Smoke & Mirrors: The Primitives of High Availability
summary
Walks through the actual primitives and tradeoffs that make “never down” plausible: DNS, TCP, HTTP, BGP routing with thousands of BGP updates/sec and a traced path of 13 hops to San Francisco, transparent proxies and load balancers, clustering via Paxos/Raft for automatic failover, synchronous vs asynchronous replication under the CAP tradeoffs, and practical monitoring gotchas like disk, entropy, and timeouts. Demonstrates a live pattern for load balancer failover that can achieve zero-request-drop failover and explains when it does and doesn’t work.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 0:53Abstraction in practice
- 7:22Defining fault tolerance
- 9:20Redundancy and proxies
- 13:35Clustering and failover
- 18:25Managing state
- 20:29Monitoring and conclusion
related talks
53mResiliency and Availability Design Patterns for the Cloud
35mDesign patterns for application resiliency
51mReplicated service mesh: hardening systems against failure modes in load balancing
22mFailing Over without Falling Over
55mGoing Resilient
34mDistributed Systems - What Can Go Wrong and How to Get It Right
41mSeven patterns to increase application availability
45m⚡A Resilient Application in a Partially Degraded World ⚡