Glossary: Reliability Terms
All the jargon explained in plain English.
A
Availability How often your system is up and working. Measured as a percentage over time.
- Example: “99.9% availability” means your system should be working 99.9% of the time, down for about 43 minutes per month.
- Related: SLA, uptime
ADR (Architecture Decision Record) A written note explaining a technical decision: what you decided, why you decided it, and what you considered. Used to prevent the same decision from being re-debated repeatedly.
B
Backup A copy of your data stored separately so you can restore it if the original is lost or damaged.
- Cold backup: Not actively used, takes time to restore
- Hot backup: Ready to use immediately
- Related: RPO, disaster recovery
BCDR (Business Continuity and Disaster Recovery) The plan for keeping your business running if something goes wrong. Includes backups, failover procedures, and recovery procedures.
Baseline Your normal state. What things look like when nothing is broken. Used to spot when something is wrong.
- Example: “Your database normally processes 1,000 requests per second. If it drops to 500, that is a baseline violation and signals a problem.”
C
Cascade (or cascading failure) When one service breaks and causes other services to break as a result, then more services break as a chain reaction.
- Example: “Payment service went down. Then Order service went down because it could not process payments. Then Billing service went down because it could not update billing.”
- Related: dependency, downtime
CDN (Content Delivery Network) A service that copies your content to multiple locations around the world so users can download from the closest location (faster).
Certificate (or SSL/TLS certificate) A digital document that proves your website is who it claims to be. Expires and must be renewed. If it expires, browsers will refuse to connect.
- Related: expiration, renewal
Circuit breaker A safeguard that stops requests being sent to a broken service temporarily, preventing your system from trying over and over again and making the problem worse.
- Works like an electrical circuit breaker: if something is wrong, flip it off instead of burning things out
Cluster A group of servers working together. If one fails, the others keep working.
- Example: A Kubernetes cluster = multiple servers running containers together
Cold start The delay when a service is not running and needs to start up from nothing. Cold starts are slower than resuming a running service.
- Related: warm start, scaling
Connection pool A group of database connections sitting ready to use, instead of creating a new connection every time. Makes things faster.
- When the pool is full (all connections in use), new requests have to wait
Control plane The part of a system that manages other parts (like the brain). Makes decisions about what should run where.
- Related: data plane
D
Dashboard A visual display showing current system status (graphs, numbers, color-coded indicators). Used to spot problems quickly.
Data corruption When your data gets changed incorrectly and you do not realize it is wrong.
- Example: Order quantity shows as 0 instead of 100, but the system looks fine (silent failure)
- Related: silent failure, data integrity
Data integrity The trustworthiness of your data. If your data is consistent and correct, you have data integrity.
Data plane The part of a system that actually does the work (serving requests, processing data). Not the management layer.
- Related: control plane
Degraded mode When a system is not fully working but is still running at reduced capacity instead of completely failing.
- Example: “Identity service is down, so we switched to cached tokens and manually approved logins (degraded mode)”
- Related: failover
Dependency Another service or system your system depends on to work.
- Example: “Our payment service depends on Stripe API. If Stripe is down, we are down.”
- Related: cascade
Deployment Releasing new code or configuration to production (putting it live).
- Blue-green deployment: Two identical environments, flip traffic between them
- Canary deployment: Send new code to 1% of users, then 10%, then 100% (test gradually)
- Rolling deployment: Replace servers one at a time
- Related: rollback, change
Disaster recovery The process of getting your system back up after something catastrophic fails.
- Related: BCDR, backup, RTO
DNS (Domain Name System) The system that converts domain names (example.com) to IP addresses (185.192.1.1). If DNS is broken, browsers cannot find your website.
E
Error budget The amount of downtime you are allowed before missing your SLA.
- Example: If your SLA is 99.9% uptime, your error budget is ~43 minutes per month
- Related: SLA, downtime
Escalation Calling in senior people when a problem is too big or urgent for on-call to handle alone.
- Example: After 15 minutes with no progress, escalate to the tech lead
F
Failover Switching traffic to a backup system when the primary system breaks.
- Example: “Primary database is down, failover to replica”
- Related: backup, BCDR, disaster recovery
Fallback A simpler or reduced version of a system that you use when the full system breaks.
- Example: “Identity provider is down, fallback to cached tokens for 2 hours”
- Related: degraded mode
Federation (or federated identity) Using credentials from one system in another system.
- Example: Sign in with Google, then use that to log into Slack
- Federation drift: When the systems get out of sync and stop trusting each other
Firewall A security device that controls which traffic is allowed to enter or leave your network.
Flow The movement of data through your system. If traffic is not flowing, data is getting stuck.
G
Graceful degradation When a system can keep working (partially) even when something breaks, instead of failing completely.
- Example: “If cache is down, read from database instead (slower but works)“
H
Health check A test that verifies a service is working. Runs frequently to catch failures early.
- Example: “Ping the API every 10 seconds, if no response, mark it as down”
- Related: monitoring, alerting
Heartbeat A regular signal from a service saying “I am alive.” If heartbeats stop, the service is assumed dead.
- Related: health check
I
IC (Incident Commander) The person in charge during an outage. Makes decisions about what to do (rollback, failover, etc.).
- Related: incident response
Identity provider (or auth provider) A system that handles logins and creates tokens/credentials for users.
- Examples: Okta, Azure AD, Auth0
- If this is down, users cannot log in to anything that depends on it
Idempotent An action that gives the same result no matter how many times you do it.
- Example: “Set color to red” is idempotent (doing it twice = same result)
- Example: “Add 1” is NOT idempotent (doing it twice adds 2)
- Related: retries
Incident An event where your system is not working as expected. Ranges from small glitches to complete outages.
Incident response The procedure for handling an incident: triage, decide on action, execute, validate recovery.
Latency How long something takes. Usually measured in milliseconds.
- Example: “Database query latency is 50ms” = queries take 50 milliseconds
- Related: performance, throughput
K
Kubernetes (K8s) A system for automatically managing containers and servers at scale.
L
LB (Load balancer) A device that splits traffic across multiple servers so no single server gets overwhelmed.
Logs Records of what your system did: who accessed it, what errors occurred, etc. Used for debugging.
M
Metric A number you track to understand system health. Examples: uptime, error rate, latency.
- Related: SLO, monitoring
Mitigation A temporary fix that stops the bleeding without fixing the root cause.
- Example: “Mitigation: Route traffic away from the broken server. Root cause fix: Deploy a new version of the code.”
Monitoring Continuously watching metrics and logs to catch problems early.
- Related: alerting, observability
O
OODA loop Observe → Orient → Decide → Act. A decision-making framework. The team that cycles through this loop fastest wins.
Observability Your ability to see what your system is doing. Built from metrics, logs, and traces.
- Related: monitoring, visibility
Outage When a service is completely down or unavailable.
- Related: incident, downtime
P
Partial failure When part of your system works and part does not. More common and more dangerous than complete failures.
- Example: “East Coast servers work, West Coast servers do not”
Patch A small update that fixes a specific problem, usually a security fix or bug.
Postmortem A meeting after an incident to understand what happened, why, and how to prevent it next time.
- Related: incident response, lessons learned
Provider (or cloud provider) The company hosting your infrastructure (AWS, Azure, Google Cloud, etc.).
- Related: SLA, outage
Provisioning Setting up new servers, databases, or resources and making them ready to use.
Q
Queue (or message queue) A system for storing messages in order until another system processes them.
- Example: “Enqueue 10,000 orders to process, then process them one by one”
- Related: throughput, scaling
R
Redundancy Having backups or duplicates so if one fails, the other keeps working.
- Example: “Primary database + replica = redundancy”
Replication (or data replication) Copying data from one place to another in real-time or near-real-time so you have a backup.
- Replication lag: The delay between primary and replica (if it is 1 second behind, lag = 1 second)
Resilience The ability to keep working or recover quickly when something breaks.
RTO (Recovery Time Objective) How fast you need to recover. The maximum downtime you can tolerate.
- Example: “RTO = 1 hour” means you must be back up within 1 hour
RPO (Recovery Point Objective) How much data loss you can tolerate. If your backup is 1 day old and you lose the last day of data, RPO = 1 day.
- Example: “RPO = 1 hour” means you can afford to lose up to 1 hour of data
Rollback Reverting to a previous version (undoing a deployment or change).
- Example: “New code broke production, rollback to the previous version”
Runbook Written procedures for handling specific problems. Step-by-step instructions for when things break.
S
SLA (Service Level Agreement) A promise to customers about uptime/availability (e.g., “99.9% uptime”).
- Related: SLO, error budget
SLO (Service Level Objective) An internal target for reliability (usually matches the SLA but may be stricter).
SPOF (Single Point of Failure) One thing that, if it breaks, breaks everything downstream.
- Example: “We only have one database server = SPOF”
- Related: redundancy, resilience
Scaling (or auto-scaling) Adding more servers when demand increases, removing them when demand decreases.
- Horizontal scaling: Add more servers
- Vertical scaling: Make each server more powerful
- Related: load balancer, throughput
Schema The structure of a database: what tables exist, what columns they have, what data types.
- Schema migration: Changing the structure (adding a column, renaming a table, etc.)
Shard (or sharding) Splitting data across multiple databases or servers by a key (e.g., split users by region).
- Related: scaling, load balancer
Silent failure When a system fails but looks like it is working (returns 200 OK with corrupted data).
- Example: “Cache returns stale data silently, you do not know it is wrong”
- Related: data corruption, monitoring
Spike A sudden, temporary increase in traffic or load.
- Example: “Error rate spiked to 50% for 2 minutes, then returned to normal”
Staging A test environment that mirrors production but is not used by real customers. Used for testing before deploying.
- Related: production, deployment
Status page A public page showing whether services are up or down and any ongoing incidents.
T
Telemetry Data collected about your system’s behavior: metrics, logs, traces.
Throughput How much work a system can do per unit time.
- Example: “Database throughput = 10,000 queries per second”
- Related: latency, capacity
Throttling Intentionally slowing down or rejecting requests when a resource is overloaded.
- Example: “API rate limit: 100 requests per second, throttle the rest”
Tier (or tier numbering) A ranking of how critical something is.
- Tier 1 (critical): If down, customers cannot use the service
- Tier 2 (important): If down, customers experience degradation
- Tier 3 (nice-to-have): If down, only internal processes affected
- Related: criticality, SLA
Timeout A set time limit for an action. If it takes longer, stop trying.
- Example: “Request timeout = 5 seconds, stop trying after 5 seconds”
Token (or auth token) A digital credential proving you are logged in. Short-lived (expires).
- Token refresh: Getting a new token before the old one expires
Trace (or distributed trace) A log of a single request as it moves through all the services in your system.
- Related: logging, observability
U
Uptime How long a system is up and working. Opposite of downtime.
- Related: availability, SLA
V
Visibility Being able to see what your system is doing. If you have no visibility, you cannot debug problems.
- Related: observability, monitoring
W
War room A room (or virtual space) where the incident response team gathers during a crisis to make decisions and coordinate.
Webhook A way for one system to automatically notify another when something happens.
- Example: “When a payment completes, webhook to our billing system to update the invoice”
Z
Zone (or availability zone) A separate data center in a region. If one zone fails, the other zones keep working.
- Example: “AWS us-east-1 has 3 zones: us-east-1a, 1b, 1c”
- Related: region, redundancy
Looking for a term?
Not in this glossary? Check your chapter for a definition in parentheses or brackets. Many terms are explained on first use.
If you find a term that should be here, let us know.