Deployment
8 pages
-
Continual Learning and Test in Production
concept
Stateless retraining vs stateful training (fine-tuning); four stages of continual learning maturity; champion/challenger pattern; test in production methods (shadow deployment, A/B testing, canary release, interleaving experiments, bandits, contextual bandits); data freshness value; continuous ML challenges (feedback loops, emergency response, stable A/B baselines)
-
Deployment Pipelines
concept
Fitness function automation; stages, fan-in/fan-out; CD vs CDP; cycle time as business metric; zero-downtime deployment (four phases: prepare/drain/apply/start); expand/contract relational schema pattern; trickle-then-batch schemaless migration; canary group evaluation; immutable infrastructure; EDM deployment patterns
-
Feature Flags
concept
Application-layer release control; Hodgson's four toggle types (release/experiment/ops/permission) with different lifecycles; flag lifecycle discipline; default behaviour and cohort routing; kill switches and load-shedding flags; flags as technical debt with a use-by date
-
Four Key Metrics (DORA)
concept
DORA four key metrics (deployment frequency, lead time for changes, change failure rate, time to restore service); empirical basis (4-year, 23K+ respondents, Accelerate); Lean conceptual basis; cluster analysis; no-tradeoff finding; CABs negatively correlated with both metrics; Goodhart's Law warning; pipeline topology variants; instrumentation; MVD; virtuous cycle mechanism; complementary metrics
-
Manageability
concept
Dynamic configuration (config store, runtime re-read); feature flags (progressive rollout, A/B testing, kill switch, deployment-release decoupling); operational triad: monitor → observe → manage
-
Parallel Run
pattern
Correctness verification: both implementations run per request; old result returned; GitHub Scientist; dark launching vs canary vs parallel run; progressive delivery
-
Progressive Delivery
pattern
Release in stages exposing change to controlled subsets; deployment vs release decoupling; canary/blue-green/ring/dark-launch/parallel-run/feature-flag spectrum; SRE order-of-fault model; "roll back first, diagnose second"; prerequisites and anti-patterns
-
Serverless Computing
concept
serverless model (pay-per-invocation, managed autoscaling); cold start by runtime; GAE autoscaling parameters; AWS Lambda (freeze/thaw, provisioned/reserved concurrency, burst limits); parameter study methodology; vendor lock-in