/*
* Acquire resource from hosts in LIBYARN mode.
*
* This function use round-robin sequence to select available hosts in HAWQ RM
* resource pool and choose suitable host to allocate containers.
*/
int RB_LIBYARN_acquireResource(uint32_t memorymb, uint32_t core, List *preferred)
{
int res = FUNC_RETURN_OK;
uint32_t MessageID = RM2RB_ALLOC_RESOURCE;
ListCell *cell = NULL;
if ( PipeReceivePending )
{
return RESBROK_PIPE_BUSY;
}
elog(LOG, "Allocating container request is triggered.");
/* Write request to pipe */
SelfMaintainBufferData sendBuffer;
initializeSelfMaintainBuffer(&sendBuffer, PCONTEXT);
appendSMBVar(&sendBuffer, MessageID);
appendSelfMaintainBufferTill64bitAligned(&sendBuffer);
RPCRequestRBAllocateResourceContainersData request;
request.ContainerCount = core;
request.MemoryMB = memorymb / core;
request.Core = 1;
request.PreferredSize = list_length(preferred);
request.MsgLength = 0;
foreach(cell, preferred)
{
PAIR pair = (PAIR)lfirst(cell);
SegResource segres = (SegResource)(pair->Key);
char *hostname = GET_SEGINFO_GRMHOSTNAME(&(segres->Stat->Info));
char *rackname = GET_SEGINFO_GRMRACKNAME(&(segres->Stat->Info));
uint32_t preferredLen = sizeof(uint32_t)*2 +
__SIZE_ALIGN64((strlen(hostname) + 1 + strlen(rackname) + 1));
request.MsgLength += preferredLen;
}
/*
* Return cluster report including all node status. The response should be
* handled asynchronously.
*/
int RB_LIBYARN_getClusterReport(const char *quename,
List **machines,
double *maxcapacity)
{
int res = FUNC_RETURN_OK;
uint32_t MessageID = RM2RB_GET_CLUSTERREPORT;
int piperes = 0;
if ( PipeReceivePending )
{
elog(DEBUG5, "YARN resource broker skip sending cluster report due to "
"not getting response of last request.");
return FUNC_RETURN_OK;
}
/* Write request to pipe */
SelfMaintainBufferData sendBuffer;
initializeSelfMaintainBuffer(&sendBuffer, PCONTEXT);
appendSMBVar(&sendBuffer, MessageID);
appendSelfMaintainBufferTill64bitAligned(&sendBuffer);
RPCRequestRBGetClusterReportHeadData request;
request.QueueNameLen = __SIZE_ALIGN64(strlen(quename) + 1);
request.Reserved = 0;
appendSMBVar(&sendBuffer, request);
appendSMBStr(&sendBuffer, quename);
appendSelfMaintainBufferTill64bitAligned(&sendBuffer);
piperes = pipewrite(ResBrokerRequestPipe[1],
sendBuffer.Buffer,
sendBuffer.Cursor + 1);
if ( piperes != sendBuffer.Cursor + 1 )
{
elog(WARNING, "YARN mode resource broker failed to generate cluster "
"report request to resource broker process through pipe. "
"Wrote length %d, expected length %d, errno %d",
piperes,
sendBuffer.Cursor + 1,
errno);
res = RESBROK_PIPE_ERROR;
}
destroySelfMaintainBuffer(&sendBuffer);
elog(LOG, "YARN mode resource broker generated cluster report request to "
"resource broker process.");
PipeReceivePending = res == FUNC_RETURN_OK;
return res;
}
int decreaseMemoryQuota(char *seghostname,
GRMContainerSet rescontainerset)
{
int res = FUNC_RETURN_OK;
AsyncCommBuffer commbuffer = NULL;
GRMContainerSet ctns = NULL;
Assert( seghostname != NULL );
Assert( rescontainerset->Containers != NULL );
/* Move resource containers that should be returned to communication context.*/
ctns = createGRMContainerSet();
moveGRMContainerSetContainerList(ctns, rescontainerset);
/* Build request */
SelfMaintainBufferData tosend;
initializeSelfMaintainBuffer(&tosend, PCONTEXT);
RPCRequestUpdateMemoryQuotaData request;
request.MemoryQuotaDelta = ctns->Allocated.MemoryMB;
GRMContainer firstcont = getGRMContainerSetContainerFirst(ctns);
Assert( firstcont != NULL );
request.MemoryQuotaTotalPending = firstcont->Resource->Allocated.MemoryMB +
firstcont->Resource->OldInuse.MemoryMB;
Assert( request.MemoryQuotaTotalPending >= 0 );
elog(LOG, "Resource manager decrease host %s memory quota "UINT64_FORMAT" MB, "
"Expected total memory quota after decreasing is " UINT64_FORMAT" MB. "
"Include %d MB old in-use memory quota.",
seghostname,
request.MemoryQuotaDelta,
request.MemoryQuotaTotalPending,
firstcont->Resource->OldInuse.MemoryMB);
appendSMBVar(&tosend, request);
/* Set content to send and add to AsyncComm framework */
AsyncCommMessageHandlerContext context =
(AsyncCommMessageHandlerContext)rm_palloc0(AsyncCommContext,
sizeof(AsyncCommMessageHandlerContextData));
context->inMessage = false;
context->UserData = ctns;
context->MessageRecvReadyHandler = NULL;
context->MessageRecvedHandler = recvDecreaseMemoryQuotaResponse;
context->MessageSendReadyHandler = NULL;
context->MessageSentHandler = sentDecreaseMemoryQuota;
context->MessageErrorHandler = sentDecreaseMemoryQuotaError;
context->MessageCleanUpHandler = sentDecreaseMemoryQuotaCleanup;
res = registerAsyncConnectionFileDesc(NULL,
seghostname,
rm_segment_port,
ASYNCCOMM_READBYTES | ASYNCCOMM_WRITEBYTES,
&AsyncCommBufferHandlersMessage,
context,
&commbuffer);
if ( res != FUNC_RETURN_OK )
{
elog(LOG, "Resource manager failed to set connection to segment host %s "
"on port %d to decrease memory quota.",
seghostname,
rm_segment_port);
processContainersAfterDecreaseMemoryQuota(ctns, false);
freeGRMContainerSet(ctns);
rm_pfree(AsyncCommContext, context);
return res;
}
else
{
elog(DEBUG3, "Resource manager succeeded set connection to segment host %s "
"on port %d to decrease memory quota.",
seghostname,
rm_segment_port);
}
buildMessageToCommBuffer(commbuffer,
tosend.Buffer,
tosend.Cursor+1,
REQUEST_RM_DECREASE_MEMORY_QUOTA,
0,
0);
destroySelfMaintainBuffer(&tosend);
context->AsyncBuffer = commbuffer;
InitHandler_Message(commbuffer);
return FUNC_RETURN_OK;
}
int sendRUAlive(char *seghostname)
{
int res = FUNC_RETURN_OK;
AsyncCommBuffer newcommbuffer = NULL;
SegResource segres = NULL;
int32_t segid = SEGSTAT_ID_INVALID;
Assert(seghostname != NULL);
res = getSegIDByHostName(seghostname, strlen(seghostname), &segid);
if ( res != FUNC_RETURN_OK )
{
elog(WARNING, "Resource manager cannot get registered host %s when send RUAlive message, "
"error code: %d",
seghostname,
res);
return res;
}
segres = getSegResource(segid);
Assert(segres != NULL);
bool oldstat = setSegResRUAlivePending(segres, true);
if( oldstat != false )
{
Assert(false);
}
/* Build request. */
SelfMaintainBufferData tosend;
initializeSelfMaintainBuffer(&tosend, PCONTEXT);
RPCRequestRUAliveData request;
request.Reserved = 0;
appendSMBVar(&tosend, request);
AsyncCommMessageHandlerContext context =
(AsyncCommMessageHandlerContext)
rm_palloc0(AsyncCommContext,
sizeof(AsyncCommMessageHandlerContextData));
context->inMessage = false;
context->MessageRecvReadyHandler = NULL;
context->MessageRecvedHandler = receivedRUAliveResponse;
context->MessageSendReadyHandler = NULL;
context->MessageSentHandler = sentRUAlive;
context->MessageErrorHandler = sentRUAliveError;
context->MessageCleanUpHandler = sentRUAliveCleanUp;
context->UserData = (void *)segres;
/* Connect to HAWQ RM server */
res = registerAsyncConnectionFileDesc(NULL,
seghostname,
rm_segment_port,
ASYNCCOMM_READBYTES | ASYNCCOMM_WRITEBYTES,
&AsyncCommBufferHandlersMessage,
context,
&newcommbuffer);
if ( res != FUNC_RETURN_OK )
{
rm_pfree(AsyncCommContext, context);
elog(WARNING, "Fail to register asynchronous connection for sending "
"RUAlive message. %d", res);
return res;
}
buildMessageToCommBuffer(newcommbuffer,
tosend.Buffer,
tosend.Cursor+1,
REQUEST_RM_RUALIVE,
0,
0);
destroySelfMaintainBuffer(&tosend);
context->AsyncBuffer = newcommbuffer;
InitHandler_Message(newcommbuffer);
return FUNC_RETURN_OK;
}
/******************************************************************************
* I aM Alive.
*
* Request:
* |<----------- 64 bits (8 bytes) ----------->|
* +----------+--------------------------------+
* | TDC | BDC | Reserved |
* +----------+----------+---------------------+
* | |
* | Machine ID info |
* | |
* +-------------------------------------------+ _____ 64bit aligned
*
* Response:
* |<----------- 64 bits (8 bytes) ----------->|
* +---------------------+---------------------+
* | heartbeat result | Reserved |
* +---------------------+---------------------+ _____ 64bit aligned
*
******************************************************************************/
int sendIMAlive(int *errorcode,
char *errorbuf,
int errorbufsize)
{
int res = FUNC_RETURN_OK;
AsyncCommBuffer newcommbuffer = NULL;
Assert( DRMGlobalInstance->LocalHostStat != NULL );
/* Build request. */
SelfMaintainBufferData tosend;
initializeSelfMaintainBuffer(&tosend, PCONTEXT);
RPCRequestHeadIMAliveData requesthead;
requesthead.TmpDirCount = getDQueueLength(&DRMGlobalInstance->LocalHostTempDirectories);
requesthead.TmpDirBrokenCount = DRMGlobalInstance->LocalHostStat->FailedTmpDirNum;
requesthead.Reserved = 0;
appendSMBVar(&tosend, requesthead);
appendSelfMaintainBuffer(&tosend,
(char *)(DRMGlobalInstance->LocalHostStat),
offsetof(SegStatData, Info) +
DRMGlobalInstance->LocalHostStat->Info.Size);
/* Set content to send and add to AsyncComm framework. */
AsyncCommMessageHandlerContext context =
rm_palloc0(AsyncCommContext,
sizeof(AsyncCommMessageHandlerContextData));
context->inMessage = false;
context->UserData = NULL;
context->MessageRecvReadyHandler = NULL;
context->MessageRecvedHandler = receivedIMAliveResponse;
context->MessageSendReadyHandler = NULL;
context->MessageSentHandler = sentIMAlive;
context->MessageErrorHandler = sentIMAliveError;
context->MessageCleanUpHandler = sentIMAliveCleanUp;
/* Connect to HAWQ RM server */
res = registerAsyncConnectionFileDesc(NULL,
DRMGlobalInstance->SendToStandby?
standby_addr_host:
master_addr_host,
rm_master_port,
ASYNCCOMM_READBYTES | ASYNCCOMM_WRITEBYTES,
&AsyncCommBufferHandlersMessage,
context,
&newcommbuffer);
if ( res != FUNC_RETURN_OK )
{
rm_pfree(AsyncCommContext, context);
elog(WARNING, "Fail to register asynchronous connection for sending "
"IMAlive message. %d", res);
return res;
}
buildMessageToCommBuffer(newcommbuffer,
tosend.Buffer,
tosend.Cursor + 1,
REQUEST_RM_IMALIVE,
0,
0);
destroySelfMaintainBuffer(&tosend);
context->AsyncBuffer = newcommbuffer;
InitHandler_Message(newcommbuffer);
return FUNC_RETURN_OK;
}
