if (fl->luxelNormals) {
fl->luxelNormals = (Vector *) calloc(fl->numluxels, sizeof( Vector ));
if ( ReadValues( pmb, fl->luxelNormals, fl->numluxels) < 0)
Error("UnSerializeFace - invalid fl->luxelNormals from %s (mb len: %d, offset: %d)", VMPI_GetMachineName( iSource ), pmb->getLen(), pmb->getOffset() );
}
}
void ShowMPIStats(
double flTimeSpent,
unsigned long nBytesSent,
unsigned long nBytesReceived,
unsigned long nMessagesSent,
unsigned long nMessagesReceived )
{
double flKSent = (nBytesSent + 511) / 1024;
double flKRecv = (nBytesReceived + 511) / 1024;
bool bShowOutput = VMPI_IsParamUsed( mpi_ShowDistributeWorkStats );
bool bOldSuppress = g_bSuppressPrintfOutput;
g_bSuppressPrintfOutput = !bShowOutput;
Msg( "\n\n--------------------------------------------------------------\n");
Msg( "Total Time : %.2f\n", flTimeSpent );
Msg( "Total Bytes Sent : %dk (%.2fk/sec, %d messages)\n", (int)flKSent, flKSent / flTimeSpent, nMessagesSent );
Msg( "Total Bytes Recv : %dk (%.2fk/sec, %d messages)\n", (int)flKRecv, flKRecv / flTimeSpent, nMessagesReceived );
if ( g_bMPIMaster )
{
Msg( "Duplicated WUs : %I64u (%.1f%%)\n", g_nDuplicatedWUs, (float)g_nDuplicatedWUs * 100.0f / g_nWUs );
Msg( "\nWU count by proc:\n" );
int nProcs = VMPI_GetCurrentNumberOfConnections();
CUtlVector<int> sortedProcs;
sortedProcs.SetSize( nProcs );
for ( int i=0; i < nProcs; i++ )
sortedProcs[i] = i;
qsort( sortedProcs.Base(), nProcs, sizeof( int ), SortByWUCount );
for ( int i=0; i < nProcs; i++ )
{
const char *pMachineName = VMPI_GetMachineName( sortedProcs[i] );
Msg( "%s", pMachineName );
char formatStr[512];
Q_snprintf( formatStr, sizeof( formatStr ), "%%%ds %I64u\n", 30 - strlen( pMachineName ), g_wuCountByProcess[ sortedProcs[i] ] );
Msg( formatStr, ":" );
}
}
Msg( "--------------------------------------------------------------\n\n ");
g_bSuppressPrintfOutput = bOldSuppress;
}
bool DistributeWorkDispatch( MessageBuffer *pBuf, int iSource, int iPacketID )
{
unsigned short iCurDistributeWorkCall = *((unsigned short*)&pBuf->data[2]);
if ( iCurDistributeWorkCall >= MAX_DW_CALLS )
Error( "Got an invalid DistributeWork packet (id: %d, sub: %d) (iCurDW: %d).", pBuf->data[0], pBuf->data[1], iCurDistributeWorkCall );
CDSInfo *pInfo = &g_DSInfo;
pBuf->setOffset( 4 );
switch ( pBuf->data[1] )
{
case DW_SUBPACKETID_MASTER_READY:
{
g_iMasterReadyForDistributeWorkCall = iCurDistributeWorkCall;
return true;
}
case DW_SUBPACKETID_WORKER_READY:
{
if ( iCurDistributeWorkCall > g_iCurDSInfo || !g_bMPIMaster )
Error( "State incorrect on master for DW_SUBPACKETID_WORKER_READY packet from %s.", VMPI_GetMachineName( iSource ) );
if ( iCurDistributeWorkCall == g_iCurDSInfo )
{
// Ok, give this guy some WUs.
if ( g_pCurDistributorMaster )
g_pCurDistributorMaster->OnWorkerReady( iSource );
}
return true;
}
case DW_SUBPACKETID_MASTER_FINISHED:
{
g_iMasterFinishedDistributeWorkCall = iCurDistributeWorkCall;
return true;
}
// Worker sends this to tell the master it has started on a work unit.
case DW_SUBPACKETID_WU_STARTED:
{
if ( iCurDistributeWorkCall != g_iCurDSInfo )
return true;
WUIndexType iWU;
pBuf->read( &iWU, sizeof( iWU ) );
VMPITracker_WorkUnitStarted( ( int ) iWU, iSource );
return true;
}
case DW_SUBPACKETID_WU_RESULTS:
{
// We only care about work results for the iteration we're in.
if ( iCurDistributeWorkCall != g_iCurDSInfo )
return true;
WUIndexType iWorkUnit;
pBuf->read( &iWorkUnit, sizeof( iWorkUnit ) );
if ( iWorkUnit >= pInfo->m_nWorkUnits )
{
Error( "DistributeWork: got an invalid work unit index (%I64u for WU count of %I64u).", iWorkUnit, pInfo->m_nWorkUnits );
}
HandleWorkUnitCompleted( pInfo, iSource, iWorkUnit, pBuf );
return true;
}
default:
{
if ( g_pCurDistributorMaster )
return g_pCurDistributorMaster->HandlePacket( pBuf, iSource, iCurDistributeWorkCall != g_iCurDSInfo );
else if ( g_pCurDistributorWorker )
return g_pCurDistributorWorker->HandlePacket( pBuf, iSource, iCurDistributeWorkCall != g_iCurDSInfo );
else
return false;
}
}
}
bool VMPITracker_WriteDebugFile( const char *pFilename )
{
FILE *fp = fopen( pFilename, "wt" );
if ( fp )
{
fprintf( fp, "# work units: %d\n", g_WorkUnits.Count() );
fprintf( fp, "# active work units: %d\n", CountActiveWorkUnits() );
fprintf( fp, "\n" );
fprintf( fp, "--- Events ---" );
fprintf( fp, "\n" );
fprintf( fp, "\n" );
for ( int i=0; i < g_WorkUnits.Count(); i++ )
{
CWorkUnit *wu = &g_WorkUnits[i];
if ( wu->m_iWorkerCompleted != -1 )
continue;
fprintf( fp, " work unit %d\n", i );
fprintf( fp, "\n" );
if ( wu->m_Events.Count() == 0 )
{
fprintf( fp, " *no events*\n" );
}
else
{
for ( int iEvent=0; iEvent < wu->m_Events.Count(); iEvent++ )
{
CWorkUnitEvent *pEvent = &wu->m_Events[iEvent];
if ( pEvent->m_iEventType == EVENT_TYPE_WU_STARTED )
{
fprintf( fp, " started (by worker %s) %.1f seconds ago\n",
VMPI_GetMachineName( wu->m_Events[iEvent].m_iWorker ),
Plat_FloatTime() - wu->m_Events[iEvent].m_flTime );
}
else if ( pEvent->m_iEventType == EVENT_TYPE_SEND_WORK_UNIT )
{
fprintf( fp, " sent (to worker %s) %.1f seconds ago\n",
VMPI_GetMachineName( wu->m_Events[iEvent].m_iWorker ),
Plat_FloatTime() - wu->m_Events[iEvent].m_flTime );
}
else if ( pEvent->m_iEventType == EVENT_TYPE_WU_COMPLETED )
{
fprintf( fp, " completed (by worker %s) %.1f seconds ago\n",
VMPI_GetMachineName( wu->m_Events[iEvent].m_iWorker ),
Plat_FloatTime() - wu->m_Events[iEvent].m_flTime );
}
}
}
fprintf( fp, "\n" );
}
fclose( fp );
return true;
}
else
{
return false;
}
}
bool SharedDispatch( MessageBuffer *pBuf, int iSource, int iPacketID )
{
char *pInPos = &pBuf->data[2];
switch ( pBuf->data[1] )
{
case VMPI_SUBPACKETID_DIRECTORIES:
{
Q_strncpy( gamedir, pInPos, sizeof( gamedir ) );
pInPos += strlen( pInPos ) + 1;
Q_strncpy( qdir, pInPos, sizeof( qdir ) );
g_bReceivedDirectoryInfo = true;
}
return true;
case VMPI_SUBPACKETID_DBINFO:
{
g_DBInfo = *((CDBInfo*)pInPos);
pInPos += sizeof( CDBInfo );
g_JobPrimaryID = *((unsigned long*)pInPos);
g_bReceivedDBInfo = true;
}
return true;
case VMPI_SUBPACKETID_CRASH:
{
char const chCrashInfoType = *pInPos;
pInPos += 2;
switch ( chCrashInfoType )
{
case 't':
Warning( "\nWorker '%s' dead: %s\n", VMPI_GetMachineName( iSource ), pInPos );
break;
case 'f':
{
int iFileSize = * reinterpret_cast< int const * >( pInPos );
pInPos += sizeof( iFileSize );
// Temp folder
char const *szFolder = NULL;
if ( !szFolder ) szFolder = getenv( "TEMP" );
if ( !szFolder ) szFolder = getenv( "TMP" );
if ( !szFolder ) szFolder = "c:";
// Base module name
char chModuleName[_MAX_PATH], *pModuleName = chModuleName;
::GetModuleFileName( NULL, chModuleName, sizeof( chModuleName ) / sizeof( chModuleName[0] ) );
if ( char *pch = strrchr( chModuleName, '.' ) )
*pch = 0;
if ( char *pch = strrchr( chModuleName, '\\' ) )
*pch = 0, pModuleName = pch + 1;
// Current time
time_t currTime = ::time( NULL );
struct tm * pTime = ::localtime( &currTime );
// Number of minidumps this run
static int s_numMiniDumps = 0;
++ s_numMiniDumps;
// Prepare the filename
char chSaveFileName[ 2 * _MAX_PATH ] = { 0 };
sprintf( chSaveFileName, "%s\\vmpi_%s_on_%s_%d%.2d%2d%.2d%.2d%.2d_%d.mdmp",
szFolder,
pModuleName,
VMPI_GetMachineName( iSource ),
pTime->tm_year + 1900, /* Year less 2000 */
pTime->tm_mon + 1, /* month (0 - 11 : 0 = January) */
pTime->tm_mday, /* day of month (1 - 31) */
pTime->tm_hour, /* hour (0 - 23) */
pTime->tm_min, /* minutes (0 - 59) */
pTime->tm_sec, /* seconds (0 - 59) */
s_numMiniDumps
);
if ( FILE *fDump = fopen( chSaveFileName, "wb" ) )
{
fwrite( pInPos, 1, iFileSize, fDump );
fclose( fDump );
Warning( "\nSaved worker crash minidump '%s', size %d byte(s).\n",
chSaveFileName, iFileSize );
}
else
{
Warning( "\nReceived worker crash minidump size %d byte(s), failed to save.\n", iFileSize );
}
}
break;
}
}
return true;
}
return false;
}