02 / 60 · Foundations

Rate Limiter — Production Architecture

All designs

Protect APIs with predictable quotas across a distributed fleet.

1M requests/s · < 2 ms decision · 99.99% availability
1. High-level architecture and animated call flow

Choose a path, then play or step through exactly how the call moves between components.

← / → step · Space play
CLOUD / MULTI-AZ DEPLOYMENT RATE LIMITER · PRODUCTION TOPOLOGY 12 named components · solid = request/data dependency · warm curve = control, event, or repair path MULTI-ZONE · VERSIONED · REPLAYABLE 01 · REQUEST & CONTROL02 · DOMAIN & COORDINATION03 · DATA, ASYNC & OPERATIONSrequestauthenticateroutevalidateread / reservecommitpublishconsumeprojectobserverepairpolicy snapshotevent streamtelemetry 01 EXTERNAL API Client interaction boundary 02 EDGE Global Edge protocol · auth · quota 03 EDGE Gateway Worker protocol · auth · quota 04 SERVICE Local Token Cache low-latency read path 05 SERVICE Limiter Library domain subsystem 06 SERVICE Key Normalizer domain subsystem 07 SERVICE Redis Cluster low-latency read path 08 SERVICE Lua/Atomic Counter domain subsystem 09 SECURITY Policy Service decision & coordination 10 SECURITY Policy Snapshot CDN low-latency read path 11 OPS Decision Metrics evidence & operations 12 STREAM Quota Audit Log durable ordered events externaledgeservicedatastreamsecurityops
Step 1 / 6API Client

The caller starts the operation at API Client with a stable request and retry identity.

  1. 01 · API ClientThe caller starts the operation at API Client with a stable request and retry identity.
  2. 02 · Global EdgeNormalize tenant, identity, route, and cost into one stable limit key.
  3. 03 · Local Token CacheEvaluate a signed local policy snapshot before touching remote state.
  4. 04 · Policy ServiceSpend a small local token lease when available; otherwise execute one atomic Redis script.
  5. 05 · Quota Audit LogReturn allowed, remaining, reset, and retry-after without blocking on metrics.
  6. 06 · Decision MetricsStream decisions for billing, anomaly detection, and lease reconciliation.
2. Critical request and recovery sequence
MUTATION / COMMITAPI ClientACTOR 01Global EdgeACTOR 02Local TokenCacheACTOR 03Lua/AtomicCounterACTOR 04ONLINE / READ + RECOVERYAPI ClientACTOR 01Gateway WorkerACTOR 02Key NormalizerACTOR 03Quota AuditLogACTOR 04Normalize tenant, identity, route, / and cost into one stable limit key.Evaluate a signed local policy / snapshot before touching remoteSpend a small local token lease / when available; otherwise executecommitted result + eventSpend a small local token lease / when available; otherwise executeReturn allowed, remaining, reset, / and retry-after without blocking onStream decisions for billing, / anomaly detection, and leasecached/read result + cursortimeout → retry same identity → read committed truth → reconcile

API Client

  • Owns client protocol, request identity, retry, and resume behavior.
  • Stateless or lease-scoped; durable progress lives elsewhere.

Global Edge

  • Owns authentication, validation, quotas, deadlines, and routing.
  • Partitioned by aggregate/key and guarded by version, uniqueness, or fencing.

Gateway Worker

  • Owns authentication, validation, quotas, deadlines, and routing.
  • Stateless or lease-scoped; durable progress lives elsewhere.

Local Token Cache

  • Serves a low-latency, disposable projection with authoritative fallback.
  • Versioned/TTL projection; safe to rebuild.

Complete System Design Tutorial

The design separates the latency-sensitive online path, the correctness-critical durable transition, and asynchronous projection or repair work. Each chapter below explains one decision shown in the diagrams.

1. Requirements and scope

Functional requirements

  • Execute the primary rate limiter workflow and return an observable result.
  • Expose the concrete operations shown below with stable retry identities.
  • Allow operators to inspect, disable, repair, or replay ambiguous state.
  • Keep derived analytics and notifications off the correctness-critical path.

Non-functional requirements

  • Design for 1M requests/s · < 2 ms decision · 99.99% availability.
  • Protect the named business invariant under concurrency and partial failure.
  • Scale by an explicit aggregate/partition key without one global mutex.
  • Degrade intentionally when cache, queue, store, or regional ownership is unavailable.
Primary invariant: each accepted rate limiter operation has one stable result and retries do not repeat its business effect.
↑ top

2. Capacity assumptions

State assumptions before choosing technology. The supplied workload is 1M requests/s · < 2 ms decision · 99.99% availability. Convert it into peak throughput, retained records, bandwidth, and failure budget; then connect each number to a component.

DimensionWorking estimateDesign implication
Traffic1M requests/s · < 2 ms decision · 99.99% availabilitySeparate online compute, durable state, and asynchronous consumers.
Peak factor5–10× averageAdmission control, autoscaling headroom, and hot-key protection.
Stored statewrite rate × retention × record bytesPartition count, indexes, compaction, backup, and archival tier.
Networkpeak operations × payload sizeBatching, compression, edge placement, and egress budget.
Availability99.9–99.99%Multi-zone replicas, tested failover, and dependency error budgets.
RecoveryRPO/RTO by invariantAck boundary, replay retention, reconciliation, and operator runbooks.
Do not cargo-cult the numbers: show the arithmetic, name the uncertain variable, and say which production metric will replace the interview estimate.
↑ top

3. API and data model

POST /v1/policiespublish versioned quota
GET /v1/policies/{id}inspect effective policy
CHECK tenant:user:routeinternal allow/deny decision
GET /v1/usage/{key}bounded usage view

Authoritative records and indexes

policy(policy_id PK, scope, algorithm, capacity, refill_rate, version)

counter(limit_key, window, used, expires_at)

decision(event_id, key_hash, policy_version, allowed, remaining)

INDEX policy(scope, active_version)

Command and event envelopes

# Command envelope
operation: POST /v1/policies
request_id: req_01J...
idempotency_key: idem_7f2...
aggregate_key: rate-limiter_123
expected_version: 17
payload: {...}

# Durable event
event_id: evt_01J...
aggregate_key: rate-limiter_123
event_type: rate_limiter_committed
schema_version: 3
trace_id: tr_8ac...

The authoritative aggregate and idempotency record share a transactional or conditional-write boundary. Event streams, search indexes, caches, analytics, and dashboards are rebuildable projections.

↑ top

4. Why this architecture and what it costs

The boxes are not a shopping list. Each boundary exists because it owns a different invariant, scaling dimension, data lifecycle, security policy, or failure domain. The table makes those choices explicit and shows when the design should remain simpler.

ComponentOwned responsibilityWhy separate itRejected alternativeCost and evolution trigger
API ClientOwns client protocol, request identity, retry, and resume behavior. In this design it participates in: Normalize tenant, identity, route, and cost into one stable limit key.Separates a distinct scale, ownership, policy, or failure boundary. It prevents the external caller from also owning the contract and failure behavior of Global Edge.Put protocol, authentication, quotas, and business transitions in one process. Acceptable at low scale, but it couples external traffic risk to domain releases. Here that simpler choice would couple “Redis unavailable forces explicit fail-open/closed policy” to the external caller.Costs another hop, policy distribution, and a high-blast-radius fleet. Introduce it when many clients or services need one protocol and protection boundary. Measure the latency, saturation, lag, or mismatch at the API Client boundary before extracting or scaling it.
Global EdgeOwns authentication, validation, quotas, deadlines, and routing. In this design it participates in: Evaluate a signed local policy snapshot before touching remote state.Separates a distinct scale, ownership, policy, or failure boundary. It prevents API Client from also owning the contract and failure behavior of Gateway Worker.Put protocol, authentication, quotas, and business transitions in one process. Acceptable at low scale, but it couples external traffic risk to domain releases. Here that simpler choice would couple “Hot enterprise tenant saturates one counter shard” to API Client.Costs another hop, policy distribution, and a high-blast-radius fleet. Introduce it when many clients or services need one protocol and protection boundary. Measure the latency, saturation, lag, or mismatch at the Global Edge boundary before extracting or scaling it.
Gateway WorkerOwns authentication, validation, quotas, deadlines, and routing. In this design it participates in: Spend a small local token lease when available; otherwise execute one atomic Redis script.Separates a distinct scale, ownership, policy, or failure boundary. It prevents Global Edge from also owning the contract and failure behavior of Local Token Cache.Put protocol, authentication, quotas, and business transitions in one process. Acceptable at low scale, but it couples external traffic risk to domain releases. Here that simpler choice would couple “Clock boundary creates fixed-window burst” to Global Edge.Costs another hop, policy distribution, and a high-blast-radius fleet. Introduce it when many clients or services need one protocol and protection boundary. Measure the latency, saturation, lag, or mismatch at the Gateway Worker boundary before extracting or scaling it.
Local Token CacheServes a low-latency, disposable projection with authoritative fallback. In this design it participates in: Return allowed, remaining, reset, and retry-after without blocking on metrics.Removes repeated work from the latency-critical path. It prevents Gateway Worker from also owning the contract and failure behavior of Limiter Library.Read authoritative storage for every request. Simpler initially, but latency and source-of-truth load scale together and a hot key becomes a database incident. Here that simpler choice would couple “Policy versions differ across gateways” to Gateway Worker.Costs invalidation, staleness, stampede protection, and memory. Add or split it only after the workload (1M requests/s · < 2 ms decision · 99.99% availability) proves repeated reads dominate. Measure the latency, saturation, lag, or mismatch at the Local Token Cache boundary before extracting or scaling it.
Limiter LibraryOwns one bounded domain decision, transition, or operational capability. In this design it participates in: Stream decisions for billing, anomaly detection, and lease reconciliation.Separates a distinct scale, ownership, policy, or failure boundary. It prevents Local Token Cache from also owning the contract and failure behavior of Key Normalizer.Keep this behavior inside an adjacent service. Prefer that simpler option until independent scale, ownership, policy, or failure isolation is measurable. Here that simpler choice would couple “Retries consume quota twice without request identity” to Local Token Cache.Costs an interface, deployment, telemetry, ownership, and failure mode. Keep it in-process first; extract when independent scale or reliability pays for the boundary. Measure the latency, saturation, lag, or mismatch at the Limiter Library boundary before extracting or scaling it.
Key NormalizerOwns one bounded domain decision, transition, or operational capability. In this design it participates in: Normalize tenant, identity, route, and cost into one stable limit key.Separates a distinct scale, ownership, policy, or failure boundary. It prevents Limiter Library from also owning the contract and failure behavior of Redis Cluster.Keep this behavior inside an adjacent service. Prefer that simpler option until independent scale, ownership, policy, or failure isolation is measurable. Here that simpler choice would couple “Redis unavailable forces explicit fail-open/closed policy” to Limiter Library.Costs an interface, deployment, telemetry, ownership, and failure mode. Keep it in-process first; extract when independent scale or reliability pays for the boundary. Measure the latency, saturation, lag, or mismatch at the Key Normalizer boundary before extracting or scaling it.
Redis ClusterServes a low-latency, disposable projection with authoritative fallback. In this design it participates in: Evaluate a signed local policy snapshot before touching remote state.Removes repeated work from the latency-critical path. It prevents Key Normalizer from also owning the contract and failure behavior of Lua / Atomic Counter.Read authoritative storage for every request. Simpler initially, but latency and source-of-truth load scale together and a hot key becomes a database incident. Here that simpler choice would couple “Hot enterprise tenant saturates one counter shard” to Key Normalizer.Costs invalidation, staleness, stampede protection, and memory. Add or split it only after the workload (1M requests/s · < 2 ms decision · 99.99% availability) proves repeated reads dominate. Measure the latency, saturation, lag, or mismatch at the Redis Cluster boundary before extracting or scaling it.
Lua / Atomic CounterOwns one bounded domain decision, transition, or operational capability. In this design it participates in: Spend a small local token lease when available; otherwise execute one atomic Redis script.Separates a distinct scale, ownership, policy, or failure boundary. It prevents Redis Cluster from also owning the contract and failure behavior of Policy Service.Keep this behavior inside an adjacent service. Prefer that simpler option until independent scale, ownership, policy, or failure isolation is measurable. Here that simpler choice would couple “Clock boundary creates fixed-window burst” to Redis Cluster.Costs an interface, deployment, telemetry, ownership, and failure mode. Keep it in-process first; extract when independent scale or reliability pays for the boundary. Measure the latency, saturation, lag, or mismatch at the Lua / Atomic Counter boundary before extracting or scaling it.
Policy ServiceOwns one bounded domain decision, transition, or operational capability. In this design it participates in: Return allowed, remaining, reset, and retry-after without blocking on metrics.Separates a distinct scale, ownership, policy, or failure boundary. It prevents Lua / Atomic Counter from also owning the contract and failure behavior of Policy Snapshot CDN.Embed policy in every caller. Rules then drift by deployment version and emergency changes require fleet-wide redeployment. Here that simpler choice would couple “Policy versions differ across gateways” to Lua / Atomic Counter.Costs an interface, deployment, telemetry, ownership, and failure mode. Keep it in-process first; extract when independent scale or reliability pays for the boundary. Measure the latency, saturation, lag, or mismatch at the Policy Service boundary before extracting or scaling it.
Policy Snapshot CDNOwns one bounded domain decision, transition, or operational capability. In this design it participates in: Stream decisions for billing, anomaly detection, and lease reconciliation.Removes repeated work from the latency-critical path. It prevents Policy Service from also owning the contract and failure behavior of Decision Metrics.Read authoritative storage for every request. Simpler initially, but latency and source-of-truth load scale together and a hot key becomes a database incident. Here that simpler choice would couple “Retries consume quota twice without request identity” to Policy Service.Costs invalidation, staleness, stampede protection, and memory. Add or split it only after the workload (1M requests/s · < 2 ms decision · 99.99% availability) proves repeated reads dominate. Measure the latency, saturation, lag, or mismatch at the Policy Snapshot CDN boundary before extracting or scaling it.
Decision MetricsOwns one bounded domain decision, transition, or operational capability. In this design it participates in: Normalize tenant, identity, route, and cost into one stable limit key.Proves correctness and supports controlled recovery. It prevents Policy Snapshot CDN from also owning the contract and failure behavior of Quota Audit Log.Infer correctness from request logs. Logs do not prove durable business state, detect silent drift, or safely repair ambiguous outcomes. Here that simpler choice would couple “Redis unavailable forces explicit fail-open/closed policy” to Policy Snapshot CDN.Costs an interface, deployment, telemetry, ownership, and failure mode. Keep it in-process first; extract when independent scale or reliability pays for the boundary. Measure the latency, saturation, lag, or mismatch at the Decision Metrics boundary before extracting or scaling it.
Quota Audit LogPreserves durable ordering, replay, and consumer checkpoints. In this design it participates in: Evaluate a signed local policy snapshot before touching remote state.Proves correctness and supports controlled recovery. It prevents Decision Metrics from also owning the contract and failure behavior of operators and repair tooling.Call every downstream consumer synchronously. That couples user latency and availability to analytics, notification, indexing, and audit dependencies. Here that simpler choice would couple “Hot enterprise tenant saturates one counter shard” to Decision Metrics.Costs eventual consistency, schema evolution, replay tooling, idempotent consumers, and lag operations. Earn it when bursts or downstream failure must not block acknowledgment. Measure the latency, saturation, lag, or mismatch at the Quota Audit Log boundary before extracting or scaling it.
Decision 01

Chosen boundary vs simpler monolith

Choose token bucket for bursts, sliding-window counter for accuracy, or fixed windows for simplicity.

A single deployable with one transactional store is the right starting point for a small rate limiter workload. The proposed boundary becomes justified when choose token bucket for bursts, sliding-window counter for accuracy, or fixed windows for simplicity. Separating it contains the named failure “Redis unavailable forces explicit fail-open/closed policy” without placing unrelated work on the critical path.

Decision 02

Correctness vs availability

Make the failure mode explicit: fail-open for low-risk traffic, fail-closed for costly or sensitive operations.

The design coordinates only where each accepted rate limiter operation has one stable result and retries do not repeat its business effect. Everywhere else it accepts bounded staleness or asynchronous propagation. The alternative—strong coordination for every projection—simplifies the mental model but increases tail latency and turns partitions into broad write unavailability.

Decision 03

Operational cost and evolution

Shard by limit key, use atomic scripts, and keep local token caches to survive store latency.

This choice adds dashboards, ownership, schema compatibility, capacity planning, and recovery runbooks. Start with the minimum topology that supports “Spend a small local token lease when available; otherwise execute one atomic Redis script.” and extract another service only when measured throughput, blast radius, data lifecycle, or team ownership changes independently.

Why this design: the synchronous path contains only the decisions required to protect each accepted rate limiter operation has one stable result and retries do not repeat its business effect. Durable projections, notifications, indexing, analytics, and repair are isolated so their load or failure cannot corrupt the acknowledged operation.
Correctness boundary

Choose token bucket for bursts, sliding-window counter for accuracy, or fixed windows for simplicity.

Scale boundary

Make the failure mode explicit: fail-open for low-risk traffic, fail-closed for costly or sensitive operations.

Operational boundary

Shard by limit key, use atomic scripts, and keep local token caches to survive store latency.

↑ top

5. Critical flow and algorithm

01

Normalize tenant, identity, route, and cost into one stable limit key.

02

Evaluate a signed local policy snapshot before touching remote state.

03

Spend a small local token lease when available; otherwise execute one atomic Redis script.

04

Return allowed, remaining, reset, and retry-after without blocking on metrics.

05

Stream decisions for billing, anomaly detection, and lease reconciliation.

Retry-safe mutation skeleton

identity = authenticate(request)
prior = idempotency.get(identity, request.key)
if prior: return prior.result

current = authoritative_store.read(partition_key)
next_state, events = domain.transition(current, request)
authoritative_store.compare_and_set(current.version, next_state)
idempotency.commit(request.key, next_state.result)
outbox.append(events)
return next_state.result
Exactly-once business effect: the network can duplicate requests. Stable identities, conditional state transitions, immutable event IDs, and reconciliation make the user-visible effect happen once.
↑ top

6. Caching, hot keys, and backpressure

Cache policy

Cache immutable or slowly changing reads with versioned keys, TTL jitter, short negative entries, and request coalescing.

Hot partition or tenant

Measure per-key and per-tenant skew. Replicate safe reads, isolate the offender, salt only when ordering permits, and enforce budgets.

Backpressure

Bound queues, buffers, connection pools, and concurrency. Shed low-priority work and scale from lag/oldest-item age rather than CPU alone.

Stampede and retry storms

Use single-flight/coalescing, jittered expiry, bounded retries, exponential backoff, and circuit breakers so one dependency failure does not multiply traffic.

Cache outage rule: never allow every miss to fall through simultaneously. Use local protection, request coalescing, circuit breakers, bounded pools, and controlled fallback.
↑ top

7. Partitioning and consistency

partition = stable_hash(primary_or_aggregate_key) → shard owner
write = conditional_update(expected_version)
read_projection = versioned and rebuildable

Strong boundary

each accepted rate limiter operation has one stable result and retries do not repeat its business effect. Use a uniqueness constraint, serializable/conditional write, expected version, or fencing token at this boundary.

Eventual boundary

Analytics, notifications, search, read models, and operational dashboards may lag when freshness is observable and replay exists.

OperationConsistencyReason
Authoritative mutationConditional / serializableProtect uniqueness, ownership, money, inventory, order, or legal state transition.
Read after own writePrimary, session token, or cache warmThe caller must observe the result it just committed.
Ordinary read projectionBounded stale where safeReplicas and caches improve latency when version/freshness is visible.
Analytics and operationsEventual + replayFreshness is less important than isolating the online path.
  • Route every mutation for one aggregate to one owner/partition epoch.
  • Use cursors rather than offsets for changing ordered collections.
  • Retain tombstones until replication and repair windows close.
  • Move shard ownership through versioned maps and dual-read verification.
↑ top

8. Failure drills and recovery

FailureRuntime behaviorRecovery proof
Redis unavailable forces explicit fail-open/closed policyStop unsafe progress, serve only explicitly safe degraded behavior, and recover from durable state.Reconciliation returns zero invariant gaps; lag/replica health and user-visible SLO return to baseline.
Hot enterprise tenant saturates one counter shardIsolate the key/tenant/priority, apply admission control, and preserve durable work.Reconciliation returns zero invariant gaps; lag/replica health and user-visible SLO return to baseline.
Clock boundary creates fixed-window burstStop unsafe progress, serve only explicitly safe degraded behavior, and recover from durable state.Reconciliation returns zero invariant gaps; lag/replica health and user-visible SLO return to baseline.
Policy versions differ across gatewaysStop unsafe progress, serve only explicitly safe degraded behavior, and recover from durable state.Reconciliation returns zero invariant gaps; lag/replica health and user-visible SLO return to baseline.
Retries consume quota twice without request identityResolve at source of truth with uniqueness, expected version, lease generation, or monotonic cursor.Reconciliation returns zero invariant gaps; lag/replica health and user-visible SLO return to baseline.

Recovery checkpoint

# Recovery checkpoint
scenario: Redis unavailable forces explicit fail-open/closed policy
operation_id: op_01J...
aggregate_key: rate-limiter_123
last_durable_version: 18
projection_checkpoint: partition-07@offset-482901
retry_policy: same_identity / bounded_backoff
reconciliation_status: pending
owner: Quota Audit Log

Every dependency needs a timeout, retry budget, circuit policy, and explicit fail-open, fail-closed, stale, or read-only mode. “Highly available” is incomplete until degraded behavior is named.

↑ top

9. Security, regions, and observability

Security and abuse

  • Authenticate at ingress and authorize the exact resource/tenant on every operation.
  • Encrypt transport and durable data; tokenize or isolate sensitive material.
  • Rate-limit by tenant, actor, resource, and network; detect enumeration and anomalous fan-out.
  • Redact secrets and personal data before logs or analytical storage.
  • Audit privileged mutations, support access, policy changes, and forced recovery.

Signals that prove correctness

  • User-visible success rate and p50/p95/p99 latency.
  • Timeouts, circuit state, saturation, and replica health by dependency.
  • Queue lag, oldest-item age, retries, and dead-letter growth.
  • Conflict, duplicate-suppression, and reconciliation mismatch counts.
  • Business drill: Redis unavailable forces explicit fail-open/closed policy — alert before the user-visible invariant is violated.

Regional evolution

  1. Begin multi-zone with tested failover, backups, and reconciliation.
  2. Add global stateless ingress and regional read projections.
  3. Keep one write owner per aggregate or partition where coordination is expensive.
  4. Move ownership with epochs/fencing; avoid silent simultaneous writers.
  5. Use expand/contract schemas, canaries, shadow reads, and reversible cutovers.
↑ top

10. Interviewer follow-ups

How do you rate-limit across regions?

Choose token bucket for bursts, sliding-window counter for accuracy, or fixed windows for simplicity. Walk the concrete lifecycle step “Normalize tenant, identity, route, and cost into one stable limit key.” Then test it against “Redis unavailable forces explicit fail-open/closed policy”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do hierarchical limits work?

Make the failure mode explicit: fail-open for low-risk traffic, fail-closed for costly or sensitive operations. Walk the concrete lifecycle step “Evaluate a signed local policy snapshot before touching remote state.” Then test it against “Hot enterprise tenant saturates one counter shard”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

What happens when Redis is unavailable?

Shard by limit key, use atomic scripts, and keep local token caches to survive store latency. Walk the concrete lifecycle step “Spend a small local token lease when available; otherwise execute one atomic Redis script.” Then test it against “Clock boundary creates fixed-window burst”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do you prevent a hot tenant key?

Choose token bucket for bursts, sliding-window counter for accuracy, or fixed windows for simplicity. Walk the concrete lifecycle step “Return allowed, remaining, reset, and retry-after without blocking on metrics.” Then test it against “Policy versions differ across gateways”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do clients learn retry timing?

Make the failure mode explicit: fail-open for low-risk traffic, fail-closed for costly or sensitive operations. Walk the concrete lifecycle step “Stream decisions for billing, anomaly detection, and lease reconciliation.” Then test it against “Retries consume quota twice without request identity”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

Answer pattern: state the invariant, choose a mechanism, walk one failure, name the recovery signal, and explain the cost of the alternative.
↑ top

11. How to present this in 45 minutes

TimeWhat to cover
0–5 minClarify scope, workload, SLOs, and the correctness invariant.
5–10 minEstimate throughput/storage and define concrete APIs and records.
10–20 minDraw online, durable-write, and asynchronous/repair paths.
20–30 minExplain the exact lifecycle, partition key, and concurrency control.
30–38 minDiscuss caching, hot keys, replication, and backpressure.
38–43 minRun failure drills, security, observability, and regional evolution.
43–45 minSummarize the five most important choices and trade-offs.

Final decision summary

Invarianteach accepted rate limiter operation has one stable result and retries do not repeat its business effect.
PartitioningStable aggregate/key hash with versioned ownership and measured hot-key isolation.
CommitExpected version or uniqueness at the authoritative state boundary.
Async workDurable outbox/log, idempotent consumers, checkpoints, and replay.
RecoverySame operation identity, durable truth lookup, and business reconciliation.

Closing summary

CLOSING “The design protects each accepted rate limiter operation has one stable result and retries do not repeat its business effect. The online path stays bounded, authoritative state is versioned and partitioned, and asynchronous work is replayable and observable.”
↑ top