org.apache.cassandra.service.reads.repair

Interface ReadRepair<E extends Endpoints<E>,P extends ReplicaPlan.ForRead<E,P>>

All Known Implementing Classes:

AbstractReadRepair, BlockingReadRepair, NoopReadRepair, ReadOnlyReadRepair

public interface ReadRepair<E extends Endpoints<E>,P extends ReplicaPlan.ForRead<E,P>>

Nested Class Summary

Nested Classes

Modifier and Type	Interface and Description
static interface	ReadRepair.Factory

Method Summary

Modifier and Type	Method and Description
void	awaitReads() Block on the reads (or timeout) sent out in startRepair(org.apache.cassandra.service.reads.DigestResolver<E, P>, java.util.function.Consumer<org.apache.cassandra.db.partitions.PartitionIterator>)
void	awaitWrites() Block on any mutations (or timeout) we sent out to repair replicas in repairPartition(org.apache.cassandra.db.DecoratedKey, java.util.Map<org.apache.cassandra.locator.Replica, org.apache.cassandra.db.Mutation>, org.apache.cassandra.locator.ReplicaPlan.ForWrite)
static <E extends Endpoints<E>,P extends ReplicaPlan.ForRead<E,P>> ReadRepair<E,P>	create(ReadCommand command, ReplicaPlan.Shared<E,P> replicaPlan, Dispatcher.RequestTime requestTime)
UnfilteredPartitionIterators.MergeListener	getMergeListener(P replicaPlan) Used by DataResolver to generate corrections as the partition iterator is consumed
void	maybeSendAdditionalReads() if it looks like we might not receive data requests from everyone in time, send additional requests to additional replicas not contacted in the initial full data read.
void	maybeSendAdditionalWrites() If it looks like we might not receive acks for all the repair mutations we sent out, combine all the unacked mutations and send them to the minority of nodes not involved in the read repair data read / write cycle.
void	repairPartition(DecoratedKey partitionKey, java.util.Map<Replica,Mutation> mutations, ReplicaPlan.ForWrite writePlan) Repairs a partition _after_ receiving data responses.
void	startRepair(DigestResolver<E,P> digestResolver, java.util.function.Consumer<PartitionIterator> resultConsumer) Called when the digests from the initial read don't match.

Method Detail

create

static <E extends Endpoints<E>,P extends ReplicaPlan.ForRead<E,P>> ReadRepair<E,P> create(ReadCommand command,
                                                                                          ReplicaPlan.Shared<E,P> replicaPlan,
                                                                                          Dispatcher.RequestTime requestTime)

getMergeListener

UnfilteredPartitionIterators.MergeListener getMergeListener(P replicaPlan)

Used by DataResolver to generate corrections as the partition iterator is consumed

startRepair

void startRepair(DigestResolver<E,P> digestResolver,
                 java.util.function.Consumer<PartitionIterator> resultConsumer)

Called when the digests from the initial read don't match. Reads may block on the repair started by this method.

Parameters:

digestResolver - supplied so we can get the original data response

resultConsumer - hook for the repair to set it's result on completion

awaitReads

void awaitReads()
         throws ReadTimeoutException

Block on the reads (or timeout) sent out in startRepair(org.apache.cassandra.service.reads.DigestResolver<E, P>, java.util.function.Consumer<org.apache.cassandra.db.partitions.PartitionIterator>)

Throws:

ReadTimeoutException

maybeSendAdditionalReads

void maybeSendAdditionalReads()

if it looks like we might not receive data requests from everyone in time, send additional requests to additional replicas not contacted in the initial full data read. If the collection of nodes that end up responding in time end up agreeing on the data, and we don't consider the response from the disagreeing replica that triggered the read repair, that's ok, since the disagreeing data would not have been successfully written and won't be included in the response the the client, preserving the expectation of monotonic quorum reads

maybeSendAdditionalWrites

void maybeSendAdditionalWrites()

If it looks like we might not receive acks for all the repair mutations we sent out, combine all the unacked mutations and send them to the minority of nodes not involved in the read repair data read / write cycle. We will accept acks from them in lieu of acks from the initial mutations sent out, so long as we receive the same number of acks as repair mutations transmitted. This prevents misbehaving nodes from killing a quorum read, while continuing to guarantee monotonic quorum reads

awaitWrites

void awaitWrites()

Block on any mutations (or timeout) we sent out to repair replicas in repairPartition(org.apache.cassandra.db.DecoratedKey, java.util.Map<org.apache.cassandra.locator.Replica, org.apache.cassandra.db.Mutation>, org.apache.cassandra.locator.ReplicaPlan.ForWrite)

repairPartition

void repairPartition(DecoratedKey partitionKey,
                     java.util.Map<Replica,Mutation> mutations,
                     ReplicaPlan.ForWrite writePlan)

Repairs a partition _after_ receiving data responses. This method receives replica list, since we will block repair only on the replicas that have responded.