Skip to content

Cloud + AI Glossary

200+ terms across cloud, AI/ML, DevOps, networking, and security - in plain English.

Use Ctrl+F. This is a reference doc, not something to read end-to-end.

Categories: - Cloud Fundamentals - AWS - Microsoft Azure - Google Cloud (GCP) - Networking - Security & Identity - DevOps & Infrastructure as Code - Containers & Kubernetes - Data & Databases - AI / Machine Learning - LLMs & Generative AI - Observability


Cloud Fundamentals

Availability Zone (AZ) - A physically separate datacenter (or cluster) inside a cloud region. AZs in the same region are connected by low-latency links. You spread workloads across AZs to survive a single datacenter failure.

Bare metal - A physical server you rent, no hypervisor, no virtualization layer. Used when you need maximum performance or specific hardware.

Cloud computing - On-demand delivery of compute, storage, networking, and software over the internet, billed by usage instead of owned outright.

Cold storage - Cheap, slow storage for data you rarely access (archives, backups, compliance retention). Examples: S3 Glacier, Azure Archive Blob, GCS Coldline.

Compute - Anything that runs code: VMs, containers, serverless functions, batch jobs.

Edge / Edge computing - Compute that runs close to the user (CDN points of presence, local devices) instead of in a central region. Lower latency for end users.

Elasticity - The ability to automatically scale resources up and down based on demand. Distinguished from "scalability" (which just means "can grow").

Hybrid cloud - Mixing on-premises infrastructure with public cloud, with networking and identity that span both.

Hypervisor - The software layer that runs VMs on physical hardware. Examples: KVM, Xen, Hyper-V, VMware ESXi.

IaaS (Infrastructure as a Service) - You rent raw VMs, networks, and storage. You manage the OS and everything above it. Examples: EC2, Azure VMs, GCE.

Multi-cloud - Using two or more public clouds (AWS + GCP, Azure + AWS, etc.) for workloads. Different from hybrid (which mixes on-prem and cloud).

Multi-tenant - A single instance of software/infrastructure serves many customers, with isolation between them. Most SaaS is multi-tenant.

On-premises (on-prem) - Infrastructure you own and run in your own datacenter (or office closet).

PaaS (Platform as a Service) - Provider manages OS, runtime, scaling. You deploy code. Examples: Heroku, Elastic Beanstalk, App Engine, Render.

Private cloud - Cloud-style infrastructure dedicated to a single organization, often on-prem or in a colo.

Public cloud - Multi-tenant cloud run by a hyperscaler (AWS, Azure, GCP, Oracle, IBM).

Region - A geographic area (e.g., us-east-1, eu-west-2) containing multiple AZs. Latency between regions is high; you typically deploy per region.

Reserved Instance / Reservation / Committed Use Discount - Pre-pay or commit to using a resource for 1-3 years in exchange for a big discount.

SaaS (Software as a Service) - You use software over the internet; provider runs everything. Examples: Gmail, Salesforce, Slack, Notion.

Serverless - You write code; the platform runs it on demand and bills per execution. No servers for you to manage. Examples: Lambda, Cloud Functions, Cloud Run, Azure Functions.

Shared responsibility model - The cloud provider secures the cloud; you secure what you put in the cloud. Where the line is depends on the service (more services managed = less for you).

Spot / Preemptible Instance - A cheap VM the provider can reclaim with little notice. Good for fault-tolerant batch work, bad for stateful production.

Tenancy - Whether your resources share hardware with others (shared tenancy) or run on dedicated hardware (dedicated tenancy).

Virtual machine (VM) - A simulated computer running on shared physical hardware via a hypervisor.

Workload - Any application, service, or batch job running in the cloud. Cloud architects love this word.


AWS

ALB (Application Load Balancer) - Layer 7 (HTTP/HTTPS) load balancer. Routes by path, host, headers.

AMI (Amazon Machine Image) - A snapshot/template of a VM disk used to launch EC2 instances.

Athena - Serverless SQL queries over data in S3.

AWS - Amazon Web Services. The largest public cloud.

CloudFormation - AWS's native infrastructure-as-code tool. JSON or YAML templates.

CloudFront - AWS's content delivery network (CDN).

CloudTrail - Logs every API call made in your AWS account. Audit trail.

CloudWatch - Metrics, logs, and alarms.

DynamoDB - Managed key-value / document NoSQL database. Single-digit-millisecond reads at any scale.

EBS (Elastic Block Store) - Block storage volumes attached to EC2 instances. Like a virtual hard drive.

EC2 (Elastic Compute Cloud) - AWS's VM service.

ECS (Elastic Container Service) - AWS's container orchestrator. Simpler than EKS; AWS-flavored.

EKS (Elastic Kubernetes Service) - Managed Kubernetes on AWS.

ElastiCache - Managed Redis or Memcached.

Fargate - Serverless compute for containers. Run ECS or EKS pods without managing nodes.

IAM (Identity and Access Management) - AWS's permissions system. Users, roles, policies.

Kinesis - Streaming data service. Kinesis Data Streams = Kafka-like; Firehose = ingest to S3/Redshift; Analytics = SQL on streams.

Lambda - AWS's serverless function service.

NLB (Network Load Balancer) - Layer 4 (TCP/UDP) load balancer. Faster, fewer features than ALB.

RDS (Relational Database Service) - Managed Postgres, MySQL, MariaDB, SQL Server, Oracle.

Route 53 - AWS's DNS service.

S3 (Simple Storage Service) - Object storage. The most fundamental AWS service.

SNS (Simple Notification Service) - Pub/sub messaging.

SQS (Simple Queue Service) - Message queue.

VPC (Virtual Private Cloud) - Your own isolated network in AWS.


Microsoft Azure

AKS (Azure Kubernetes Service) - Managed Kubernetes on Azure.

App Service - PaaS for web apps and APIs. Easier than Kubernetes; less flexible.

ARM (Azure Resource Manager) - The deployment and management layer for Azure. Templates are like CloudFormation.

Azure - Microsoft's public cloud.

Azure AD / Entra ID - Microsoft's cloud identity provider. Used for SSO, B2B, B2C, conditional access.

Azure DevOps - CI/CD, repos, boards, artifacts. The Azure-flavored GitHub.

Azure Functions - Serverless functions on Azure.

Bicep - A friendlier DSL that compiles to ARM templates.

Blob Storage - Azure's object storage (equivalent to S3).

Cosmos DB - Globally-distributed multi-model database. NoSQL, document, graph, key-value, all in one.

Event Grid / Event Hubs / Service Bus - Azure's three messaging services. Event Grid = events, Event Hubs = high-throughput streaming, Service Bus = enterprise queues.

Resource Group - A logical container for related Azure resources. Permissions and lifecycle apply to the group.

Subscription - A billing/permission boundary in Azure. You usually have several (dev, prod, etc.) under a tenant.

Synapse Analytics - Azure's data warehouse + analytics platform.

Tenant - The top-level identity boundary in Microsoft 365 / Azure (one per organization).

Virtual Network (VNet) - Azure's equivalent of a VPC.


Google Cloud (GCP)

BigQuery - GCP's serverless data warehouse. Pay per query and storage.

Cloud Functions - GCP's FaaS (function-as-a-service).

Cloud Run - Run containers serverlessly on GCP. Scales to zero.

Cloud SQL - Managed Postgres, MySQL, SQL Server.

Cloud Storage (GCS) - GCP's object storage (equivalent to S3 / Blob).

Dataflow - Managed Apache Beam (stream + batch processing).

Firestore - Serverless NoSQL document DB. Real-time client SDKs.

GCE (Google Compute Engine) - GCP's VM service.

GCP - Google Cloud Platform.

GKE (Google Kubernetes Engine) - Managed Kubernetes on GCP. The original; Google invented Kubernetes.

IAM (GCP) - Same name as AWS, different model. Roles attach to identities and apply to resources.

Pub/Sub - GCP's globally-distributed pub/sub messaging.

Spanner - Globally-consistent relational database. Distributed but ACID.

VPC (GCP) - GCP's virtual network. Notably, GCP VPCs are global (not regional like AWS).

Vertex AI - GCP's unified ML platform. Training, deployment, AutoML, foundation models.


Networking

API Gateway - A managed front door for APIs. Handles auth, rate limiting, routing, transforms.

ASN (Autonomous System Number) - The unique ID for a network on the internet (used in BGP).

BGP (Border Gateway Protocol) - The protocol the internet uses to route between networks.

CDN (Content Delivery Network) - Edge cache for static (and sometimes dynamic) content. CloudFront, Cloud CDN, Azure CDN, Cloudflare, Fastly.

CIDR (Classless Inter-Domain Routing) - The notation for IP ranges (e.g., 10.0.0.0/16). The number after the slash is how many bits are fixed.

DNS (Domain Name System) - Translates names (example.com) to IPs (93.184.216.34).

Egress - Traffic leaving your network/cloud. Often the part that costs money.

Firewall - Filters traffic by rules. Cloud-native versions are usually called security groups, NACLs, or firewall policies.

HTTP/HTTPS - The protocols the web runs on. HTTPS = HTTP over TLS.

Ingress - Traffic coming into your network. In Kubernetes, also a resource type that exposes services to the internet.

Latency - The time it takes a request to round-trip. Measured in milliseconds.

Load balancer - Distributes incoming traffic across many backend servers.

NAT (Network Address Translation) - Lets private IPs reach the internet by sharing a public IP.

OSI model - 7-layer abstraction for networking. Layer 4 = TCP/UDP, Layer 7 = HTTP. Not literally how things work, but useful vocabulary.

Peering - A direct network connection between two networks (VPCs, datacenters, ISPs).

Private link / Private endpoint - A way to access a managed service over a private IP without traversing the public internet.

Public IP - An IP routable on the internet. Public IPs are scarce (IPv4) and often cost money.

Reverse proxy - A server that takes inbound requests and forwards them to backend services. Nginx, HAProxy, Envoy.

Route table - Defines where packets go based on destination IP. Each subnet has one.

Security group - A stateful firewall attached to a VM/pod/service. Rules are usually allow-only.

Subnet - A subdivision of a VPC into a smaller IP range, typically scoped to one AZ.

TCP/UDP - Transport protocols. TCP = reliable, ordered, connection-based. UDP = best-effort, fast, no connection.

TLS (Transport Layer Security) - The encryption protocol that powers HTTPS. SSL is the deprecated predecessor.

VPN (Virtual Private Network) - Encrypted tunnel between networks (or a user and a network).

VPC peering - Connect two VPCs so they can route to each other. Non-transitive.


Security & Identity

Access key - A pair of credentials (key ID + secret) that authenticates programmatic access. Treat like a password.

ACL (Access Control List) - A list of rules saying who can do what to a resource.

API key - A long token used to authenticate API calls. Less powerful than a credential pair; widely used for SaaS.

Audit log - A record of who did what, when. Required for most compliance frameworks.

Authentication (AuthN) - Proving who you are. (Username/password, MFA, SSO.)

Authorization (AuthZ) - What you're allowed to do once authenticated. (RBAC, ABAC, policies.)

Bearer token - Anyone who holds it can use it. JWTs are typically bearer tokens.

Certificate - A signed public key that proves identity. Used by TLS.

CIA triad - Confidentiality, Integrity, Availability. The three goals of infosec.

CSPM (Cloud Security Posture Management) - Tools that scan your cloud for misconfigurations. Wiz, Prisma, AWS Security Hub.

Encryption at rest - Data on disk is encrypted. Decryption happens on read.

Encryption in transit - Data over the network is encrypted (typically TLS).

IAM (Identity and Access Management) - Generic term for controlling who can do what. Each cloud has its own IAM service.

JWT (JSON Web Token) - A signed JSON blob carrying claims (user ID, expiry, etc.). Common in API auth.

KMS (Key Management Service) - Cloud service for managing encryption keys. AWS KMS, Azure Key Vault, GCP KMS.

Least privilege - Give an identity only the permissions it needs, no more.

MFA (Multi-Factor Authentication) - Requires more than just a password (e.g., TOTP, hardware key, SMS).

OAuth 2.0 - The protocol that powers "Sign in with Google." Authorization framework.

OIDC (OpenID Connect) - Authentication layer on top of OAuth 2.0. Adds identity tokens.

Penetration test (pentest) - Authorized hacking. Hire someone to break in so you can fix it.

RBAC (Role-Based Access Control) - Permissions are grouped into roles; you assign roles to users.

Secret - Sensitive data (passwords, API keys, certificates) stored separately from code.

Secrets manager - A service for storing and rotating secrets. AWS Secrets Manager, HashiCorp Vault, Azure Key Vault, GCP Secret Manager.

Service account - An identity that belongs to a service or app, not a human.

SIEM (Security Information and Event Management) - Aggregates logs and alerts on suspicious patterns. Splunk, Sentinel, Elastic SIEM.

SSO (Single Sign-On) - Log in once, access many apps. Usually SAML or OIDC.

WAF (Web Application Firewall) - Filters HTTP requests for attacks (SQL injection, XSS, etc.).

Zero Trust - Don't trust the network perimeter. Verify every request, regardless of source.


DevOps & Infrastructure as Code

Ansible - Agentless configuration management tool. YAML playbooks, runs over SSH.

Artifact - The output of a build (a JAR, a Docker image, a binary). Stored in an artifact registry.

Blue/green deployment - Two production environments (blue = current, green = new). Switch traffic once green is verified.

Canary deployment - Roll out a new version to a small percentage of users, watch metrics, expand if good.

CD (Continuous Deployment) - Every commit that passes tests goes to production. Automated end-to-end.

CI (Continuous Integration) - Every commit triggers a build and test run. Catches regressions fast.

CICD pipeline - Automated workflow that builds, tests, and deploys code. GitHub Actions, GitLab CI, Jenkins, CircleCI.

Docker - The most common container runtime / image format.

Drift - When live infrastructure no longer matches what's in code. Bad. IaC tools detect it.

GitOps - Git is the source of truth for infrastructure. Tools like ArgoCD or Flux sync the cluster to git.

IaC (Infrastructure as Code) - Define infrastructure in text files, version it, deploy it. Terraform, CloudFormation, Bicep, Pulumi.

Idempotent - Running the operation multiple times produces the same result. IaC tools rely on this.

Immutable infrastructure - Servers are never modified after deploy. To "update," you replace.

Pulumi - IaC using real programming languages (TypeScript, Python, Go).

Rolling deployment - Replace instances gradually rather than all at once.

SRE (Site Reliability Engineering) - Google-coined discipline for running production. SLOs, error budgets, postmortems.

Terraform - The dominant IaC tool. HCL syntax, multi-cloud, state-driven.

Terragrunt - A wrapper around Terraform for managing many environments / modules.


Containers & Kubernetes

Container - A packaged app + dependencies that runs in an isolated process. Lighter than a VM, heavier than a function.

Container registry - Where Docker images live. Docker Hub, ECR, GCR/Artifact Registry, ACR.

ConfigMap - A Kubernetes object holding non-secret config. Mount as files or env vars.

CRD (Custom Resource Definition) - Lets you extend Kubernetes with your own resource types.

Deployment - Kubernetes resource that manages replica sets and rolling updates.

Helm - The package manager for Kubernetes. Charts = templated manifests.

Image - A built container blueprint. You run an image, you get a container.

Ingress - A Kubernetes resource that exposes HTTP services to the outside world. Backed by an ingress controller (Nginx, Traefik, etc.).

Kubernetes (K8s) - Open-source container orchestrator. Schedules pods across nodes, restarts failures, handles networking and storage.

Kustomize - Built-in Kubernetes templating without Helm. Layer-based overlays.

Manifest - A YAML file describing a Kubernetes resource.

Namespace - A logical partition inside a cluster. Used for tenancy, RBAC, and resource quotas.

Node - A worker machine in a Kubernetes cluster. Runs pods.

Operator - A custom controller that automates managing a stateful app (databases, caches) on Kubernetes.

Pod - The smallest deployable unit in Kubernetes. One or more containers sharing network and storage.

Replica - One copy of a pod. Deployments manage replicas.

Secret (Kubernetes) - Like a ConfigMap but for sensitive data. Base64-encoded by default (not actually encrypted).

Service (Kubernetes) - A stable network endpoint that routes to pods. ClusterIP, NodePort, LoadBalancer.

Service mesh - Layer that adds traffic management, security, and observability to service-to-service calls. Istio, Linkerd, Consul.

StatefulSet - Like a Deployment but for stateful apps. Stable identities, persistent storage.

Sidecar - A second container in a pod that augments the main one (logging, proxy, etc.).


Data & Databases

ACID - Atomicity, Consistency, Isolation, Durability. The promises of traditional relational databases.

BASE - Basically Available, Soft state, Eventual consistency. The relaxed promises of many NoSQL systems.

Columnar storage - Data on disk is organized by column, not row. Great for analytics. Used by data warehouses.

Data lake - A bunch of raw files in object storage. Schema-on-read.

Data lakehouse - Lake + warehouse hybrid. Open table formats (Delta, Iceberg, Hudi) on object storage with warehouse-like queries.

Data warehouse - Optimized for analytical queries over large datasets. BigQuery, Redshift, Snowflake, Synapse.

Document DB - Stores JSON-like documents. MongoDB, Firestore, DynamoDB (sort of).

ETL / ELT - Extract, Transform, Load (transform before loading) vs Extract, Load, Transform (load raw, transform inside the warehouse). ELT is increasingly common.

Eventual consistency - Reads might be stale right after a write but will eventually catch up.

Graph DB - Stores nodes and edges. Neo4j, Neptune, ArangoDB.

Index - A separate data structure that makes queries faster. Costs disk and write speed.

Key-value store - Simplest DB model. Get/put by key. Redis, DynamoDB, etcd.

NoSQL - Anything that isn't a traditional SQL relational database. Document, key-value, graph, time-series, etc.

OLAP (Online Analytical Processing) - Big aggregate queries over historical data. Warehouses.

OLTP (Online Transaction Processing) - Fast, small transactions. Operational databases.

ORM (Object-Relational Mapper) - Library that maps DB rows to objects in your code. SQLAlchemy, Prisma, ActiveRecord.

Relational DB - Tables with rows and columns, joined via keys. Postgres, MySQL, SQL Server, Oracle.

Replica - A read-only copy of a database for scaling reads or disaster recovery.

Schema - The structure of your data (tables, columns, types).

Sharding - Splitting a dataset across multiple servers based on a key.

Time-series DB - Optimized for timestamped data. InfluxDB, TimescaleDB, Prometheus.

Transaction - A group of operations that succeed or fail together.

Vector database - Stores high-dimensional vectors (embeddings) for similarity search. Pinecone, Weaviate, pgvector, Qdrant, Milvus.


AI / Machine Learning

Algorithm (ML) - The recipe used to learn from data. Linear regression, decision trees, neural networks.

Bias (ML) - Systematic error from a model. Often means the model unfairly favors some groups.

Classification - Predicting a discrete label (spam / not spam).

Clustering - Unsupervised grouping of similar items.

Cross-validation - Splitting your training data multiple ways to check that performance generalizes.

Dataset - The collection of examples a model learns from.

Deep learning - ML using neural networks with many layers.

Epoch - One pass over the training set.

Feature - An input variable used by a model.

Feature engineering - Crafting input variables to make a model learn better.

GPU (Graphics Processing Unit) - Originally for graphics, now the workhorse of ML training. Massively parallel.

Gradient descent - The optimization algorithm at the core of most neural network training.

Hyperparameter - A setting you choose before training (learning rate, batch size, etc.). Distinguished from learned parameters (weights).

Inference - Running a trained model on new data to get predictions. Distinguished from training.

Label - The correct answer for a training example. Used in supervised learning.

Loss function - Measures how wrong the model's predictions are. Training minimizes this.

ML (Machine Learning) - Algorithms that improve with data, instead of being explicitly programmed.

Model - The output of training. Takes inputs, gives predictions.

Neural network - A model loosely inspired by neurons. Weighted connections in layers.

Overfitting - Model memorizes training data and performs badly on new data.

Regression - Predicting a continuous number (price, temperature).

Reinforcement learning - Agent learns by acting in an environment and getting rewards. AlphaGo, robotics, game AI.

Supervised learning - Training on labeled examples.

TPU (Tensor Processing Unit) - Google's custom chip for ML. Comparable to a GPU.

Training - Adjusting model weights based on data to minimize loss.

Underfitting - Model is too simple to capture the pattern.

Unsupervised learning - Finding structure in unlabeled data (clustering, dimensionality reduction).

Validation set - Held-out data used during training to tune hyperparameters.


LLMs & Generative AI

Agent (AI) - An LLM-powered system that can take actions (call tools, read files, send messages) toward a goal.

Attention - The core mechanism in transformers. Lets each token "look at" other tokens when computing its representation.

Benchmark - A standard test for models. MMLU, HumanEval, SWE-bench, GSM8K, etc.

Chain-of-thought (CoT) - Asking the model to "think step by step" before answering. Often improves reasoning.

Chunking - Splitting documents into smaller pieces for embedding / retrieval. Bad chunking ruins RAG.

Claude - Anthropic's family of LLMs (Haiku, Sonnet, Opus).

Context window - The max number of tokens the model can consider at once. Bigger = more text in, but slower and more expensive.

Distillation - Training a smaller model to mimic a larger one.

Embedding - A vector that represents the meaning of text (or images, audio). Similar meanings = nearby vectors.

Eval - Automated test of model quality. Can be deterministic (does it match a regex?) or model-graded.

Fine-tuning - Continuing training of a base model on your own data to specialize it.

Foundation model - A large, general-purpose model trained on broad data. GPT, Claude, Gemini, Llama.

Function calling / Tool use - The model returns a structured request to invoke a function, then resumes once the result is fed back in.

Generative AI (GenAI) - Models that generate new content (text, images, audio, video) instead of classifying.

GPT - OpenAI's family of LLMs (GPT-3.5, 4, 4o, etc.). The first widely-known LLM.

Grounding - Constraining a model's answers to verified sources (often via RAG).

Hallucination - When a model confidently makes up something false.

Inference cost - Per-token (or per-second) cost of running a model. Often tracked as $/M tokens in/out.

LangChain / LlamaIndex - Frameworks for building LLM apps. LangChain = general orchestration, LlamaIndex = retrieval-focused.

Latent space - The high-dimensional space embeddings live in.

LLM (Large Language Model) - A model trained on huge amounts of text to predict the next token.

LoRA (Low-Rank Adaptation) - Cheap fine-tuning by training only a small adapter, not the whole model.

MCP (Model Context Protocol) - Anthropic's open protocol for connecting LLMs to tools and data sources.

Multimodal - Models that handle multiple input/output types (text + image, text + audio, etc.).

Parameter - One of the learned numbers in a model. GPT-3 had 175B; modern flagship models can have hundreds of billions.

Prompt - The text you give the model.

Prompt caching - Reusing a precomputed prompt prefix to save cost and latency. Anthropic and OpenAI both support it.

Prompt engineering - The craft of writing prompts that get good outputs reliably.

Prompt injection - An attack where malicious text tells the model to ignore its instructions.

RAG (Retrieval-Augmented Generation) - Look up relevant documents, stuff them into the prompt, then generate. Standard pattern for "chat with your docs."

RLHF (Reinforcement Learning from Human Feedback) - Training step where humans rank model outputs to teach it preferences.

Reranker - A second-stage model that re-scores retrieved chunks for relevance.

Sampling - The process of choosing the next token. Temperature, top-p, top-k are sampling parameters.

Semantic search - Search by meaning, not keywords. Powered by embeddings.

SFT (Supervised Fine-Tuning) - Fine-tuning on labeled examples.

System prompt - Top-level instructions to the model. Sets the role/behavior.

Temperature - Controls randomness in sampling. 0 = deterministic, higher = more varied.

Token - The unit a model reads/writes. Usually ~4 characters of English. Costs are per-token.

Tokenizer - Splits text into tokens. Different models use different tokenizers.

Top-k / Top-p sampling - Restrict the model's choices to the most likely next tokens.

Transformer - The neural network architecture behind nearly all modern LLMs. Introduced in "Attention Is All You Need" (2017).

Vibe coding - Writing software primarily by prompting an AI model rather than editing code yourself. Term popularized in 2025.


Observability

Alert - A rule that fires when a metric crosses a threshold. Wakes someone up (or not).

APM (Application Performance Monitoring) - Tools that trace requests through your app and surface slowness. Datadog, New Relic, AppDynamics.

Cardinality - The number of unique label combinations on a metric. High cardinality = expensive metrics.

Distributed tracing - Following a single request across many services. OpenTelemetry, Jaeger, Tempo, X-Ray.

Error budget - The amount of failure tolerated by an SLO. If you blow it, no risky deploys.

Log - A timestamped text record of an event.

Metric - A numeric measurement over time (request count, error rate, latency).

Observability - The ability to understand what's happening inside a system from its outputs (logs, metrics, traces).

OpenTelemetry (OTel) - The open standard for instrumenting apps with logs, metrics, and traces.

SLO (Service Level Objective) - The target for a service's reliability (e.g., "99.9% of requests under 500ms").

Span - One unit of work in a distributed trace.

Trace - The full journey of a request across services. Made up of spans.


Want a term added?

Open an issue or PR. See CONTRIBUTING.md.