Continuous Delivery Foundations and Value Streams
Continuous Delivery (CD) keeps software in a deployable state at all times, so any commit that passes the pipeline can reach production on demand. The exam tests whether you can separate CD from continuous deployment, read a value stream map, and reason correctly about the four DORA metrics.
Core definitions
- Continuous integration: every commit merges to trunk and triggers build plus automated verification within minutes.
- Continuous delivery: every change passing the pipeline is releasable; the business chooses when to release.
- Continuous deployment: every passing change ships automatically. It is an optional extension of CD, not a synonym.
- Deployment vs release: deployment moves bits into an environment; release exposes behaviour to users. Feature toggles decouple the two.
The pipeline is the unit of design
The deployment pipeline is the only validated path to production. Steps that bypass it produce a route nobody has tested, which is where most change failures originate.
DORA metrics
Four outcomes: lead time for changes, deployment frequency (throughput); change failure rate, time to restore (stability). They are system-level measures, never individual performance targets.
Value stream mapping
Map every step from commit to value in production, recording process time, wait time, and %C/A (percent complete and accurate - the share of handoffs needing no rework). Lead time is dominated by queues, not by hands-on work.
Worked example
A team records: coding 4 h; build and unit tests 20 min work with a 15 min queue; manual regression queue 2 days containing 6 h of tester work; UAT wait 1 day with 2 h work; deploy 30 min. Process time is roughly 13 h, lead time roughly 3.6 days. %C/A at the regression handoff is 40%, so most builds bounce back for rework. The constraint is the regression queue, not developer speed: halving coding time saves 2 h out of 3.6 days, while automating regression removes two days.
Traps that cost marks
- Calling a staging deployment a release, or claiming CD mandates continuous deployment.
- Optimising process time while the longest queue is untouched.
- Using DORA metrics as individual KPIs, which invites gaming.
- Assuming a manual approval gate adds quality; it adds wait time and a handoff, not verification.
- Mapping the value stream at too coarse a level, hiding the queue that is the real constraint.