Progressive Delivery Engineer — Interview Questions

Option B is strongest: it uses the coal mine analogy origin but focuses on the operational reality, specifies exactly what is being watched (error rate, latency, business metrics with a concrete PM-relevant example), provides the exact traffic progression (5% → 20% → 50% → 100%), gives the rollback time ('under a minute'), and ends with a four-word framing (real production, real users, real metrics, reversible) that a PM can actually use to understand the value. The business metric example (checkout conversion) makes it meaningful to a PM, not just a deployment technique. Canary deployment vocabulary: Canary deployment — a progressive rollout strategy where a new version is first released to a small percentage of traffic. Canary weight — the percentage of traffic currently routed to the canary version. Traffic shifting — the mechanism by which traffic percentage is moved between versions. Rollback — the process of shifting all traffic back to the previous stable version. Error rate — the percentage of requests that result in an error response. Options C and D are accurate but lack the business metric example and the PM-framing conclusion.

Option B is strongest: it distinguishes automated from manual triggers (and explains why each category exists), provides concrete threshold examples with units (0.5% vs 0.1% baseline = 5x), introduces the rolling baseline methodology and explains exactly why absolute thresholds are inadequate (variable baseline across the day), specifies the 7-day same-time-of-day baseline calculation, and provides the team conversation script that quantifies the value of automated triggers as 'maximum exposure window shrinks to 3 minutes'. That conversation script is what makes the answer interview-ready. Progressive delivery vocabulary: Rollback trigger — an automated or manual signal that initiates a traffic shift back to the previous version. Rolling baseline — a dynamic reference metric calculated over a recent historical window. p99 latency — the latency value below which 99% of requests fall; used as a worst-case latency indicator. Exposure window — the time period during which users are exposed to a potentially broken canary version. MTTR (Mean Time to Recovery) — the average time to restore service after a failure. Options C and D are accurate but lack the rolling baseline explanation and the team conversation script.

Option B is strongest: it provides the exact status update sentence in full, explains the role of each sentence (not just what it says), and — critically — explains why 'rollback is armed' is a more confidence-giving phrase than 'we will roll back if needed' (armed = automatic, the other = manual). The explanation of why the next decision time and rollback state are required in a status update is the key insight: those elements allow stakeholders to disengage from monitoring rather than continuing to watch. That is the communication goal of a status update in an ongoing deployment. Deployment status vocabulary: Canary weight — the current percentage of traffic routed to the canary version. Baseline comparison — presenting a metric alongside its historical reference value to give context for whether the current value is normal. Armed rollback — an automated rollback trigger that is configured and active, requiring no human action to execute. Advancement — the act of increasing canary weight to the next stage. Decision point — the scheduled time or metric threshold at which a deployment will either advance or roll back. Options C and D are accurate but lack the explanation of why each sentence exists and the 'armed vs. manual' distinction.

Option B is strongest: it opens with the most important insight for a new team (flags must be removed — they are not permanent), provides all five stages with descriptive names and the purpose of each stage, introduces the concept of dark launch and explicitly states why it matters ('decouples deployment from release' — a core progressive delivery principle), names the production risk of skipping stage 5 (flag debt) with a concrete explanation of why it is dangerous (unknown interactions with new flags), and ends with the team framing that turns the principle into an actionable habit (death date on the creation ticket). That habit prescription is what makes the answer practical, not just conceptual. Feature flag vocabulary: Feature flag (feature toggle) — a configuration mechanism that controls whether a code path is active without a new deployment. Dark launch — deploying code to production with the feature disabled; separates deployment from release. Flag debt — the accumulation of unremoved feature flags that create code complexity and unpredictable interactions. Targeted rollout — enabling a feature for specific user segments before general availability. Death date — the planned date for removing a feature flag and its associated old code path. Options C and D are accurate but lack the explanation of why dark launch matters and the flag debt consequence detail.

Option B is strongest: it opens with the correct framing question (failure exposure — how much impact is acceptable), provides symmetric trade-off analysis for both strategies (infrastructure cost, exposure scope, rollback speed), introduces the critical 'suitable for' guidance that distinguishes which risk type each strategy addresses, and ends with a real-world usage pattern ('we run canary for app deployments and blue-green for infrastructure') that shows the strategies are complementary, not competing. The 'they solve different problems' closing is the key insight — most teams frame it as a choice when in practice both are used for different deployment types. Deployment strategy vocabulary: Blue-green deployment — a deployment strategy using two identical environments, with instant full traffic switchover between them. Canary deployment — a deployment strategy with incremental traffic exposure to a new version. Traffic splitting — the mechanism that routes a defined percentage of requests to each version. Exposure window — the period during which users are served by a potentially problematic version. Infrastructure configuration change — a change to environment-level settings (server configuration, network rules) as opposed to application code. Options C and D are accurate but lack the 'suitable for' risk-type guidance and the complementary-not-competing framing.