/// <summary>
/// Assign a container to this node to facilitate
/// <paramref name="request"/>
/// . If node does
/// not have enough memory, create a reservation. This is called once we are
/// sure the particular request should be facilitated by this node.
/// </summary>
/// <param name="node">The node to try placing the container on.</param>
/// <param name="request">The ResourceRequest we're trying to satisfy.</param>
/// <param name="type">The locality of the assignment.</param>
/// <param name="reserved">Whether there's already a container reserved for this app on the node.
/// </param>
/// <returns>
/// If an assignment was made, returns the resources allocated to the
/// container. If a reservation was made, returns
/// FairScheduler.CONTAINER_RESERVED. If no assignment or reservation was
/// made, returns an empty resource.
/// </returns>
private Org.Apache.Hadoop.Yarn.Api.Records.Resource AssignContainer(FSSchedulerNode
node, ResourceRequest request, NodeType type, bool reserved)
{
// How much does this request need?
Org.Apache.Hadoop.Yarn.Api.Records.Resource capability = request.GetCapability();
// How much does the node have?
Org.Apache.Hadoop.Yarn.Api.Records.Resource available = node.GetAvailableResource
();
Container container = null;
if (reserved)
{
container = node.GetReservedContainer().GetContainer();
}
else
{
container = CreateContainer(node, capability, request.GetPriority());
}
// Can we allocate a container on this node?
if (Resources.FitsIn(capability, available))
{
// Inform the application of the new container for this request
RMContainer allocatedContainer = Allocate(type, node, request.GetPriority(), request
, container);
if (allocatedContainer == null)
{
// Did the application need this resource?
if (reserved)
{
Unreserve(request.GetPriority(), node);
}
return(Resources.None());
}
// If we had previously made a reservation, delete it
if (reserved)
{
Unreserve(request.GetPriority(), node);
}
// Inform the node
node.AllocateContainer(allocatedContainer);
// If this container is used to run AM, update the leaf queue's AM usage
if (GetLiveContainers().Count == 1 && !GetUnmanagedAM())
{
((FSLeafQueue)GetQueue()).AddAMResourceUsage(container.GetResource());
SetAmRunning(true);
}
return(container.GetResource());
}
else
{
if (!FairScheduler.FitsInMaxShare(((FSLeafQueue)GetQueue()), capability))
{
return(Resources.None());
}
// The desired container won't fit here, so reserve
Reserve(request.GetPriority(), node, container, reserved);
return(FairScheduler.ContainerReserved);
}
}
/// <summary>
/// Remove the reservation on
/// <paramref name="node"/>
/// at the given
/// <see cref="Org.Apache.Hadoop.Yarn.Api.Records.Priority"/>
/// .
/// This dispatches SchedulerNode handlers as well.
/// </summary>
public virtual void Unreserve(Priority priority, FSSchedulerNode node)
{
RMContainer rmContainer = node.GetReservedContainer();
UnreserveInternal(priority, node);
node.UnreserveResource(this);
GetMetrics().UnreserveResource(GetUser(), rmContainer.GetContainer().GetResource(
));
}
/// <summary>Helper method to check if the queue should attempt assigning resources</summary>
/// <returns>true if check passes (can assign) or false otherwise</returns>
protected internal virtual bool AssignContainerPreCheck(FSSchedulerNode node)
{
if (!Resources.FitsIn(GetResourceUsage(), scheduler.GetAllocationConfiguration().
GetMaxResources(GetName())) || node.GetReservedContainer() != null)
{
return(false);
}
return(true);
}
public virtual RMContainer Allocate(NodeType type, FSSchedulerNode node, Priority
priority, ResourceRequest request, Container container)
{
lock (this)
{
// Update allowed locality level
NodeType allowed = allowedLocalityLevel[priority];
if (allowed != null)
{
if (allowed.Equals(NodeType.OffSwitch) && (type.Equals(NodeType.NodeLocal) || type
.Equals(NodeType.RackLocal)))
{
this.ResetAllowedLocalityLevel(priority, type);
}
else
{
if (allowed.Equals(NodeType.RackLocal) && type.Equals(NodeType.NodeLocal))
{
this.ResetAllowedLocalityLevel(priority, type);
}
}
}
// Required sanity check - AM can call 'allocate' to update resource
// request without locking the scheduler, hence we need to check
if (GetTotalRequiredResources(priority) <= 0)
{
return(null);
}
// Create RMContainer
RMContainer rmContainer = new RMContainerImpl(container, GetApplicationAttemptId(
), node.GetNodeID(), appSchedulingInfo.GetUser(), rmContext);
// Add it to allContainers list.
newlyAllocatedContainers.AddItem(rmContainer);
liveContainers[container.GetId()] = rmContainer;
// Update consumption and track allocations
IList <ResourceRequest> resourceRequestList = appSchedulingInfo.Allocate(type, node
, priority, request, container);
Resources.AddTo(currentConsumption, container.GetResource());
// Update resource requests related to "request" and store in RMContainer
((RMContainerImpl)rmContainer).SetResourceRequests(resourceRequestList);
// Inform the container
rmContainer.Handle(new RMContainerEvent(container.GetId(), RMContainerEventType.Start
));
if (Log.IsDebugEnabled())
{
Log.Debug("allocate: applicationAttemptId=" + container.GetId().GetApplicationAttemptId
() + " container=" + container.GetId() + " host=" + container.GetNodeId().GetHost
() + " type=" + type);
}
RMAuditLogger.LogSuccess(GetUser(), RMAuditLogger.AuditConstants.AllocContainer,
"SchedulerApp", GetApplicationId(), container.GetId());
return(rmContainer);
}
}
/// <summary>
/// Whether this app has containers requests that could be satisfied on the
/// given node, if the node had full space.
/// </summary>
public virtual bool HasContainerForNode(Priority prio, FSSchedulerNode node)
{
ResourceRequest anyRequest = GetResourceRequest(prio, ResourceRequest.Any);
ResourceRequest rackRequest = GetResourceRequest(prio, node.GetRackName());
ResourceRequest nodeRequest = GetResourceRequest(prio, node.GetNodeName());
return(anyRequest != null && anyRequest.GetNumContainers() > 0 && (anyRequest.GetRelaxLocality
() || (rackRequest != null && rackRequest.GetNumContainers() > 0)) && (rackRequest
== null || rackRequest.GetRelaxLocality() || (nodeRequest != null && nodeRequest
.GetNumContainers() > 0)) && Resources.LessThanOrEqual(ResourceCalculator, null,
anyRequest.GetCapability(), node.GetRMNode().GetTotalCapability()));
}
/// <summary>
/// Create and return a container object reflecting an allocation for the
/// given appliction on the given node with the given capability and
/// priority.
/// </summary>
public virtual Container CreateContainer(FSSchedulerNode node, Org.Apache.Hadoop.Yarn.Api.Records.Resource
capability, Priority priority)
{
NodeId nodeId = node.GetRMNode().GetNodeID();
ContainerId containerId = BuilderUtils.NewContainerId(GetApplicationAttemptId(),
GetNewContainerId());
// Create the container
Container container = BuilderUtils.NewContainer(containerId, nodeId, node.GetRMNode
().GetHttpAddress(), capability, priority, null);
return(container);
}
public override Org.Apache.Hadoop.Yarn.Api.Records.Resource AssignContainer(FSSchedulerNode
node)
{
Org.Apache.Hadoop.Yarn.Api.Records.Resource assigned = Resources.None();
if (Log.IsDebugEnabled())
{
Log.Debug("Node " + node.GetNodeName() + " offered to queue: " + GetName());
}
if (!AssignContainerPreCheck(node))
{
return(assigned);
}
IComparer <Schedulable> comparator = policy.GetComparator();
writeLock.Lock();
try
{
runnableApps.Sort(comparator);
}
finally
{
writeLock.Unlock();
}
// Release write lock here for better performance and avoiding deadlocks.
// runnableApps can be in unsorted state because of this section,
// but we can accept it in practice since the probability is low.
readLock.Lock();
try
{
foreach (FSAppAttempt sched in runnableApps)
{
if (SchedulerAppUtils.IsBlacklisted(sched, node, Log))
{
continue;
}
assigned = sched.AssignContainer(node);
if (!assigned.Equals(Resources.None()))
{
break;
}
}
}
finally
{
readLock.Unlock();
}
return(assigned);
}
/// <summary>
/// Reserve a spot for
/// <paramref name="container"/>
/// on this
/// <paramref name="node"/>
/// . If
/// the container is
/// <paramref name="alreadyReserved"/>
/// on the node, simply
/// update relevant bookeeping. This dispatches ro relevant handlers
/// in
/// <see cref="FSSchedulerNode"/>
/// ..
/// </summary>
private void Reserve(Priority priority, FSSchedulerNode node, Container container
, bool alreadyReserved)
{
Log.Info("Making reservation: node=" + node.GetNodeName() + " app_id=" + GetApplicationId
());
if (!alreadyReserved)
{
GetMetrics().ReserveResource(GetUser(), container.GetResource());
RMContainer rmContainer = base.Reserve(node, priority, null, container);
node.ReserveResource(this, priority, rmContainer);
}
else
{
RMContainer rmContainer = node.GetReservedContainer();
base.Reserve(node, priority, rmContainer, container);
node.ReserveResource(this, priority, rmContainer);
}
}
public override Org.Apache.Hadoop.Yarn.Api.Records.Resource AssignContainer(FSSchedulerNode
node)
{
Org.Apache.Hadoop.Yarn.Api.Records.Resource assigned = Resources.None();
// If this queue is over its limit, reject
if (!AssignContainerPreCheck(node))
{
return(assigned);
}
childQueues.Sort(policy.GetComparator());
foreach (FSQueue child in childQueues)
{
assigned = child.AssignContainer(node);
if (!Resources.Equals(assigned, Resources.None()))
{
break;
}
}
return(assigned);
}
/// <summary>
/// Called when this application already has an existing reservation on the
/// given node.
/// </summary>
/// <remarks>
/// Called when this application already has an existing reservation on the
/// given node. Sees whether we can turn the reservation into an allocation.
/// Also checks whether the application needs the reservation anymore, and
/// releases it if not.
/// </remarks>
/// <param name="node">Node that the application has an existing reservation on</param>
public virtual Org.Apache.Hadoop.Yarn.Api.Records.Resource AssignReservedContainer
(FSSchedulerNode node)
{
RMContainer rmContainer = node.GetReservedContainer();
Priority priority = rmContainer.GetReservedPriority();
// Make sure the application still needs requests at this priority
if (GetTotalRequiredResources(priority) == 0)
{
Unreserve(priority, node);
return(Resources.None());
}
// Fail early if the reserved container won't fit.
// Note that we have an assumption here that there's only one container size
// per priority.
if (!Resources.FitsIn(node.GetReservedContainer().GetReservedResource(), node.GetAvailableResource
()))
{
return(Resources.None());
}
return(AssignContainer(node, true));
}
private void UnreserveInternal(Priority priority, FSSchedulerNode node)
{
lock (this)
{
IDictionary <NodeId, RMContainer> reservedContainers = this.reservedContainers[priority
];
RMContainer reservedContainer = Sharpen.Collections.Remove(reservedContainers, node
.GetNodeID());
if (reservedContainers.IsEmpty())
{
Sharpen.Collections.Remove(this.reservedContainers, priority);
}
// Reset the re-reservation count
ResetReReservations(priority);
Org.Apache.Hadoop.Yarn.Api.Records.Resource resource = reservedContainer.GetContainer
().GetResource();
Resources.SubtractFrom(currentReservation, resource);
Log.Info("Application " + GetApplicationId() + " unreserved " + " on node " + node
+ ", currently has " + reservedContainers.Count + " at priority " + priority +
"; currentReservation " + currentReservation);
}
}
public virtual Org.Apache.Hadoop.Yarn.Api.Records.Resource AssignContainer(FSSchedulerNode
node)
{
return(AssignContainer(node, false));
}
private Org.Apache.Hadoop.Yarn.Api.Records.Resource AssignContainer(FSSchedulerNode
node, bool reserved)
{
if (Log.IsDebugEnabled())
{
Log.Debug("Node offered to app: " + GetName() + " reserved: " + reserved);
}
ICollection <Priority> prioritiesToTry = (reserved) ? Arrays.AsList(node.GetReservedContainer
().GetReservedPriority()) : GetPriorities();
// For each priority, see if we can schedule a node local, rack local
// or off-switch request. Rack of off-switch requests may be delayed
// (not scheduled) in order to promote better locality.
lock (this)
{
foreach (Priority priority in prioritiesToTry)
{
if (GetTotalRequiredResources(priority) <= 0 || !HasContainerForNode(priority, node
))
{
continue;
}
AddSchedulingOpportunity(priority);
// Check the AM resource usage for the leaf queue
if (GetLiveContainers().Count == 0 && !GetUnmanagedAM())
{
if (!((FSLeafQueue)GetQueue()).CanRunAppAM(GetAMResource()))
{
return(Resources.None());
}
}
ResourceRequest rackLocalRequest = GetResourceRequest(priority, node.GetRackName(
));
ResourceRequest localRequest = GetResourceRequest(priority, node.GetNodeName());
if (localRequest != null && !localRequest.GetRelaxLocality())
{
Log.Warn("Relax locality off is not supported on local request: " + localRequest);
}
NodeType allowedLocality;
if (scheduler.IsContinuousSchedulingEnabled())
{
allowedLocality = GetAllowedLocalityLevelByTime(priority, scheduler.GetNodeLocalityDelayMs
(), scheduler.GetRackLocalityDelayMs(), scheduler.GetClock().GetTime());
}
else
{
allowedLocality = GetAllowedLocalityLevel(priority, scheduler.GetNumClusterNodes(
), scheduler.GetNodeLocalityThreshold(), scheduler.GetRackLocalityThreshold());
}
if (rackLocalRequest != null && rackLocalRequest.GetNumContainers() != 0 && localRequest
!= null && localRequest.GetNumContainers() != 0)
{
return(AssignContainer(node, localRequest, NodeType.NodeLocal, reserved));
}
if (rackLocalRequest != null && !rackLocalRequest.GetRelaxLocality())
{
continue;
}
if (rackLocalRequest != null && rackLocalRequest.GetNumContainers() != 0 && (allowedLocality
.Equals(NodeType.RackLocal) || allowedLocality.Equals(NodeType.OffSwitch)))
{
return(AssignContainer(node, rackLocalRequest, NodeType.RackLocal, reserved));
}
ResourceRequest offSwitchRequest = GetResourceRequest(priority, ResourceRequest.Any
);
if (offSwitchRequest != null && !offSwitchRequest.GetRelaxLocality())
{
continue;
}
if (offSwitchRequest != null && offSwitchRequest.GetNumContainers() != 0)
{
if (!HasNodeOrRackLocalRequests(priority) || allowedLocality.Equals(NodeType.OffSwitch
))
{
return(AssignContainer(node, offSwitchRequest, NodeType.OffSwitch, reserved));
}
}
}
}
return(Resources.None());
}
public abstract Org.Apache.Hadoop.Yarn.Api.Records.Resource AssignContainer(FSSchedulerNode
arg1);