Glossary

302 Found: The HTTP status that tells the browser the target is temporary. The browser follows the Location header and keeps using the short URL next time.; A URL shortener
429 Too Many Requests: The HTTP status for a client that sent too many requests in a given time. It may carry a Retry-After header saying when to try again.; A rate limiter
Anchor: A copy of the latest hash written to a place the database admins cannot change, used later to check the chain.; An audit log
At-least-once: A sender retries until it gets an acknowledgement, so the receiver may see the same event more than once.; Notifications
At-least-once: Each job runs one or more times. A crash can make it run again, and it never runs zero times.; A job scheduler
At-least-once delivery: Every event reaches the customer one or more times: the sender retries until it gets a 2xx, so the customer can get a copy twice.; Webhook delivery
Authorize and capture: Two card calls: the authorization holds the amount on the card, the capture takes the money later.; An order across services
Backfill: A job that fills a new column for rows that existed before it, in small batches.; Deploying without downtime
Backlog: The jobs waiting in the queue that no worker has taken yet.; A queue that fills up
Backpressure: A signal from the consumer side back to the producer that says "send less", so the backlog stops growing.; A queue that fills up
Base62: Writing a value with the 62 characters 0-9, a-z and A-Z. A code of n characters has 62 to the power n possible values.; A URL shortener
Binlog: The MySQL binary log: every committed change, in commit order. MySQL uses it for replication.; Moving to a new database
Block: A fixed-size slice of a file, named by the hash of its bytes, so two devices can compare files block by block.; File sync
Breaking change: A change after which code that worked against the API stops working: a removed field, a renamed field, a field with a new type.; Changing a public API
Burn rate: How fast the service spends its error budget. A burn rate of 1 spends exactly the whole budget in the 30-day window.; Measuring a service
Capped list: A list trimmed to a fixed length after each insert, so it keeps only the newest entries.; An activity feed
Central authority: The one process that decides, for one document, the order in which steps are applied.; Shared document editing
Change data capture (CDC): Reading a database’s log of changes and sending each change somewhere else as an event.; Moving to a new database
Channel: One way to reach a user: email, push to a phone, or the list of notifications inside the app.; Notifications
Check: The question "does user U have relation R on object O?", answered by following tuples from O to U.; A permissions system
Circuit breaker: A wrapper around a call that counts recent failures. Above a threshold it stops making the call for a while and fails at once, so the caller goes straight to its fallback.; When a dependency is down
Collapse id: The APNs header apns-collapse-id. Push requests with the same value show as one notification on the phone.; Notifications
Compensating transaction: The step that undoes an earlier one: release the reserved stock, cancel the card authorization.; An order across services
Completion suggester: An Elasticsearch feature that returns indexed inputs starting with a given prefix, from a structure kept in memory.; Search autocomplete
Conditional update: An UPDATE whose WHERE clause states the row must still be free. The returned row count tells you whether you got it.; Booking concert tickets
Conflict: Two revisions made from the same parent revision on two devices, which sync then puts side by side under one document id.; An offline-first mobile app
Conflict copy: The second version of a file, kept under a new name when two edits cannot be ordered.; File sync
Cutover: The moment the application stops writing to the old database and starts writing to the new one.; Moving to a new database
Dead row version: In PostgreSQL, the old copy of a deleted or updated row. It keeps its disk space until VACUUM marks that space free for reuse.; A paste service
Deadlock: Two transactions each hold a lock the other one waits for, so neither can finish.; Booking concert tickets
Debounce: The client waits a short time after the last keystroke and sends one request for the text at that moment.; Search autocomplete
Delivery: One row meaning "tell this user about this event on this channel". Its key is (event_id, user_id, channel).; Notifications
DNS rebinding: A domain answers with a public IP when you check it, then with an internal IP when you connect.; Link previews
Epoch: The moment the timestamp part of an id counts from. Snowflake counted from November 2010 instead of 1970, so the 41 bits last longer.; Unique ids across servers
Error budget: The failures the SLO allows in its window. At 99.9%, it is 0.1% of the requests.; Measuring a service
Event notification: A message the storage service sends when an object is created in a bucket. Here it goes to a queue.; Image thumbnails
Expand and contract: A schema change split into steps. First add the new shape next to the old one, then move code and data to it, and only then remove the old shape.; Deploying without downtime
Expired event: A Redis keyspace notification published when Redis deletes a key whose TTL ran out.; Online status
Expiry time: The moment after which a paste must no longer be served. The author picks it when creating the paste.; A paste service
Exponential backoff: Each retry waits twice as long as the previous one, starting at 1 minute.; Webhook delivery
Fail open: When the limiter itself is down, let requests through instead of refusing them all.; A rate limiter
Fallback: What the code returns when a call fails or times out, like a stored list of best sellers instead of personal recommendations.; When a dependency is down
Fan-out: Turning one sent message into one delivery per member of the group.; Group chat
Fan-out: Sending one incoming message to many receivers. Here, one goal goes out to 2 million open connections.; Live sports scores
Fan-out on read: The event is stored once, under its author, and each reader collects events from the accounts they follow when they open the feed.; An activity feed
Fan-out on write: When an account posts, the system copies the event id into the feed of each follower, so a read only fetches one list.; An activity feed
Feature flag: A named switch the code reads at run time, like new-checkout, whose rules live outside the code.; Feature flags
Fixed window: A counter per key for each calendar minute (12:00, 12:01...). The count starts again at 0 when the minute changes.; A rate limiter
Flattened membership: For each user, the precomputed list of every team they belong to, directly or through nested teams.; A permissions system
Flush: The moment an app server writes the counts it holds in memory to the database and starts again from zero.; Counting page views
Gateway: A server that holds open WebSocket connections to clients and pushes new messages down them.; Group chat
Geospatial index: A structure that answers "which points lie within 3 km of here?" without reading every point. Redis stores one as a sorted set, filled with GEOADD.; Matching riders and drivers
Grace: In Varnish, extra time after the TTL during which the cache still serves the stale page while it fetches a new one in the background.; Caching a news site
Gradual rollout: A rule that turns a flag on for a percentage of users, chosen by a hash of the user id.; Feature flags
Hash chain: Each record stores a hash computed over its own content and the previous record's hash, so changing one record changes every hash after it.; An audit log
Heartbeat: A small message a connection sends on a fixed interval to say "this user is still here".; Online status
Hold: A seat reserved for one fan for 10 minutes while they pay. Nobody else can hold or buy it until the hold ends.; Booking concert tickets
Hot standby: A PostgreSQL replica that accepts read-only queries while it replays the changes it receives from the primary.; Serving two regions
Idempotency key: A value sent with a request so the receiver can recognise a repeat and do the work only once. Here, the job id.; A job scheduler
Idempotency key: A unique value sent in the Idempotency-Key header. The provider runs the first request with a given key and answers every retry with the saved result.; Taking payments
Idempotency key: A unique string sent with a request, so the server can recognise a retry and return the first result instead of doing the work again.; Capstone: an online shop’s checkout
Inbound reference: A place in another module's code that uses a class of this module. Each one must change if this module moves out.; Splitting a monolith
Last-Event-ID: The HTTP header a browser’s EventSource sends when it reconnects, holding the id of the last event it received.; Live sports scores
Leader key: A key in a small coordination store, like etcd, that only the current primary may hold. It has a time to live, and the primary must renew it before it expires.; Database failover
Lease: A claim on a job that ends at a set time, stored in the job row as locked_until. If the worker does not finish before then, another worker may take the job.; A job scheduler
Lock with expiry: A key that only one process can create, and that deletes itself after a set time. In Redis: SET key value NX EX 15.; Matching riders and drivers
LSN: Log sequence number: a position in PostgreSQL’s write-ahead log, like 16/B374D848. A replica that has replayed up to an LSN has every change written before it.; Serving two regions
Machine id: A number given to each server, unique among all servers. It goes into every id that server makes.; Unique ids across servers
Modular monolith: One codebase and one deploy, with enforced boundaries between modules: a module calls another only through that module's public code.; Splitting a monolith
Multipart upload: An S3 upload sent as numbered parts. S3 joins the parts into one object when the client calls CompleteMultipartUpload.; Video upload pipeline
New enemy problem: A user removed from a team still gets access because a server checked against data from before the removal.; A permissions system
Offline-first: The app reads and writes a local database on the device, and syncs with the server whenever a network is there.; An offline-first mobile app
Open Graph tags: <meta property="og:title"> and og:image tags in a page’s <head> that give the title and image a preview should show.; Link previews
Orchestrator: One component that stores the saga's state and tells each service which step to run next.; An order across services
Origin: The servers that build the pages, here the 6 CMS servers behind the cache.; Caching a news site
Outbox: A table written in the same transaction as the order, holding the messages (like emails) that a worker sends afterwards.; Capstone: an online shop’s checkout
Packwerk: A Ruby gem from Shopify that groups files into packages and reports a violation when a package references a constant from a package it has not declared as a dependency.; Splitting a monolith
Partition key: The part of a row’s key that decides which node stores the row. Rows with the same partition key sit together on disk.; Group chat
Pool and silo: Pool: many tenants share the same tables. Silo: one tenant gets a database of its own.; Multi-tenant SaaS data
Pre-aggregation: Adding up counts in memory on the app server and writing the sums, instead of writing each event.; Counting page views
Presigned URL: A URL your API signs so that a phone can upload one object straight to the bucket, without having storage credentials of its own.; Image thumbnails
Presigned URL: A URL that carries a signature allowing one S3 operation for a limited time. A browser can use it with no AWS credentials of its own.; Video upload pipeline
Pub/Sub: Redis channels: a publisher sends a message to a channel name, and Redis passes it to every client subscribed to that name at that moment.; Group chat
Publisher confirm: A RabbitMQ message from the broker to the producer that says whether it accepted a message (basic.ack) or refused it (basic.nack).; A queue that fills up
Rank: A player’s position on the board, counted from 1 for the best score.; A game leaderboard
Rebase: Rewriting your own unsent steps so they apply on top of steps other people had accepted first.; Shared document editing
Reconnect storm: Thousands of clients reconnecting in the same second, after the server they were connected to stops.; Live sports scores
Refresh interval: How often each server fetches the flag rules from the flag service.; Feature flags
Relation tuple: One stored fact of the form object, relation, user, like project:p1#editor@team:t1#member (members of team t1 are editors of p1).; A permissions system
Replication lag: The time between a commit on the primary and the moment that change is visible on a replica.; Serving two regions
Request coalescing: When many clients ask for the same missing page, the cache sends one request to the backend and holds the others until it answers.; Caching a news site
Reservation: Stock taken off the available count when an order is placed, before it is paid, and put back if payment never happens.; Capstone: an online shop’s checkout
Revision: In CouchDB and PouchDB, one saved version of a document, named like 2-c1592ce7..., where the number counts edits since the first one.; An offline-first mobile app
Rolling deploy: Servers get the new version a few at a time, while the others keep serving. For a while, old and new code run side by side against the same database.; Deploying without downtime
Row-level security: A PostgreSQL feature: a policy on a table decides which rows each query can see or change, whatever the query's own WHERE says.; Multi-tenant SaaS data
RPO: Recovery point objective: the most data, counted in time, the business accepts to lose. Here, 5 minutes.; A backup plan
RTO: Recovery time objective: the longest the business accepts to wait from the decision to restore until the database is back. Here, 2 hours.; A backup plan
Saga: A sequence of local transactions, one per service, where each completed step has a compensating step that undoes it if a later step fails.; An order across services
Sequence: A counter inside one server that numbers the ids made in the same millisecond.; Unique ids across servers
Server-sent events (SSE): One HTTP response that the server keeps open and writes events into, as text/event-stream. The browser reads it with EventSource.; Live sports scores
Short code: The few characters after the domain, like k3Xa9Qz. The service maps each code to one long URL.; A URL shortener
Signature: An HMAC of the request body, computed with a secret shared with the customer, that lets the customer check who sent the request.; Webhook delivery
SKIP LOCKED: A PostgreSQL option of SELECT ... FOR UPDATE: rows that another transaction has locked are left out of the result instead of making the query wait.; A job scheduler
SLI: Service level indicator: a measured ratio, the number of good events divided by the total number of events.; Measuring a service
SLO: Service level objective: the target for an SLI over a time window, such as 99.9% of checkouts succeed over 30 days.; Measuring a service
Sorted set: A Redis type that keeps unique members ordered by a numeric score, with each lookup in O(log N).; A game leaderboard
Split brain: Two servers both accept writes as primary at the same time. Their data diverges, and one side has to be thrown away or merged by hand.; Database failover
SSRF: Server-side request forgery: a user makes your server send a request to an address they choose, such as an internal one.; Link previews
Stale position: The last position a driver sent before their app stopped reporting. The driver may be far away, or offline.; Matching riders and drivers
Status change: A user going from offline to online or from online to offline. A heartbeat from a user who is already online does not count as one.; Online status
Step: One small change to the document, such as "insert 2026-11-03 at position 412". A client sends steps in batches.; Shared document editing
Stickiness: The same user gets the same answer for a flag on every request and on every server.; Feature flags
Synchronous replication: The primary reports a commit as done only after a replica confirms it has the change. A write the client saw succeed is then on two machines.; Database failover
Tamper evident: A change to stored records can be detected afterwards. It does not mean the change was prevented.; An audit log
Tenant: One customer company. Every row of data belongs to exactly one tenant, marked by a tenant_id column.; Multi-tenant SaaS data
Tie-break: The rule that orders two players with the same points, here the one who reached them first.; A game leaderboard
Timeout: The longest a caller waits for an answer before it gives up and moves on. Here, a setting on the HTTP client that calls the recommendation service.; When a dependency is down
TOAST: PostgreSQL’s storage for large field values: it compresses them or moves them into a side table, without any change to your queries.; A paste service
Transcoding: Decoding a video and encoding it again at another resolution or bitrate. Here, ffmpeg makes three smaller versions of each original.; Video upload pipeline
TTL: Time to live: the number of seconds after which Redis removes a key on its own.; Online status
TTL: Time to live: how many seconds a cache serves a stored page before it counts as stale.; Caching a news site
Unknown outcome: The state after a request times out: the provider may have charged the card, or may never have seen the request.; Taking payments
Upsert: One statement that inserts a row, or updates it if a row with the same key already exists.; Counting page views
Upsert: A write that inserts the row if its primary key is new and updates it if the key already exists.; Moving to a new database
Version: An integer: the number of steps the server has accepted for a document. Each accepted step adds 1.; Shared document editing
Version change module: A piece of code that turns a response from the current shape into the shape of one older version.; Changing a public API
Version pinning: The API records which version each API key uses and answers that key in that version until its owner changes it.; Changing a public API
Version vector: One counter per device on each file. A device adds 1 to its own counter each time it changes the file.; File sync
Visibility timeout: In Amazon SQS, the time a received message stays hidden from other workers. If the worker does not delete it in that time, the message shows up again.; Image thumbnails
Waiting room: A queue in front of the seat map that lets fans in at a fixed rate and shows everyone else their place in line.; Booking concert tickets
WAL archiving: PostgreSQL hands each finished write-ahead log segment, 16 MB by default, to a command that copies it off the server. A base backup plus those segments can rebuild the database at any moment in between.; A backup plan
Webhook: An HTTP POST that your service sends to a URL the customer registered, when an event happens on your side.; Webhook delivery
Webhook: An HTTP request the provider sends to your server when something happens on its side, such as a payment that succeeded.; Taking payments
Weight: A number stored with each suggestion that ranks it, here the product’s sales over 30 days.; Search autocomplete