Distributed Systems

Nothing here yet.

Core: start here

• CAP TheoremDuring a network partition, you must choose consistency or availability.
• Eventual ConsistencyReplicas converge once updates stop, but reads may be stale in the meantime.

Core: start here

• Consistency ModelsLinearizable, sequential, causal, eventual: what readers are allowed to see.
• Distributed SystemMany machines cooperating over an unreliable network.
• Fallacies of Distributed ComputingEight false assumptions, starting with "the network is reliable".
• Saga PatternA chain of local transactions with compensating actions instead of a distributed transaction.
• Transactional OutboxSaving events in the same database transaction, then publishing them reliably.

Nothing here yet.

• Byzantine FaultNodes that behave arbitrarily or maliciously.
• Fencing TokenA counter that stops stale lock holders from writing.
• Hedged RequestsSending a duplicate request when the first is slow, to cut tail latency.
• LinearizabilityOperations appear to happen instantly, in real-time order.
• Logical ClocksLamport and vector clocks for ordering events without real time.
• PaxosThe classic, notoriously hard-to-understand consensus algorithm.
• Total Order BroadcastDelivering the same messages in the same order to every node.

Nothing here yet.

Core: start here

• Change Data CaptureStreaming database changes out as events.

• Choreography vs OrchestrationServices reacting to events vs a central coordinator directing them.
• Clock SkewMachines' clocks disagreeing, which breaks time-based ordering.
• ConsensusGetting nodes to agree on a value despite failures.
• Consistency ModelsLinearizable, sequential, causal, eventual: what readers are allowed to see.
• Data Consistency Across ServicesKeeping data correct when each service owns its own database.
• Database per ServiceEach microservice owning its data exclusively.
• Distributed LockMutual exclusion across machines, and its pitfalls.
• Distributed SystemMany machines cooperating over an unreliable network.
• Fallacies of Distributed ComputingEight false assumptions, starting with "the network is reliable".
• Gossip ProtocolNodes spreading information by talking to random peers.
• HeartbeatPeriodic "I'm alive" signals used to detect failed nodes.
• Idempotency in Distributed SystemsMaking retries safe when you can't know whether the first attempt worked.
• Leader ElectionChoosing one node to coordinate the others.
• LeaseA lock or role that expires automatically unless renewed.
• PACELCExtends CAP: even without partitions, you trade latency against consistency.
• QuorumRequiring a majority of nodes to agree on reads or writes.
• RaftAn understandable consensus algorithm, used in etcd.
• Request CoalescingCollapsing identical concurrent requests into one.
• Saga PatternA chain of local transactions with compensating actions instead of a distributed transaction.
• Service DiscoveryHow services find each other's addresses.
• Service MeshInfrastructure handling service-to-service traffic, retries and mTLS.
• SidecarA helper process deployed alongside each service.
• Split BrainTwo nodes both believing they're the leader.
• Strong ConsistencyEvery read sees the latest write.
• Transactional OutboxSaving events in the same database transaction, then publishing them reliably.
• Two Generals' ProblemWhy agreement over an unreliable link can't be guaranteed.

Nothing here yet.

• Byzantine FaultNodes that behave arbitrarily or maliciously.
• Fencing TokenA counter that stops stale lock holders from writing.
• Hedged RequestsSending a duplicate request when the first is slow, to cut tail latency.
• LinearizabilityOperations appear to happen instantly, in real-time order.
• Logical ClocksLamport and vector clocks for ordering events without real time.
• PaxosThe classic, notoriously hard-to-understand consensus algorithm.
• Total Order BroadcastDelivering the same messages in the same order to every node.

Nothing here yet.

• CAP TheoremDuring a network partition, you must choose consistency or availability.
• Eventual ConsistencyReplicas converge once updates stop, but reads may be stale in the meantime.
• IdempotenceDoing something twice has the same effect as doing it once.

• Consistency ModelsLinearizable, sequential, causal, eventual: what readers are allowed to see.
• Distributed SystemMany machines cooperating over an unreliable network.
• Fallacies of Distributed ComputingEight false assumptions, starting with "the network is reliable".
• Idempotency in Distributed SystemsMaking retries safe when you can't know whether the first attempt worked.
• Request CoalescingCollapsing identical concurrent requests into one.
• Strong ConsistencyEvery read sees the latest write.

Nothing here yet.

Nothing here yet.