What changes once a call can be slow, fail, or arrive twice — and the patterns that keep a multi-service system honest.
16 articles
Listed inRetries, Timeouts, and BackoffDistributed Systemson
Why naive retries turn a blip into an outage, how jitter fixes it, and where the deadline should live so a retry is never pointless.
Listed inRetries, Timeouts, and BackoffDistributed Systemson
The Amazon Builders Library on why retries amplify failure, how jitter breaks synchronisation, and where to put the deadline.
Listed inCAP and Consistency ModelsDistributed Systemson
Jepsen maps the whole lattice, from linearizable down to eventual, and shows exactly which anomalies each level still permits.
Listed inConsensus and Leader ElectionDistributed Systemson
The Raft paper. Written specifically to be followable, and still the fastest route to understanding leader election and log replication.
Listed inLoad BalancingDistributed Systemson
L4 versus L7, health checks, connection draining, and consistent hashing for sticky work.
Listed inCAP and Consistency ModelsDistributed Systemson
Linearizable, causal, and eventual consistency — what CAP really says and what it does not.
Listed inRetries, Timeouts, and BackoffDistributed Systemson
Exponential backoff with jitter, deadline propagation, and the retry storm you caused.
Listed inClocks and OrderingDistributed Systemson
Wall clocks lie; logical clocks and vector clocks tell you what happened before what.
Listed inThe FallaciesDistributed Systemson
The eight assumptions about networks that are always false and always assumed anyway.
Listed inCachingDistributed Systemson
Cache-aside, write-through, TTLs, stampedes, and the invalidation everyone quotes about.
Listed inConsensus and Leader ElectionDistributed Systemson
Raft in enough detail to reason about it, quorums, and why split-brain is the failure to fear.
Listed inQueues and BrokersDistributed Systemson
Queues versus logs, consumer groups, ordering guarantees, and dead-letter handling.
Listed inDistributed TransactionsDistributed Systemson
Sagas, compensating actions, and the outbox pattern for writing to a database and a queue at once.
Listed inService DiscoveryDistributed Systemson
How a service finds another one, DNS-based versus registry-based, and the stale-endpoint failure.
Listed inDelivery GuaranteesDistributed Systemson
At-most-once, at-least-once, and why exactly-once is a property of your consumer, not the broker.
Listed inCircuit Breakers and BulkheadsDistributed Systemson
Failing fast on a sick dependency, and isolating resources so one slow call cannot take the process.