Glossary
Short definitions for terms used across Foundations and the case studies. For the full mechanism behind a term, follow its “Read more” link.
A
- Authentication
- Verifying who is making a request. Checked once, typically at login, producing a session or token that every subsequent request can reference. Read more
- Authorization
- Verifying whether an already-authenticated caller is allowed to do a specific thing to a specific resource. Has to be checked per resource and per action, not once. Read more
- Availability
- Every request gets a non-error response, though not necessarily one that reflects the latest write. Distinct from correctness and from durability. Read more
B
- Burn Rate
- How fast an error budget is being consumed right now, not just whether it's already gone. A well-designed symptom-based alert is usually built on burn rate instead of a single breached threshold, so a page can fire while there's still budget left to react within. Read more
C
- Cache Invalidation
- Removing or updating a cached value once the underlying data it represents has changed, so the cache stops serving something stale. Read more
- CAP Theorem
- During a network partition, a distributed system has to choose between staying consistent and staying available. It can't do both at once. Read more
- Cause-Based Alert
- An alert that fires on internal system state, CPU, memory, disk space, a pod restarting, queue depth, values that move constantly as an ordinary consequence of autoscaling and deploys. Still useful for dashboards and debugging once something's known to be wrong, just not reliable enough to page a human on its own. Read more
- CDN
- Content delivery network: servers spread across many physical locations that cache content close to users, so a request doesn't have to travel all the way to the origin server. Read more
- Circuit Breaker
- A mechanism that tracks failures to a downstream dependency and, past a threshold, stops calling it entirely for a while instead of letting every caller wait out a timeout. Read more
- Consistency
- Every read reflects the most recent acknowledged write, or the system says it can't guarantee that. One of the three properties in tension in CAP theorem. Read more
- Conway's Law
- A system's architecture ends up mirroring the communication structure of the organization that built it, whether or not anyone planned it that way. Read more
- CQRS
- Command query responsibility segregation: splitting a system's write model, the source of truth, from its read model, a separate copy shaped for the queries that actually get run. Read more
D
- Dead-Letter Queue
- A separate holding place for messages that failed processing after exhausting their retries, instead of being retried forever or dropped silently. Read more
E
- Error Budget
- The SLO flipped around: 100% minus the target, the amount of failure still allowed before the promised number is broken. Spending it is supposed to be a decision rule, not just a statistic: once it's gone, the agreed response is to freeze feature work and fix reliability first. Read more
- Event-Driven Architecture
- Structuring a system around services publishing events describing what happened, with other services subscribing to whichever ones they care about, instead of calling each other directly. Read more
I
- Idempotency
- A property where repeating an operation produces the same result as doing it once, so a retried request doesn't cause a duplicate effect. Read more
J
- JWT
- JSON Web Token: a signed, self-contained token that carries its own claims and can be verified locally, without a lookup against any server-side store. Read more
L
- Liveness
- Whether a process is alive at all, not stuck in a way it will never recover from on its own, a deadlock or an infinite loop. Failing a liveness check has one correct response: kill the process and start a fresh one. Read more
- Load Balancer
- A component that sits in front of multiple instances of a service and decides which instance handles each incoming request. Read more
O
- OAuth
- A delegated authorization protocol: lets a user grant a third-party app a scoped permission without ever handing that app a password. Read more
- OIDC
- OpenID Connect: an identity layer built on top of OAuth, adding a specific token, the ID token, that proves who a user is rather than just what an app is allowed to do. Read more
- Outbox Pattern
- Writing an event to a table in the same database transaction as the actual data, so the write and the intent to publish commit atomically, with a separate process relaying it to the real message broker afterward. Read more
P
- Partition Tolerance
- A system keeps functioning even when network messages between nodes are dropped or delayed. Not really optional for anything that runs on more than one machine. Read more
R
- Rate Limiting
- Capping how many requests a client is allowed to make in some period, rejecting or delaying anything over that cap. Read more
- Readiness
- Whether a specific instance can actually handle traffic right now, a different question from whether the process is merely alive. Failing readiness should stop new traffic from arriving without touching the process itself. Read more
- Replication Lag
- The delay between a write landing on a primary database and that same write becoming visible on a replica. Read more
- RPO
- Recovery point objective: the maximum acceptable data loss, expressed as a window of time, how far back the most recent recoverable copy goes if disaster strikes right now. Read more
- RTO
- Recovery time objective: the maximum acceptable time to actually restore service after a disaster, from the moment it starts to traffic being fully redirected to something working again. Read more
S
- Salting (password hashing)
- Mixing a random, unique value into a password before hashing it, so two users with the same password get different stored hashes and a precomputed table of common hashes doesn't work against the whole database at once. Read more
- Shard Key
- The field used to decide which shard a given row lives on. Get it wrong and you get a hot shard, or a query that has to fan out to every shard. Read more
- Sharding
- Splitting one dataset across multiple database instances by a shard key, so no single machine has to hold or write all of it. Read more
- SLI
- Service level indicator: the specific thing measured, like percentage of requests that succeed or percentage served under 200ms, chosen because it reflects what a user actually experiences. Read more
- SLO
- Service level objective: the target chosen for an SLI over some window, like 99.9% of requests succeeding over a rolling 30 days. A deliberate ceiling on acceptable failure, not a target to always beat. Read more
- Span
- One service's own recorded contribution to a trace: which operation ran, how long it took, and which span it descended from. A trace is every span sharing one trace id, reassembled into a single timeline. Read more
- Symptom-Based Alert
- An alert that fires on what a user actually experiences: error rate, latency, whether an SLI has breached its SLO. Moves far less often than a cause-based signal, because it only fires when something genuinely affects someone on the other end of the request. Read more
T
- TLS
- Transport Layer Security: encrypts the connection between a client and a server so anything traveling over it is unreadable to anyone intercepting it in the middle. Read more
- Token Bucket
- A rate-limiting algorithm where a client has a bucket of tokens that refills continuously, and each request consumes one, allowing a bounded burst instead of a hard per-window cutoff. Read more
- Trace
- The full path one specific request took across every service it touched, built by tagging every hop with the same trace id generated once at the edge. What metrics and logs can't do alone: show exactly where the time went for that one request. Read more
- TTL
- Time to live: how long a cached or stored value is allowed to sit before it's treated as expired, regardless of whether it's still actually correct. Read more