Platform Engineering Glossary
Definitions for Kafka, rate limiting, latency, and API versioning terms used by platform engineers. Click any term to learn more with detailed explanations, formulas, and practical examples.
Latency budget
Total time allocated for a complete user-facing request across all architectural hops.
SLA and SLO (service level agreement vs objective)
SLA is a contract with guarantees; SLO is the internal target set stricter than SLA to create error budget.
p50 latency (median)
50th percentile of request durations: half of requests complete faster, half slower.
p99 latency
99th percentile of request durations: captures experience of users under stress.
Headroom (latency)
Gap between measured latency and SLA ceiling; buffer for unpredictable spikes.
Cold start (serverless)
Extra latency when serverless function starts on new execution environment.
Cross-AZ latency
Round-trip time between availability zones; adds cost and latency to replication.
Tail latency
High-percentile latency values (p99, p99.9) representing slowest requests.
Fan-out (distributed systems)
Single request triggers multiple parallel downstream calls.
N+1 query problem
One query to fetch list plus N queries for each item; causes database round-trips.
Speed of light in optical fibre
~200,000 km/s; physical lower bound on network latency regardless of optimization.
Thundering herd
Many clients simultaneously access resource that just became available.
Percentile latency (p50 / p99 / p999)
Statistical measure of request duration distribution.
Fixed window rate limiting
Counts requests in discrete, non-overlapping time buckets.
Sliding window rate limiting
Maintains rolling count of requests over recent N seconds.
Token bucket
Accumulates tokens at fixed rate; requests consume tokens.
Leaky bucket
Processes requests at constant output rate, queueing excess.
Burst limit
Maximum requests above steady-state rate in short period.
Rate limit window
Time period over which request counts are measured.
Global vs per-consumer rate limit
Global limit shared across all clients vs per-consumer independent counters.
Jitter (retries)
Adds randomness to retry wait times to prevent synchronized retries.
Exponential backoff
Retry wait times grow multiplicatively after each failure.
Retry storm
Failure amplification where synchronized retries recreate overload.
HTTP 429 Too Many Requests
Standard status code for rate limiting responses.
RateLimit headers (IETF)
Standardized headers for communicating quota information.
X-RateLimit headers
Legacy de facto standard for rate limit state communication.
Retry-After header
Specifies how long client should wait before retrying.
API throttling vs rate limiting
Rate limiting rejects excess; throttling delays to preserve availability.
nginx limit_req
Nginx directive implementing leaky bucket rate limiting (token-bucket-like with nodelay).
Kong rate-limiting plugin
Enforces request quotas on routes and services.
Kafka partition
Fundamental unit of parallelism and ordering within a topic.
Consumer lag (Kafka)
Difference between latest message and consumer committed offset.
Consumer group (Kafka)
Set of consumers cooperatively consuming a topic.
Consumer offset (Kafka)
Position consumer group has committed as processed.
Effective consumers (Kafka lag)
Actual consumers processing partitions, capped at partition count.
Consumer poll loop (Kafka)
Central control loop: fetch records, process, repeat.
Consumer rebalance (Kafka)
Redistributing partition assignments among group consumers.
Replication factor (Kafka)
Number of brokers storing copies of each partition.
Log retention (Kafka)
How long messages kept before deletion eligibility.
Producer compression (Kafka)
Reduces batch size before sending to broker.
Throughput vs bandwidth (Kafka)
Actual processed rate vs theoretical maximum capacity.
max.poll.interval.ms
Maximum time between poll() calls before consumer considered dead.
max.poll.records
Maximum records returned by single poll() call.
Kafka batch.size
Maximum size of message batch before sending.
kafka linger.ms
How long producer waits to accumulate messages into batch.
replica.fetch.max.bytes
Maximum bytes follower fetches from leader per request.
message.max.bytes
Maximum size of single message batch broker accepts.
Kafka rebalance protocol (Eager vs Cooperative)
Eager stops all consumers; Cooperative only moves needed partitions.
CooperativeStickyAssignor
Incremental cooperative rebalancing with minimal partition movement.
Kafka partition skew
Uneven message distribution across partitions.
Kafka time lag
Elapsed time between message production and consumption.
Kafka claim check pattern
Stores large payloads externally, puts reference in Kafka.
Kafka RecordBatch
Wire format container grouping records for transmission.
Semantic Versioning (SemVer)
MAJOR.MINOR.PATCH scheme communicating compatibility changes.
Major, minor, and patch bumps (API)
Patch fixes bugs, Minor adds features, Major breaks compatibility.
Breaking change (API)
Modification causing existing clients to fail or require updates.
Calendar Versioning (CalVer)
Version tied to release dates rather than compatibility signals.
URL versioning vs header versioning
URL embeds version in path; header uses custom headers.
Sunset header (API)
Communicates when resource or version will be decommissioned.
API deprecation strategy
Plan for retiring API while giving consumers migration time.
Consumer-driven contracts
Consumers define expectations; provider runs contracts as tests.
PgBouncer
Lightweight PostgreSQL connection pooler that multiplexes many clients onto a small pool of DB connections.