Continuing through failure.
WHAT A CLUSTER IS
Several machines cooperating so that a service survives the loss of one.
WHAT ACTIVE-PASSIVE MEANS
One node serving, another standing by.
WHAT ACTIVE-ACTIVE MEANS
Several nodes serving simultaneously.
WHAT THE SECOND ADDITIONALLY PROVIDES
Capacity, and confidence that the other nodes work.
WHY THAT CONFIDENCE MATTERS
A passive node that has never served may not work when needed.
WHAT CLUSTERING REQUIRES
Failure detection An agreed decision about who is active A mechanism for taking over the service address Shared or replicated data
WHAT SPLIT-BRAIN IS
Nodes losing contact and each believing it should be active.
WHAT THAT CAUSES
Both writing, and data diverging irreconcilably.
WHAT PREVENTS IT
Quorum: a majority required to act
Fencing: forcibly isolating a node believed failed
WHY QUORUM REQUIRES AN ODD NUMBER
Two nodes cannot establish a majority when they disagree.
WHAT TO ADD TO A TWO-NODE CLUSTER
A third arbitrating participant.
WHAT TO TEST
Failover, deliberately, under load, repeatedly.
