ACE-Specialist flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

Recovery Time Objective

The maximum acceptable elapsed time to restore a service after a disruption. It drives standby topology choices such as warm standby versus active-active and is independent of backup frequency.

Recovery Point Objective

The maximum age of data that may be lost, expressed in time. Backup or replication frequency must be at least as often as the RPO demands.

Loose coupling

A design where components communicate through queues or events rather than direct calls, so a failing consumer does not fail the producer and each side scales independently.

Synchronous replication

Writes are acknowledged only after the replica confirms them, producing near-zero data loss at the cost of added write latency and reduced availability during a network partition.

Stateless tier

Application instances hold no session data between requests; state lives in an external cache or database, so any instance can serve any request and the tier scales horizontally.

Readiness probe

A health check reporting whether an instance can serve traffic. Failing instances are removed from load balancer or service endpoints without being restarted.

Blue/green deployment

Two identical environments where traffic switches to the new one after validation and rollback is a switch back, so both stacks must stay warm during the bake period.

Canary release

A small percentage of traffic routes to the new version while error rate and latency are compared to baseline; automated thresholds trigger promotion or rollback.

Expand-contract migration

A backward-compatible schema sequence: add the new column or table, backfill and dual-write, switch reads, then drop the old structure once no code references it.

Immutable artifact

A build output identified by content digest and never modified after publishing. The same digest is promoted across environments, making rollback deterministic.

Feature flag

A runtime toggle enabling or disabling code paths without redeploying. It separates deployment from release and provides an instant kill switch during incidents.

Least privilege

Granting only the permissions a task requires, scoped to specific resources and duration. Roles issuing short-lived credentials replace long-lived access keys.

Envelope encryption

A data key encrypts the payload and a root key encrypts the data key, so rotating the root key does not require re-encrypting the stored data.

Identity federation

Trusting an external identity provider via SAML or OIDC so users authenticate once and receive temporary role credentials, with joiners and leavers managed centrally.

Secrets manager

A managed store holding credentials that rotates them on schedule and issues short-lived access, keeping secrets out of code, images, and state files.

Defense in depth

Layering independent controls such as edge filtering, network segmentation, private data subnets, and least-privilege identity, so one failed control does not expose the system.

Immutable audit log

Control-plane and access logs written to a write-once, separate-account store with a retention period matching the compliance obligation and restricted delete rights.

Error budget

One minus the service level objective: the amount of unreliability permitted in a period. Releases are frozen or slowed once the budget is exhausted.

Four golden signals

Latency, traffic, errors, and saturation. Instrumenting these four per service surfaces most user-visible problems before customers report them.

Burn rate alert

An alert that fires when the error budget is consumed faster than the objective allows, usually combining a short fast window with a longer slow window.

Distributed tracing

Propagating a correlation identifier across service calls so a single request's spans can be reconstructed and the slow hop identified during troubleshooting.

Blameless postmortem

A review documenting timeline, impact, contributing causes, and tracked remediation without attributing fault to individuals, which improves the honesty of future incident reporting.

Remote state locking

A mechanism preventing two concurrent applies from writing the same state file, typically a lock record in a shared backend, avoiding corruption and orphaned resources.

Idempotent automation

An operation that produces the same end state no matter how many times it runs, making retries and recovery from partial failure safe and predictable.

Plan review

A dry-run diff between desired configuration and recorded state showing creates, updates, and destroys. It should be inspected before every apply.

State drift

A difference between real infrastructure and the recorded state, usually caused by manual console changes and detected by a scheduled refresh and plan.

Version pinning

Locking reusable modules and their providers to exact versions so plans stay reproducible and an upstream release cannot silently break production.

Mandatory tagging

A policy requiring keys such as owner, environment, cost center, and data classification at creation time, enforced by policy-as-code rather than after-the-fact cleanup.

Chargeback

Allocating actual cloud spend back to the teams or business units that incurred it, using a consistent account or project hierarchy plus enforced cost tags.

Commitment discount

A promise to consume a steady amount of usage over a term in exchange for a lower rate. It is cost-effective only when baseline usage is predictable.

Storage lifecycle policy

Rules that transition data to colder tiers or expire it based on age, noncurrent version count, and compliance retention rather than storage convenience.

Policy as code

Guardrails expressed as versioned, testable rules evaluated at provisioning time, denying public buckets or unapproved Regions so violations are blocked at the source.