CCOA (IJCAHPO) key terms

The division of security duties between a cloud provider and its customer. The provider secures the physical facility, hypervisor, and managed service internals, while the customer secures its own data, identities, configurations, and access policies.

Six design dimensions used to evaluate cloud workloads: operational excellence, security, reliability, performance efficiency, cost optimization, and sustainability. Reviewing a workload against each pillar surfaces tradeoffs and improvement opportunities before they become incidents.

An isolated data center group within a cloud region, with independent power, cooling, and networking. Deploying replicas across multiple zones protects a workload from a single-site failure without the latency of cross-region traffic.

A geographic area containing multiple availability zones, chosen for latency, data residency, and service availability. Workloads spanning regions gain resilience against large-scale outages but must handle replication lag and cross-region data transfer costs.

A legal or policy requirement that certain data remain stored and processed within specified geographic boundaries. It constrains region selection, backup locations, and replication targets, so architects must encode it as an explicit design constraint.

A design principle where components interact through stable interfaces such as queues, events, or APIs rather than direct dependencies. Failures and scaling changes in one component then propagate less to others, improving overall resilience.

A service that keeps no session data between requests, storing state externally in a database, cache, or object store. Any instance can handle any request, which simplifies horizontal scaling, rolling deployments, and failure recovery.

Adding or removing identical instances of a component to match demand, as opposed to resizing a single instance. It pairs naturally with stateless services and autoscaling policies that react to metrics like queue depth or CPU.

Rules that determine when a platform adds or removes capacity, based on metrics, schedules, or predicted load. Good policies include cooldown periods and sensible minimums and maximums to avoid thrashing and runaway spend.

Matching provisioned compute, memory, and storage to actual measured demand rather than peak guesses. It is typically the largest lever for reducing cloud waste and requires usage telemetry before any instance class change.

A consistent scheme of key-value metadata applied to cloud resources to attribute cost, ownership, environment, and compliance scope. Without enforced tags, chargeback reports and automated policy checks become unreliable or impossible.

Managing cloud resources through versioned, machine-readable definition files instead of manual console actions. It enables peer review, repeatable environments, drift detection, and reliable rollback when changes are applied through a pipeline.

A record maintained by an IaC tool mapping declared resources to real cloud identifiers and attributes. It must be stored remotely with locking so concurrent pipeline runs cannot corrupt it or duplicate resources.

The process of comparing deployed infrastructure against its declared configuration to find manual or out-of-band changes. Detecting drift regularly prevents the next automated apply from silently reverting or conflicting with emergency fixes.

A practice where servers or containers are replaced rather than patched in place. Each deployment produces a fresh artifact from a known image, eliminating configuration snowflakes and making rollbacks a matter of redeploying a prior version.

Tools and practices that enforce a desired state on servers or systems, such as installed packages, file contents, and running services. It is often used for mutable hosts, while image-building pipelines handle immutable workloads.

A property where repeating an operation produces the same end state as running it once. It is essential for safe retries in automation, configuration runs, and API calls that may be delivered more than once.

A defined sequence of automated tasks with dependencies, branching, retries, and error handling, executed by a scheduler or workflow engine. It coordinates multi-step operations such as provisioning, data loading, and deployment approvals.

A packaged, layered filesystem plus metadata that provides everything needed to run an application. Images are built once, tagged immutably, scanned for vulnerabilities, and pulled by any runtime that supports the same format.

A service that stores and distributes container images, often with access control, replication, and vulnerability scanning. Pipelines push versioned images there and clusters pull from it, making it a critical supply-chain component.

The smallest deployable unit in Kubernetes, wrapping one or more containers that share a network namespace and storage volumes. Pods are ephemeral; controllers recreate them rather than repairing them in place.

A Kubernetes workload resource that manages replica sets to achieve a declared pod count and supports rolling updates. It records revision history, allowing a rollout to be paused or reverted when health checks fail.

The mechanism by which components locate each other's network endpoints as instances scale or move. In Kubernetes, Services and DNS names provide stable addresses in front of constantly changing pod IPs.

Kubernetes settings declaring the CPU and memory a container needs and the ceiling it may consume. Requests drive scheduling decisions, while limits enforce caps that protect node stability but can cause throttling or eviction.

Defining build, test, and deploy stages in a version-controlled file rather than through a UI. It makes CI/CD behavior reviewable, reusable across repositories, and reproducible when recreating tooling.

A release pattern where a new version is deployed to a parallel environment and traffic is switched over once verified. Rollback is fast because the previous environment remains intact, though it doubles infrastructure during the switch.

Gradually routing a small share of traffic to a new version while monitoring error rates and latency before widening exposure. It limits blast radius but requires comparable metrics and automated analysis to be trustworthy.

Restoring a previously known-good version of code or infrastructure after a failed change. Effective rollback depends on immutable artifacts, backward-compatible data migrations, and a rehearsed procedure rather than improvisation.

A pre-built, hardened machine or container template used as the starting point for all deployments. It encodes patching, agents, and baseline configuration once, reducing variance and speeding provisioning.

The ability to infer a system's internal state from its external outputs, mainly metrics, logs, and traces. It differs from monitoring by supporting open-ended questions about failures that were never explicitly anticipated.

Recording a single request's path across services using propagated trace and span identifiers. Traces reveal latency contributions and failure points in microservice architectures where logs alone cannot show causal order.

Emitting log records as machine-parseable fields such as JSON with consistent keys, instead of free text. It enables filtering, aggregation, and correlation with trace identifiers at scale.

A measurable target for reliability or performance, such as 99.9 percent of requests succeeding within 300 milliseconds over 30 days. SLOs guide alerting and release decisions more usefully than raw infrastructure metrics.

The amount of unreliability an SLO permits over a period, derived by subtracting the target from 100 percent. Teams spend it on releases and experiments, and exhausted budgets typically trigger a change freeze.

The policy defining how long log and metric data is kept and at what storage tier. It balances investigation needs and compliance requirements against storage cost, often using tiered or sampled retention.

Granting identities only the permissions required for their tasks, for the shortest practical duration. It limits blast radius from compromised credentials and is implemented through scoped roles, conditions, and just-in-time elevation.

Protecting stored data by encrypting volumes, buckets, and databases with managed or customer-provided keys. Key management and rotation policies matter as much as the cipher, since key compromise defeats the protection.

Expressing security and compliance rules as versioned, automatically evaluated definitions rather than manual review. It enables pre-deployment blocking of misconfigured resources and continuous auditing of deployed infrastructure.

Automatically inspecting container images for known vulnerabilities in operating system packages and libraries. Running scans in the pipeline and on a schedule catches newly disclosed flaws in images already in production.

A documented, rehearsed procedure for restoring services after a major failure, including roles, dependencies, and validation steps. Untested runbooks are unreliable, so periodic game days should exercise them.

A durable buffer that decouples producers from consumers, absorbing traffic spikes and allowing retries after consumer failure. It requires attention to ordering, duplicate delivery, and dead-letter handling.