[merge from 2.1.x] Bug 5781: Only grab capacity global lock when accessing capacity tables. Once the queries are done, release the lock so that others can query for the data or update the data as needed.

This commit is contained in:
Kris McQueen 2010-08-23 13:41:03 -07:00
parent f98e272d48
commit adce18b2c0
2 changed files with 139 additions and 131 deletions

View File

@ -154,47 +154,23 @@ public class UserConcentratedAllocator implements PodAllocator {
}
private boolean dataCenterAndPodHasEnoughCapacity(long dataCenterId, long podId, long capacityNeeded, short capacityType, long[] hostCandidate) {
List<CapacityVO> capacities = null;
if (m_capacityCheckLock.lock(120)) { // 2 minutes
try {
SearchCriteria<CapacityVO> sc = _capacityDao.createSearchCriteria();
sc.addAnd("capacityType", SearchCriteria.Op.EQ, capacityType);
sc.addAnd("dataCenterId", SearchCriteria.Op.EQ, dataCenterId);
sc.addAnd("podId", SearchCriteria.Op.EQ, podId);
List<CapacityVO> capacities = _capacityDao.search(sc, null);
boolean enoughCapacity = false;
if (capacities != null) {
for (CapacityVO capacity : capacities) {
if(capacityType == CapacityVO.CAPACITY_TYPE_CPU || capacityType == CapacityVO.CAPACITY_TYPE_MEMORY) {
//
// for CPU/Memory, we now switch to static allocation
//
if ((capacity.getTotalCapacity() -
calcHostAllocatedCpuMemoryCapacity(capacity.getHostOrPoolId(), capacityType)) >= capacityNeeded) {
hostCandidate[0] = capacity.getHostOrPoolId();
enoughCapacity = true;
break;
}
} else {
if ((capacity.getTotalCapacity() - capacity.getUsedCapacity()) >= capacityNeeded) {
hostCandidate[0] = capacity.getHostOrPoolId();
enoughCapacity = true;
break;
}
}
}
}
return enoughCapacity;
capacities = _capacityDao.search(sc, null);
} finally {
m_capacityCheckLock.unlock();
}
} else {
s_logger.error("Unable to acquire synchronization lock for pod allocation");
s_logger.error("Unable to acquire synchronization lock for pod allocation");
// we now try to enforce reservation-style allocation, waiting time has been adjusted
// to 2 minutes
return false;
// we now try to enforce reservation-style allocation, waiting time has been adjusted
// to 2 minutes
return false;
/*
// If we can't lock the table, just return that there is enough capacity and allow instance creation to fail on the agent
@ -203,6 +179,31 @@ public class UserConcentratedAllocator implements PodAllocator {
return true;
*/
}
boolean enoughCapacity = false;
if (capacities != null) {
for (CapacityVO capacity : capacities) {
if(capacityType == CapacityVO.CAPACITY_TYPE_CPU || capacityType == CapacityVO.CAPACITY_TYPE_MEMORY) {
//
// for CPU/Memory, we now switch to static allocation
//
if ((capacity.getTotalCapacity() -
calcHostAllocatedCpuMemoryCapacity(capacity.getHostOrPoolId(), capacityType)) >= capacityNeeded) {
hostCandidate[0] = capacity.getHostOrPoolId();
enoughCapacity = true;
break;
}
} else {
if ((capacity.getTotalCapacity() - capacity.getUsedCapacity()) >= capacityNeeded) {
hostCandidate[0] = capacity.getHostOrPoolId();
enoughCapacity = true;
break;
}
}
}
}
return enoughCapacity;
}
private boolean skipCalculation(VMInstanceVO vm) {

View File

@ -327,113 +327,120 @@ public class AlertManagerImpl implements AlertManager {
// the amount allocated. Hopefully it's limited to 3 entry points and will keep the amount allocated
// per host accurate.
if (m_capacityCheckLock.lock(5)) { // 5 second timeout
if (s_logger.isTraceEnabled()) {
s_logger.trace("recalculating system capacity");
if (s_logger.isTraceEnabled()) {
s_logger.trace("recalculating system capacity");
}
List<CapacityVO> newCapacities = new ArrayList<CapacityVO>();
// get all hosts..
SearchCriteria sc = _hostDao.createSearchCriteria();
sc.addAnd("status", SearchCriteria.Op.EQ, Status.Up.toString());
List<HostVO> hosts = _hostDao.search(sc, null);
// prep the service offerings
List<ServiceOfferingVO> offerings = _offeringsDao.listAll();
Map<Long, ServiceOfferingVO> offeringsMap = new HashMap<Long, ServiceOfferingVO>();
for (ServiceOfferingVO offering : offerings) {
offeringsMap.put(offering.getId(), offering);
}
for (HostVO host : hosts) {
if (host.getType() != Host.Type.Routing) {
continue;
}
long cpu = 0;
long usedMemory = 0;
List<DomainRouterVO> domainRouters = _routerDao.listUpByHostId(host.getId());
if (s_logger.isDebugEnabled()) {
s_logger.debug("Found " + domainRouters.size() + " router domains on host " + host.getId());
}
for (DomainRouterVO router : domainRouters) {
usedMemory += router.getRamSize() * 1024L * 1024L;
}
List<ConsoleProxyVO> proxys = _consoleProxyDao.listUpByHostId(host.getId());
if (s_logger.isDebugEnabled()) {
s_logger.debug("Found " + proxys.size() + " console proxy on host " + host.getId());
}
for(ConsoleProxyVO proxy : proxys) {
usedMemory += proxy.getRamSize() * 1024L * 1024L;
}
List<SecondaryStorageVmVO> secStorageVms = _secStorgaeVmDao.listUpByHostId(host.getId());
if (s_logger.isDebugEnabled()) {
s_logger.debug("Found " + secStorageVms.size() + " secondary storage VM on host " + host.getId());
}
for(SecondaryStorageVmVO secStorageVm : secStorageVms) {
usedMemory += secStorageVm.getRamSize() * 1024L * 1024L;
}
List<UserVmVO> vms = _userVmDao.listUpByHostId(host.getId());
if (s_logger.isDebugEnabled()) {
s_logger.debug("Found " + vms.size() + " user VM on host " + host.getId());
}
for (UserVmVO vm : vms) {
ServiceOffering so = offeringsMap.get(vm.getServiceOfferingId());
usedMemory += so.getRamSize() * 1024L * 1024L;
cpu += so.getCpu() * (so.getSpeed() * 0.99);
}
long totalMemory = host.getTotalMemory();
CapacityVO newMemoryCapacity = new CapacityVO(host.getId(), host.getDataCenterId(), host.getPodId(), usedMemory, totalMemory, CapacityVO.CAPACITY_TYPE_MEMORY);
CapacityVO newCPUCapacity = new CapacityVO(host.getId(), host.getDataCenterId(), host.getPodId(), cpu, (long)(host.getCpus()*host.getSpeed()* _cpuOverProvisioningFactor), CapacityVO.CAPACITY_TYPE_CPU);
newCapacities.add(newMemoryCapacity);
newCapacities.add(newCPUCapacity);
}
// Calculate storage pool capacity
List<StoragePoolVO> storagePools = _storagePoolDao.listAllActive();
for (StoragePoolVO pool : storagePools) {
long disk = 0l;
Pair<Long, Long> sizes = _volumeDao.getCountAndTotalByPool(pool.getId());
disk = sizes.second();
int provFactor = 1;
if( pool.getPoolType() == StoragePoolType.NetworkFilesystem ) {
provFactor = _overProvisioningFactor;
}
CapacityVO newStorageCapacity = new CapacityVO(pool.getId(), pool.getDataCenterId(), pool.getPodId(), disk, pool.getCapacityBytes() * provFactor, CapacityVO.CAPACITY_TYPE_STORAGE_ALLOCATED);
newCapacities.add(newStorageCapacity);
continue;
}
// Calculate new Public IP capacity
List<DataCenterVO> datacenters = _dcDao.listAll();
for (DataCenterVO datacenter : datacenters) {
long dcId = datacenter.getId();
int totalPublicIPs = _publicIPAddressDao.countIPs(dcId, -1, false);
int allocatedPublicIPs = _publicIPAddressDao.countIPs(dcId, -1, true);
CapacityVO newPublicIPCapacity = new CapacityVO(null, dcId, null, allocatedPublicIPs, totalPublicIPs, CapacityVO.CAPACITY_TYPE_PUBLIC_IP);
newCapacities.add(newPublicIPCapacity);
}
// Calculate new Private IP capacity
List<HostPodVO> pods = _podDao.listAll();
for (HostPodVO pod : pods) {
long podId = pod.getId();
long dcId = pod.getDataCenterId();
int totalPrivateIPs = _privateIPAddressDao.countIPs(podId, dcId, false);
int allocatedPrivateIPs = _privateIPAddressDao.countIPs(podId, dcId, true);
CapacityVO newPrivateIPCapacity = new CapacityVO(null, dcId, podId, allocatedPrivateIPs, totalPrivateIPs, CapacityVO.CAPACITY_TYPE_PRIVATE_IP);
newCapacities.add(newPrivateIPCapacity);
}
if (m_capacityCheckLock.lock(5)) { // 5 second timeout
try {
// delete the old records
_capacityDao.clearNonStorageCapacities();
// get all hosts..
SearchCriteria<HostVO> sc = _hostDao.createSearchCriteria();
sc.addAnd("status", SearchCriteria.Op.EQ, Status.Up.toString());
List<HostVO> hosts = _hostDao.search(sc, null);
// prep the service offerings
List<ServiceOfferingVO> offerings = _offeringsDao.listAll();
Map<Long, ServiceOfferingVO> offeringsMap = new HashMap<Long, ServiceOfferingVO>();
for (ServiceOfferingVO offering : offerings) {
offeringsMap.put(offering.getId(), offering);
}
for (HostVO host : hosts) {
if (host.getType() != Host.Type.Routing) {
continue;
}
long cpu = 0;
long usedMemory = 0;
List<DomainRouterVO> domainRouters = _routerDao.listUpByHostId(host.getId());
if (s_logger.isDebugEnabled()) {
s_logger.debug("Found " + domainRouters.size() + " router domains on host " + host.getId());
}
for (DomainRouterVO router : domainRouters) {
usedMemory += router.getRamSize() * 1024L * 1024L;
}
List<ConsoleProxyVO> proxys = _consoleProxyDao.listUpByHostId(host.getId());
if (s_logger.isDebugEnabled()) {
s_logger.debug("Found " + proxys.size() + " console proxy on host " + host.getId());
}
for(ConsoleProxyVO proxy : proxys) {
usedMemory += proxy.getRamSize() * 1024L * 1024L;
}
List<SecondaryStorageVmVO> secStorageVms = _secStorgaeVmDao.listUpByHostId(host.getId());
if (s_logger.isDebugEnabled()) {
s_logger.debug("Found " + secStorageVms.size() + " secondary storage VM on host " + host.getId());
}
for(SecondaryStorageVmVO secStorageVm : secStorageVms) {
usedMemory += secStorageVm.getRamSize() * 1024L * 1024L;
}
List<UserVmVO> vms = _userVmDao.listUpByHostId(host.getId());
if (s_logger.isDebugEnabled()) {
s_logger.debug("Found " + vms.size() + " user VM on host " + host.getId());
}
for (UserVmVO vm : vms) {
ServiceOffering so = offeringsMap.get(vm.getServiceOfferingId());
usedMemory += so.getRamSize() * 1024L * 1024L;
cpu += so.getCpu() * (so.getSpeed() * 0.99);
}
long totalMemory = host.getTotalMemory();
CapacityVO newMemoryCapacity = new CapacityVO(host.getId(), host.getDataCenterId(), host.getPodId(), usedMemory, totalMemory, CapacityVO.CAPACITY_TYPE_MEMORY);
CapacityVO newCPUCapacity = new CapacityVO(host.getId(), host.getDataCenterId(), host.getPodId(), cpu, (long)(host.getCpus()*host.getSpeed()* _cpuOverProvisioningFactor), CapacityVO.CAPACITY_TYPE_CPU);
_capacityDao.persist(newMemoryCapacity);
_capacityDao.persist(newCPUCapacity);
}
// Calculate storage pool capacity
List<StoragePoolVO> storagePools = _storagePoolDao.listAllActive();
for (StoragePoolVO pool : storagePools) {
long disk = 0l;
Pair<Long, Long> sizes = _volumeDao.getCountAndTotalByPool(pool.getId());
disk = sizes.second();
int provFactor = 1;
if( pool.getPoolType() == StoragePoolType.NetworkFilesystem ) {
provFactor = _overProvisioningFactor;
}
CapacityVO newStorageCapacity = new CapacityVO(pool.getId(), pool.getDataCenterId(), pool.getPodId(), disk, pool.getCapacityBytes() * provFactor, CapacityVO.CAPACITY_TYPE_STORAGE_ALLOCATED);
_capacityDao.persist(newStorageCapacity);
continue;
}
// Calculate new Public IP capacity
List<DataCenterVO> datacenters = _dcDao.listAll();
for (DataCenterVO datacenter : datacenters) {
long dcId = datacenter.getId();
int totalPublicIPs = _publicIPAddressDao.countIPs(dcId, -1, false);
int allocatedPublicIPs = _publicIPAddressDao.countIPs(dcId, -1, true);
CapacityVO newPublicIPCapacity = new CapacityVO(null, dcId, null, allocatedPublicIPs, totalPublicIPs, CapacityVO.CAPACITY_TYPE_PUBLIC_IP);
_capacityDao.persist(newPublicIPCapacity);
}
// Calculate new Private IP capacity
List<HostPodVO> pods = _podDao.listAll();
for (HostPodVO pod : pods) {
long podId = pod.getId();
long dcId = pod.getDataCenterId();
int totalPrivateIPs = _privateIPAddressDao.countIPs(podId, dcId, false);
int allocatedPrivateIPs = _privateIPAddressDao.countIPs(podId, dcId, true);
CapacityVO newPrivateIPCapacity = new CapacityVO(null, dcId, podId, allocatedPrivateIPs, totalPrivateIPs, CapacityVO.CAPACITY_TYPE_PRIVATE_IP);
_capacityDao.persist(newPrivateIPCapacity);
for (CapacityVO newCapacity : newCapacities) {
_capacityDao.persist(newCapacity);
}
} finally {
m_capacityCheckLock.unlock();