Technical terms, cross-domain

Glossary

Terms that recur across incident analyses on this site. When an incident page uses jargon, the first occurrence links here.

JUMP TOA B C D F I M N R S T

Alert fatigue: The progressive ignoring of alerts due to false-positive volume. High-frequency false alarms train operators to dismiss signals, including the one that mattered.
Authority gradient: A power differential in a decision-making context that suppresses correction by junior members. Extensively studied in aviation; applies anywhere hierarchy meets time pressure.
Backpressure: A feedback signal from a slow consumer to a fast producer to reduce its rate. Systems without backpressure convert temporary slowness into total collapse.
Blast radius: The scope of impact from a failure, measured in affected parties. One of our six classification dimensions.
BGP (Border Gateway Protocol): The routing protocol that connects autonomous systems on the internet. BGP misconfigurations, leaks, and withdrawals appear in a large fraction of our infrastructure incidents.
Cascading failure: A failure whose primary effect is triggering additional failures downstream. Distinguished from a "big bang" failure by the propagation pattern.
Circuit breaker: A mechanism that halts activity once a failure threshold is exceeded. Originated in electrical engineering; now a broad pattern in software, finance, and trading systems.
Control plane: The system that manages configuration and state of a service, as distinct from the data plane that serves requests. Control plane failures often cause slow-burning outages.
Defense in depth: Multiple independent safety layers such that no single failure is catastrophic. The Swiss Cheese Model describes the failure mode.
DNS (Domain Name System): The service that translates hostnames to addresses. DNS failures appear in a disproportionate share of catastrophic cloud incidents because many other services depend on it.
Fishbone diagram: Also Ishikawa diagram. A cause-categorization tool showing contributing factors as branches off a central "spine." See the methodology page.
Five Whys: An iterative questioning technique to descend past the first plausible cause. Originated at Toyota.
Ironic automation: Lisanne Bainbridge's observation that automation increases the skill required of the human operator, while decreasing the human's practice. When automation fails, the human is least prepared to take over.
Metastable failure: A failure state that, once entered, prevents recovery through normal means. The system is stable in the failed state and stable in the healthy state, with a barrier between them that routine recovery cannot cross.
Normalization of deviance: The progressive acceptance of behavior outside the defined safety envelope when it fails to produce immediate consequence. Diane Vaughan's term, originating in her Challenger analysis.
NXDOMAIN: A DNS response indicating the queried name does not exist. When a valid service suddenly returns NXDOMAIN, downstream systems that treated the name as permanent crash.
Race condition: A software defect where program behavior depends on the relative timing of concurrent operations. By nature, hard to reproduce; by corollary, hard to test for.
Retry storm: A cascade where failing requests are retried faster than they can succeed, concentrating load on the already-failing component and preventing recovery.
Single point of failure (SPOF): A component whose failure causes the whole system to fail. Present in more systems than their owners believe.
Swiss Cheese Model: Reason's model: each defensive layer has holes; failures happen when holes line up. See the methodology page.
Thundering herd: A pattern where many clients wake simultaneously to retry a resource, overwhelming it. Related to retry storms, distinguished by the trigger being time (cache expiry, scheduler tick) rather than a prior failure.
Tight coupling: A design where components depend strongly on each other's behavior and timing. Tightly coupled systems propagate failure; loosely coupled systems contain it.