ClRcT clCpmShutDown(ClUint32T argc, ClCharT *argv[], ClCharT **retStr)
{
ClUint32T rc = CL_OK;
if (!CL_CPM_IS_ACTIVE())
{
cpmCliPrint(retStr,
"Node shutdown operation can be performed "
"only on CPM/G active");
rc = CL_CPM_RC(CL_ERR_OP_NOT_PERMITTED);
goto failure;
}
if (2 == argc)
{
ClIocNodeAddressT nodeAddress = 0;
nodeAddress = (ClIocNodeAddressT) cpmCliStrToInt(argv[1]);
if (nodeAddress <= 0)
{
cpmCliPrint(retStr, "Invalid node address [%s]", argv[1]);
rc = CL_CPM_RC(CL_ERR_INVALID_PARAMETER);
goto failure;
}
rc = clCpmNodeShutDown(nodeAddress);
if (CL_OK != rc)
{
cpmCliPrint(retStr,
"Failed to shutdown the node, error [%#x]",
rc);
goto failure;
}
}
else
{
rc = CL_CPM_RC(CL_ERR_INVALID_PARAMETER);
cpmCliPrint(retStr,
"Usage : nodeShutdown <node address>\n"
"\tnode address[in decimal] - IOC address "
"of the node");
goto failure;
}
return CL_OK;
failure:
return rc;
}
static void cpmMakeSCActiveOrDeputy(const ClGmsClusterNotificationBufferT *notificationBuffer,
ClCpmLocalInfoT *pCpmLocalInfo)
{
ClUint32T rc = CL_OK;
ClGmsNodeIdT prevMasterNodeId = gpClCpm->activeMasterNodeId;
ClBoolT leadershipChanged = notificationBuffer->leadershipChanged;
/*
* Check for initial leadership state incase the cluster track from AMF was issued
* after GMS leader election was done and GMS responded back with a track with a leadership changed
* set to FALSE for a CURRENT async request from AMF.
*/
if(leadershipChanged == CL_FALSE &&
(ClInt32T) notificationBuffer->leader == pCpmLocalInfo->nodeId &&
(ClInt32T) prevMasterNodeId != pCpmLocalInfo->nodeId &&
gpClCpm->haState == CL_AMS_HA_STATE_NONE)
leadershipChanged = CL_TRUE;
if (leadershipChanged == CL_TRUE)
{
if ( (ClInt32T) notificationBuffer->leader == pCpmLocalInfo->nodeId)
{
clLogInfo(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_GMS,
"Node [%d] has become the leader of the cluster",
pCpmLocalInfo->nodeId);
if(gpClCpm->haState != CL_AMS_HA_STATE_ACTIVE)
{
rc = cpmUpdateTL(CL_AMS_HA_STATE_ACTIVE);
if (rc != CL_OK)
{
clLogWrite(CL_LOG_HANDLE_APP, CL_LOG_SEV_DEBUG, NULL,
CL_CPM_LOG_1_TL_UPDATE_FAILURE, rc);
}
}
gpClCpm->deputyNodeId = notificationBuffer->deputy;
}
else if ((ClInt32T) notificationBuffer->deputy == pCpmLocalInfo->nodeId)
{
clLogInfo(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_GMS,
"Node [%d] has become the deputy of the cluster",
pCpmLocalInfo->nodeId);
/*
* Deregister the active registration if going from active to standby.
*/
if(gpClCpm->haState == CL_AMS_HA_STATE_ACTIVE)
{
clIocTransparencyDeregister(pCpmLocalInfo->nodeId << CL_CPM_IOC_SLOT_BITS);
}
rc = cpmUpdateTL(CL_AMS_HA_STATE_STANDBY);
if (rc != CL_OK)
{
clLogWrite(CL_LOG_HANDLE_APP, CL_LOG_SEV_DEBUG, NULL,
CL_CPM_LOG_1_TL_UPDATE_FAILURE, rc);
}
}
if ((gpClCpm->haState == CL_AMS_HA_STATE_ACTIVE) &&
( (ClInt32T) notificationBuffer->leader != pCpmLocalInfo->nodeId))
{
clLogDebug(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_GMS,
"Node [%d] is changing HA state from active to standby",
pCpmLocalInfo->nodeId);
/*
* Deregister the entry during a state change.
*/
if ( (ClInt32T) notificationBuffer->deputy != pCpmLocalInfo->nodeId)
{
clIocTransparencyDeregister((pCpmLocalInfo->nodeId) << CL_CPM_IOC_SLOT_BITS);
}
/*
* Inform AMS to become standby and read the checkpoint.
*/
if ((gpClCpm->cpmToAmsCallback != NULL) &&
(gpClCpm->cpmToAmsCallback->amsStateChange != NULL))
{
clLogDebug(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_GMS,
"Informing AMS on node [%d] to change state "
"from active to standby...",
pCpmLocalInfo->nodeId);
rc = gpClCpm->cpmToAmsCallback->amsStateChange(CL_AMS_STATE_CHANGE_ACTIVE_TO_STANDBY |
CL_AMS_STATE_CHANGE_USE_CHECKPOINT);
if (CL_OK != rc)
{
clLogWarning(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_AMS,
"AMS state change from active to standby "
"returned [%#x]",
rc);
}
cpmWriteNodeStatToFile("AMS", CL_NO);
if(!gClAmsSwitchoverInline)
{
if (((notificationBuffer->numberOfItems == 0) &&
(notificationBuffer->notification == NULL)) &&
gpClCpm->polling &&
(gpClCpm->nodeLeaving == CL_FALSE))
{
/*
* This indicates that leader election API of
* GMS was called. Since this involves
* interaction among only system controllers,
* we don't need to restart the worker nodes
* like in the case of split brain handling.
*/
cpmActive2Standby(CL_NO);
}
else if (( (ClInt32T) notificationBuffer->deputy == pCpmLocalInfo->nodeId) && gpClCpm->polling && (gpClCpm->nodeLeaving == CL_FALSE))
{
/*
* We try and handle a possible split brain
* since presently GMS shouldnt be reelecting.
* And even if it does, its pretty much
* invalid with respect to AMS where you could
* land up with 2 actives.
*/
/*
* We arent expected to return back.
*/
cpmActive2Standby(CL_NO);
}
}
}
/*
* Bug 4168:
* Updating the data structure gpClCpm, when the active becomes
* standby.
*/
gpClCpm->haState = CL_AMS_HA_STATE_STANDBY;
gpClCpm->activeMasterNodeId = notificationBuffer->leader;
gpClCpm->deputyNodeId = notificationBuffer->deputy;
if(gClAmsSwitchoverInline)
{
/*
*Re-register with active.
*/
clOsalMutexUnlock(&gpClCpm->clusterMutex);
cpmSwitchoverActive();
clOsalMutexLock(&gpClCpm->clusterMutex);
}
}
else if ((gpClCpm->haState == CL_AMS_HA_STATE_STANDBY) &&
( (ClInt32T) notificationBuffer->leader == pCpmLocalInfo->nodeId))
{
rc = cpmStandby2Active(prevMasterNodeId,
notificationBuffer->deputy);
if (CL_OK != rc)
{
return;
}
}
else if ((gpClCpm->haState == CL_AMS_HA_STATE_NONE) &&
( (ClInt32T) notificationBuffer->leader ==
pCpmLocalInfo->nodeId))
{
/*
* Bug 4411:
* Added the if-else block.
* Calling the AMS initialize only if both the callback pointers
* are not NULL.
* FIXME: This change is sort of workaround for 2.2.
* Neat solution is to protect gpClCpm structure and fields properly
*/
if ((gpClCpm->amsToCpmCallback != NULL) &&
(gpClCpm->cpmToAmsCallback != NULL))
{
clLogDebug(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_AMS,"Starting AMS in active mode...");
rc = clAmsStart(&gAms,CL_AMS_INSTANTIATE_MODE_ACTIVE | CL_AMS_INSTANTIATE_USE_CHECKPOINT);
/*
* Bug 4092:
* If the AMS intitialize fails then do the
* self shut down.
*/
if (CL_OK != rc)
{
clLogCritical(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_AMS,"Unable to initialize AMF, error = [%#x]", rc);
gpClCpm->amsToCpmCallback = NULL;
cpmReset(NULL,NULL);
return;
}
}
else
{
rc = CL_CPM_RC(CL_ERR_NULL_POINTER);
if (!gpClCpm->polling)
{
clLogCritical(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_AMS, "AMF finalize called before AMF initialize during node shutdown.");
}
else
{
clLogCritical(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_AMS, "Unable to initialize AMF, error = [%#x]", rc);
cpmRestart(NULL,NULL);
}
return;
}
gpClCpm->haState = CL_AMS_HA_STATE_ACTIVE;
gpClCpm->activeMasterNodeId = notificationBuffer->leader;
gpClCpm->deputyNodeId = notificationBuffer->deputy;
}
/*
* There could be more than 1 standby SC.
* Every none HA-state SC except the master is updated to become to standby SC.
*/
else if (gpClCpm->haState == CL_AMS_HA_STATE_NONE)
{
/*
* Bug 4411:
* Added the if-else block.
* Calling the clAmsStart only if both the callback pointers
* are not NULL.
* FIXME: This change is sort of workaround for 2.2.
* Neat solution is to protect gpClCpm structure and fields properly
*/
if ((gpClCpm->amsToCpmCallback != NULL) &&
(gpClCpm->cpmToAmsCallback != NULL))
{
clLogDebug(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_AMS,
"Starting AMS in standby mode...");
rc = clAmsStart(&gAms,CL_AMS_INSTANTIATE_MODE_STANDBY);
/*
* Bug 4092:
* If the AMS initialize fails then do the
* self shut down.
*/
if (CL_OK != rc)
{
clLogCritical(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_AMS,"Unable to initialize AMS, error = [%#x]", rc);
gpClCpm->amsToCpmCallback = NULL;
cpmRestart(NULL,NULL);
return;
}
}
else
{
rc = CL_CPM_RC(CL_ERR_NULL_POINTER);
if (!gpClCpm->polling)
{
clLogCritical(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_AMS,"AMS finalize called before AMS initialize during node shutdown.");
}
else
{
clLogCritical(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_AMS,"Unable to initialize AMS, error = [%#x]", rc);
cpmRestart(NULL,NULL);
}
return;
}
gpClCpm->haState = CL_AMS_HA_STATE_STANDBY;
gpClCpm->activeMasterNodeId = notificationBuffer->leader;
gpClCpm->deputyNodeId = notificationBuffer->deputy;
if(gpClCpm->bmTable->currentBootLevel == pCpmLocalInfo->defaultBootLevel)
{
cpmInitializeStandby();
}
}
else
{
gpClCpm->activeMasterNodeId = notificationBuffer->leader;
gpClCpm->deputyNodeId = notificationBuffer->deputy;
}
clLogNotice(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_CPM,
"HA state of node [%s] with node ID [%d] is [%s], "
"Master node is [%d]",
pCpmLocalInfo->nodeName,
pCpmLocalInfo->nodeId,
gpClCpm->haState == CL_AMS_HA_STATE_ACTIVE ? "Active":
gpClCpm->haState == CL_AMS_HA_STATE_STANDBY ? "Standby":
"None",
gpClCpm->activeMasterNodeId);
}
else
{
/*
* Always update the deputy node ID. It may be that this
* path is reached because a deputy node failed.
*/
gpClCpm->deputyNodeId = notificationBuffer->deputy;
if (CL_CPM_IS_ACTIVE())
{
if(notificationBuffer->notification &&
notificationBuffer->numberOfItems == 1)
{
if (CL_GMS_NODE_LEFT ==
notificationBuffer->notification->clusterChange)
{
cpmFailoverNode(notificationBuffer->notification->
clusterNode.nodeId, CL_FALSE);
}
else
{
clLogNotice(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_AMS,
"Ignoring notification of type [%d] for node [%d]",
notificationBuffer->notification->clusterChange,
notificationBuffer->notification->clusterNode.nodeId);
}
}
else if(notificationBuffer->notification)
{
clLogNotice(CPM_LOG_AREA_CPM, CPM_LOG_CTX_CPM_AMS,
"Ignoring notification with number of items [%d], first type [%d]",
notificationBuffer->numberOfItems,
notificationBuffer->notification->clusterChange);
}
}
else if ((gpClCpm->haState == CL_AMS_HA_STATE_NONE) && ( (ClInt32T) notificationBuffer->deputy == pCpmLocalInfo->nodeId))
{
if(CL_GMS_NODE_JOINED ==
notificationBuffer->notification->clusterChange)
{
cpmStandbyRecover(notificationBuffer);
}
else if(gpClCpm->bmTable->currentBootLevel == pCpmLocalInfo->defaultBootLevel)
{
cpmStandbyRecover(notificationBuffer);
cpmInitializeStandby();
}
}
}
return ;
}