5 exercises on the adverbs that define how deployments are released and how systems grow — essential vocabulary for SREs, DevOps engineers, and any developer who writes incident reports or architecture docs.
Key verb–adverb patterns in this set
deploy gradually — incremental rollout; canary and blue-green strategies
scale automatically — controller-driven; no human intervention required
scale horizontally / vertically — the canonical scaling pair (out vs. up)
roll back immediately — urgent incident response; restore previous stable version
provision dynamically — infrastructure allocated at runtime based on demand
0 / 5 completed
1 / 5
A senior engineer explains a production release strategy in a team meeting:
"We never push directly to 100% of users. We deploy ___ — starting with 5% of traffic, monitoring error rates and latency for 30 minutes, then incrementally increasing to 25%, 50%, and 100%."
Which adverb describes a staged, incremental rollout that limits blast radius?
Deploy gradually means rolling out a change in incremental stages rather than all at once — a core risk-mitigation strategy in modern CI/CD.
This is the standard adverb for canary and blue-green deployments:
Canary deploy gradually — route a small slice of traffic to the new version first
Roll out gradually — increase exposure in timed steps
Release gradually — feature flags let you expose to 1%, 10%, 50%, 100%
deploy gradually is contrasted with deploy all at once (big-bang release) or deploy immediately (urgent hotfix). "Automatically" describes the trigger mechanism (CI/CD vs. manual press), not the rollout speed. "Remotely" describes where the engineer is, not how the deployment proceeds. "Manually" is the opposite of automation — also unrelated to pace.
2 / 5
An SRE documents the cluster autoscaler configuration:
"HPA is configured to scale ___ based on CPU utilisation. When the 5-minute average exceeds 70%, the controller adds replicas without any human intervention."
Which adverb is the exact technical term for scaling that requires no operator action?
Scale automatically is the precise adverb when the scaling is triggered and executed by a controller or policy engine — not by a human.
Key collocations in cloud/Kubernetes contexts:
scale automatically — Horizontal Pod Autoscaler (HPA), AWS Auto Scaling Groups, GKE cluster autoscaler
provision automatically — Terraform or Pulumi spinning up resources on demand
restart automatically — liveness probe triggers automatic pod restart
"Dynamically" describes real-time adaptation but does not specifically exclude human involvement — it is a weaker technical claim. "Horizontally" describes the direction of scaling (more instances) — orthogonal to who or what triggers it. "Rapidly" describes speed, not automation. In SLO documentation you will often see: "scales automatically within 90 seconds of the metric threshold being breached."
3 / 5
Architecture RFC states:
"The API tier is stateless and designed to scale ___. Each additional node receives an equal share of traffic from the load balancer. Contrast this with the database tier, which scales ___ by increasing CPU and RAM on a single instance."
Fill both blanks with the correct pair of adverbs:
Scale horizontally (add more nodes) / scale vertically (enlarge one node) is the canonical distinction in distributed systems architecture.
Terminology reference:
Horizontal scaling — also called scaling out; stateless services are ideal candidates; Kubernetes, EC2 Auto Scaling Groups
Vertical scaling — also called scaling up; single-instance DBs (PostgreSQL primary), in-memory caches; has a physical ceiling
Always state the pair together in design documents so the intent is explicit. Common gotcha: developers say "scaling up" when they mean "scaling out" — these are different operations with different cost profiles and availability characteristics. "Outward / upward" are informal but sometimes used — however horizontally / vertically are the standard technical terms used in cloud vendor documentation, AWS Well-Architected Framework, and CNCF materials.
4 / 5
An on-call engineer writes in the incident channel:
"P0 incident confirmed on payment service. Initiating rollback now. Rolling back ___ to v2.14.1 to restore service — investigation of v2.15.0 will continue post-incident."
Which adverb conveys that the rollback must happen without delay due to the severity?
Roll back immediately is the correct collocation for an urgent incident response where restoring the previous stable state must happen without any delay.
In incident management collocations:
roll back immediately — P0/P1 incidents; customer impact is live
roll back gradually — used in canary rollbacks when partial rollout is in progress and you want to drain traffic in steps
revert immediately — git-level equivalent (revert the commit)
escalate immediately — notify senior engineers or incident commander
"Carefully" is contextually wrong in a P0 — the speed is the priority; a careful staged rollback would be used in a lower-severity situation. "Gradually" is contradicted by the context ("P0 incident"). "Safely" describes the risk level, not the urgency — and in an incident, immediately wins over safely in priority.
5 / 5
An infrastructure-as-code PR description reads:
"This Terraform module provisions EC2 instances ___. Rather than creating a fixed pool of servers, the module creates instances on demand when job queues exceed thresholds, and terminates them when the queue is drained."
Which adverb describes resource provisioning that responds to real-time demand?
Provision dynamically means allocating infrastructure resources at runtime in response to actual demand — as opposed to pre-allocating a fixed set of servers.
route dynamically — service meshes (Istio, Linkerd) that update routing tables at runtime
Contrast: provision statically = fixed capacity decided at design time (traditional on-premise model). "Automatically" is close — it excludes human intervention — but "dynamically" additionally carries the meaning of real-time responsiveness. The two can co-occur: "provisions automatically and dynamically based on queue depth."