org.apache.ignite.internal.processors.cluster

Class BaselineTopology

java.lang.Object

org.apache.ignite.internal.processors.cluster.BaselineTopology

All Implemented Interfaces:

Serializable

public class BaselineTopology
extends Object
implements Serializable

BaselineTopology represents a set of "database nodes" - nodes responsible for holding persistent-enabled caches and persisting their information to durable storage. Two major features BaselineTopology allows are:

1. Protection from conflicting updates.
2. Cluster auto activation.

Protection from conflicting updates

Consider en example: there is a cluster of three nodes, A, B and C, all nodes store persistent data. Cluster history looks like this: [A,B,C] | \ (1)cluster segmentation, two parts get activated independently | | [A,B] [C] | | (2)updates to both parts After independent updates applied to both parts of cluster at point(2) node C should not be allowed to join [A,B] part. The following algorithm makes sure node C will never join [A,B] part back:

1. When the cluster is initially activated BaselineTopology is created; its property branchingPntHash is calculated based on consistent IDs of all nodes.
2. At point(1) two parts are activated separately; BaselineTopology in each is updated: old branchingPntHash is moved to branchingHistory list; new one is calculated based on new set of nodes.
3. If node C tries to join other part of cluster (e.g. after network connectivity repair) [A,B] cluster checks its branchingPntHash and branchingHistory to see if branching history of C's BaselineTopology has diverged from [A,B]. If divergence is detected [A,B] cluster refuses to let C into topology.

All new nodes joining a cluster should pass validation process before join; detailed algorithm of the process is described below.

Joining node validation algorithm

Node joining a cluster firstly reads its local BaselineTopology from metastore and sends it to the cluster.

1. When BaselineTopology is set (e.g. on first activation) or recreated (e.g. with set-baseline command) its ID on all nodes is incremented by one. So when cluster receives a join request with BaselineTopology it firstly compares joining node BlT ID with local BlT ID. If joining node has a BaselineTopology with ID greater than one in cluster it means that BlT was changed more times there; therefore new node is not allowed to join the cluster.
2. If user manually activates cluster when some of Baseline nodes are offline no new BlT is created. Instead current set of online nodes from BaselineTopology is used to update branchingPntHash property of current BaselineTopology. Old value of the property is moved to branchingHist list. If joining node and local BlT IDs are the same then cluster takes branchingPntHash of joining node and verifies that its local branchingHist contains that hash. If joining node hash is not presented in cluster branching history list it means that joining node was activated independently of currently running cluster; therefore new node is not allowed to join the cluster. If joining node hash is presented in the history, that it is safe to let the node join the cluster.
3. When BaselineTopology is recreated (e.g. with set-baseline command) previous BaselineTopology is moved to BaselineHistory (consult source code of GridClusterStateProcessor for more details). If cluster sees that joining node BlT ID is less than cluster BlT ID it looks up for BaselineHistory item for new node ID. Having this BaselineHistory item cluster verifies that branching history of the item contains branching point hash of joining node (similar check as in the case above with only difference that joining node BlT is compared against BaselineHistory item instead of BaselineTopology). If new node branching point hash is found in the history than node is allowed to join; otherwise it is rejected.

Cluster auto activation

See Also:

Serialized Form

Method Summary

Modifier and Type	Method and Description
Map<String,Object>	attributes(Object consId)
ClusterNode	baselineNode(Object consId)
List<Long>	branchingHistory()
long	branchingPointHash()
static @Nullable BaselineTopology	build(Collection<? extends BaselineNode> nodes, int id)
Map<Short,Object>	compactIdMapping()
Map<Object,Short>	consistentIdMapping()
Set<Object>	consistentIds()
List<ClusterNode>	createBaselineView(Collection<ClusterNode> aliveNodes, @Nullable IgnitePredicate<ClusterNode> nodeFilter)
List<BaselineNode>	currentBaseline()
static boolean	equals(BaselineTopology blt1, BaselineTopology blt2)
boolean	equals(Object o)
int	hashCode()
int	id()
boolean	isNewTopology()
boolean	isSatisfied(@NotNull Collection<ClusterNode> presentedNodes)
Object	resolveConsistentId(Short constId)
Short	resolveShortConsistentId(Object constId)
int	size()
String	toString()

Methods inherited from class java.lang.Object

clone, finalize, getClass, notify, notifyAll, wait, wait, wait

Method Detail

id

public int id()

Returns:

id of this BaselineTopology.

consistentIds

public Set<Object> consistentIds()

Returns:

Set of consistent IDs.

branchingHistory

public List<Long> branchingHistory()

Returns:

Activation history.

compactIdMapping

public Map<Short,Object> compactIdMapping()

Returns:

Compact IDs mapping.

consistentIdMapping

public Map<Object,Short> consistentIdMapping()

Returns:

Consistent IDs mapping.

resolveShortConsistentId

public Short resolveShortConsistentId(Object constId)

Returns:

Short consistent Id.

resolveConsistentId

public Object resolveConsistentId(Short constId)

Returns:

Object consistent Id.

branchingPointHash

public long branchingPointHash()

Returns:

Activation hash.

attributes

public Map<String,Object> attributes(Object consId)

Parameters:

consId - Node consistent ID.

Returns:

Node attributes map.

currentBaseline

public List<BaselineNode> currentBaseline()

baselineNode

public ClusterNode baselineNode(Object consId)

Parameters:

consId - Consistent ID.

Returns:

Baseline node, if present in the baseline, or null if absent.

createBaselineView

public List<ClusterNode> createBaselineView(Collection<ClusterNode> aliveNodes,
                                            @Nullable
                                            @Nullable IgnitePredicate<ClusterNode> nodeFilter)

Parameters:

aliveNodes - Sorted list of currently alive nodes.

nodeFilter - Node filter.

Returns:

Sorted list of baseline topology nodes.

isSatisfied

public boolean isSatisfied(@NotNull
                           @NotNull Collection<ClusterNode> presentedNodes)

Parameters:

presentedNodes - Nodes present in cluster.

Returns:

True if current topology satisfies baseline.

size

public int size()

Returns:

Size of the baseline topology.

equals

public boolean equals(Object o)

Overrides:

equals in class Object

hashCode

public int hashCode()

Overrides:

hashCode in class Object

equals

public static boolean equals(BaselineTopology blt1,
                             BaselineTopology blt2)

Parameters:

blt1 - Baseline topology instance.

blt2 - Baseline topology instance.

Returns:

True if equals.

isNewTopology

public boolean isNewTopology()

Returns:

true If it is new baseline topology.

build

@Nullable
public static @Nullable BaselineTopology build(Collection<? extends BaselineNode> nodes,
                                                         int id)

Parameters:

nodes - Nodes.

id - ID of BaselineTopology to build.

Returns:

Baseline topology consisting of given nodes.

toString

public String toString()

Overrides:

toString in class Object