Feature Flag Service — System Design Academy

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 feature flag service 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 Billions evals/s · < 1 ms local · updates in seconds.
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 feature flag service 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 Billions evals/s · < 1 ms local · updates in seconds. Convert it into peak throughput, retained records, bandwidth, and failure budget; then connect each number to a component.

Dimension	Working estimate	Design implication
Traffic	Billions evals/s · < 1 ms local · updates in seconds	Separate online compute, durable state, and asynchronous consumers.
Peak factor	5–10× average	Admission control, autoscaling headroom, and hot-key protection.
Stored state	write rate × retention × record bytes	Partition count, indexes, compaction, backup, and archival tier.
Network	peak operations × payload size	Batching, compression, edge placement, and egress budget.
Availability	99.9–99.99%	Multi-zone replicas, tested failover, and dependency error budgets.
Recovery	RPO/RTO by invariant	Ack 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

PUT /v1/flags/{key}validated version

POST /v1/flags/{key}/rolloutsstaged rollout

GET /v1/sdk/snapshotsigned snapshot

STREAM /v1/sdk/updatesordered deltas

Authoritative records and indexes

flag(project, key, version, rules, default, state)

environment(project, env, active_snapshot)

rollout(flag_key, stage, cohort, health_gate, state)

exposure(subject_hash, flag_key, variation, version)

Command and event envelopes

# Command envelope
operation: PUT /v1/flags/{key}
request_id: req_01J...
idempotency_key: idem_7f2...
aggregate_key: feature-flags_123
expected_version: 17
payload: {...}

# Durable event
event_id: evt_01J...
aggregate_key: feature-flags_123
event_type: feature_flags_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.

Component	Owned responsibility	Why separate it	Rejected alternative	Cost and evolution trigger
Application SDK	Owns client protocol, request identity, retry, and resume behavior. In this design it participates in: Validate rule syntax, prerequisites, and permissions before storing immutable version.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents the external caller from also owning the contract and failure behavior of In-Process Evaluator.	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 “Prerequisite cycle enters config” to the external caller.	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 Application SDK boundary before extracting or scaling it.
In-Process Evaluator	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Build and sign a complete environment snapshot, then append its version to distribution log.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Application SDK from also owning the contract and failure behavior of Local Snapshot Cache.	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 “SDK misses deltas during disconnect” to Application SDK.	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 In-Process Evaluator boundary before extracting or scaling it.
Local Snapshot Cache	Serves a low-latency, disposable projection with authoritative fallback. In this design it participates in: Relays stream ordered deltas; SDK atomically swaps snapshots and retains last-known-good.	Removes repeated work from the latency-critical path. It prevents In-Process Evaluator from also owning the contract and failure behavior of Streaming Relay.	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 “Percentage assignment changes after rule edit” to In-Process Evaluator.	Costs invalidation, staleness, stampede protection, and memory. Add or split it only after the workload (Billions evals/s · < 1 ms local · updates in seconds) proves repeated reads dominate. Measure the latency, saturation, lag, or mismatch at the Local Snapshot Cache boundary before extracting or scaling it.
Streaming Relay	Preserves durable ordering, replay, and consumer checkpoints. In this design it participates in: Evaluator hashes stable subject attributes for deterministic percentage assignment.	Decouples bursts and failure from user acknowledgment. It prevents Local Snapshot Cache from also owning the contract and failure behavior of Edge Relay Tier.	Call every downstream consumer synchronously. That couples user latency and availability to analytics, notification, indexing, and audit dependencies. Here that simpler choice would couple “Bad rollout harms one cohort” to Local Snapshot Cache.	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 Streaming Relay boundary before extracting or scaling it.
Edge Relay Tier	Owns authentication, validation, quotas, deadlines, and routing. In this design it participates in: Rollout controller advances cohorts only while health gates pass; kill path promotes safe default.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Streaming Relay from also owning the contract and failure behavior of Config Distribution Log.	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 “Control plane unavailable during emergency” to Streaming Relay.	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 Edge Relay Tier boundary before extracting or scaling it.
Config Distribution Log	Preserves durable ordering, replay, and consumer checkpoints. In this design it participates in: Validate rule syntax, prerequisites, and permissions before storing immutable version.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Edge Relay Tier from also owning the contract and failure behavior of Flag Control Plane.	Call every downstream consumer synchronously. That couples user latency and availability to analytics, notification, indexing, and audit dependencies. Here that simpler choice would couple “Prerequisite cycle enters config” to Edge Relay Tier.	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 Config Distribution Log boundary before extracting or scaling it.
Flag Control Plane	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Build and sign a complete environment snapshot, then append its version to distribution log.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Config Distribution Log from also owning the contract and failure behavior of Validation / Policy.	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 “SDK misses deltas during disconnect” to Config Distribution Log.	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 Flag Control Plane boundary before extracting or scaling it.
Validation / Policy	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Relays stream ordered deltas; SDK atomically swaps snapshots and retains last-known-good.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Flag Control Plane from also owning the contract and failure behavior of Versioned Config DB.	Embed policy in every caller. Rules then drift by deployment version and emergency changes require fleet-wide redeployment. Here that simpler choice would couple “Percentage assignment changes after rule edit” to Flag Control Plane.	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 Validation / Policy boundary before extracting or scaling it.
Versioned Config DB	Owns authoritative versioned state or immutable evidence. In this design it participates in: Evaluator hashes stable subject attributes for deterministic percentage assignment.	Defines the source of truth and commit boundary. It prevents Validation / Policy from also owning the contract and failure behavior of Rollout Controller.	Let each service persist its own interpretation. That removes one clear commit boundary and makes retries, support, and reconciliation ambiguous. Here that simpler choice would couple “Bad rollout harms one cohort” to Validation / Policy.	Costs schema migration, replication, backup, and shard ownership. Split storage only when access patterns, retention, or compliance boundaries materially differ. Measure the latency, saturation, lag, or mismatch at the Versioned Config DB boundary before extracting or scaling it.
Rollout Controller	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Rollout controller advances cohorts only while health gates pass; kill path promotes safe default.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Versioned Config DB from also owning the contract and failure behavior of Experiment Assigner.	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 “Control plane unavailable during emergency” to Versioned Config DB.	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 Rollout Controller boundary before extracting or scaling it.
Experiment Assigner	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Validate rule syntax, prerequisites, and permissions before storing immutable version.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Rollout Controller from also owning the contract and failure behavior of Exposure Stream.	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 “Prerequisite cycle enters config” to Rollout Controller.	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 Experiment Assigner boundary before extracting or scaling it.
Exposure Stream	Preserves durable ordering, replay, and consumer checkpoints. In this design it participates in: Build and sign a complete environment snapshot, then append its version to distribution log.	Decouples bursts and failure from user acknowledgment. It prevents Experiment Assigner from also owning the contract and failure behavior of Audit Log.	Call every downstream consumer synchronously. That couples user latency and availability to analytics, notification, indexing, and audit dependencies. Here that simpler choice would couple “SDK misses deltas during disconnect” to Experiment Assigner.	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 Exposure Stream boundary before extracting or scaling it.
Audit Log	Preserves durable ordering, replay, and consumer checkpoints. In this design it participates in: Relays stream ordered deltas; SDK atomically swaps snapshots and retains last-known-good.	Proves correctness and supports controlled recovery. It prevents Exposure Stream from also owning the contract and failure behavior of Emergency Kill Path.	Call every downstream consumer synchronously. That couples user latency and availability to analytics, notification, indexing, and audit dependencies. Here that simpler choice would couple “Percentage assignment changes after rule edit” to Exposure Stream.	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 Audit Log boundary before extracting or scaling it.
Emergency Kill Path	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Evaluator hashes stable subject attributes for deterministic percentage assignment.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Audit Log from also owning the contract and failure behavior of operators and repair tooling.	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 “Bad rollout harms one cohort” to Audit Log.	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 Emergency Kill Path boundary before extracting or scaling it.

Decision 01

Chosen boundary vs simpler monolith

Keep evaluation in-process using versioned snapshots; the network distributes config but is not on the hot path.

A single deployable with one transactional store is the right starting point for a small feature flag service workload. The proposed boundary becomes justified when keep evaluation in-process using versioned snapshots; the network distributes config but is not on the hot path. Separating it contains the named failure “Prerequisite cycle enters config” without placing unrelated work on the critical path.

Decision 02

Correctness vs availability

Model targeting, percentage rollout, prerequisites, and defaults deterministically by stable subject hash.

The design coordinates only where each accepted feature flag service 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

Sign snapshots, preserve audit history, and provide kill switches with fast propagation.

This choice adds dashboards, ownership, schema compatibility, capacity planning, and recovery runbooks. Start with the minimum topology that supports “Relays stream ordered deltas; SDK atomically swaps snapshots and retains last-known-good.” 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 feature flag service 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

Keep evaluation in-process using versioned snapshots; the network distributes config but is not on the hot path.

Scale boundary

Model targeting, percentage rollout, prerequisites, and defaults deterministically by stable subject hash.

Operational boundary

Sign snapshots, preserve audit history, and provide kill switches with fast propagation.

↑ top

5. Critical flow and algorithm

Validate rule syntax, prerequisites, and permissions before storing immutable version.

Build and sign a complete environment snapshot, then append its version to distribution log.

Relays stream ordered deltas; SDK atomically swaps snapshots and retains last-known-good.

Evaluator hashes stable subject attributes for deterministic percentage assignment.

Rollout controller advances cohorts only while health gates pass; kill path promotes safe default.

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 feature flag service 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.

Operation	Consistency	Reason
Authoritative mutation	Conditional / serializable	Protect uniqueness, ownership, money, inventory, order, or legal state transition.
Read after own write	Primary, session token, or cache warm	The caller must observe the result it just committed.
Ordinary read projection	Bounded stale where safe	Replicas and caches improve latency when version/freshness is visible.
Analytics and operations	Eventual + replay	Freshness 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

Failure	Runtime behavior	Recovery proof
Prerequisite cycle enters config	Stop 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.
SDK misses deltas during disconnect	Stop 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.
Percentage assignment changes after rule edit	Stop 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.
Bad rollout harms one cohort	Stop 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.
Control plane unavailable during emergency	Stop 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.

Recovery checkpoint

# Recovery checkpoint
scenario: Prerequisite cycle enters config
operation_id: op_01J...
aggregate_key: feature-flags_123
last_durable_version: 18
projection_checkpoint: partition-07@offset-482901
retry_policy: same_identity / bounded_backoff
reconciliation_status: pending
owner: Emergency Kill Path

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: Prerequisite cycle enters config — alert before the user-visible invariant is violated.

Regional evolution

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

↑ top

10. Interviewer follow-ups

How do percentage rollouts stay stable?

Keep evaluation in-process using versioned snapshots; the network distributes config but is not on the hot path. Walk the concrete lifecycle step “Validate rule syntax, prerequisites, and permissions before storing immutable version.” Then test it against “Prerequisite cycle enters config”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

What if the control plane is down?

Model targeting, percentage rollout, prerequisites, and defaults deterministically by stable subject hash. Walk the concrete lifecycle step “Build and sign a complete environment snapshot, then append its version to distribution log.” Then test it against “SDK misses deltas during disconnect”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do flags compose?

Sign snapshots, preserve audit history, and provide kill switches with fast propagation. Walk the concrete lifecycle step “Relays stream ordered deltas; SDK atomically swaps snapshots and retains last-known-good.” Then test it against “Percentage assignment changes after rule edit”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do you prevent stale unsafe flags?

Keep evaluation in-process using versioned snapshots; the network distributes config but is not on the hot path. Walk the concrete lifecycle step “Evaluator hashes stable subject attributes for deterministic percentage assignment.” Then test it against “Bad rollout harms one cohort”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do experiments use exposure events?

Model targeting, percentage rollout, prerequisites, and defaults deterministically by stable subject hash. Walk the concrete lifecycle step “Rollout controller advances cohorts only while health gates pass; kill path promotes safe default.” Then test it against “Control plane unavailable during emergency”. 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

Time	What to cover
0–5 min	Clarify scope, workload, SLOs, and the correctness invariant.
5–10 min	Estimate throughput/storage and define concrete APIs and records.
10–20 min	Draw online, durable-write, and asynchronous/repair paths.
20–30 min	Explain the exact lifecycle, partition key, and concurrency control.
30–38 min	Discuss caching, hot keys, replication, and backpressure.
38–43 min	Run failure drills, security, observability, and regional evolution.
43–45 min	Summarize the five most important choices and trade-offs.

Final decision summary

Invariant	each accepted feature flag service operation has one stable result and retries do not repeat its business effect.
Partitioning	Stable aggregate/key hash with versioned ownership and measured hot-key isolation.
Commit	Expected version or uniqueness at the authoritative state boundary.
Async work	Durable outbox/log, idempotent consumers, checkpoints, and replay.
Recovery	Same operation identity, durable truth lookup, and business reconciliation.

Closing summary

CLOSING “The design protects each accepted feature flag service 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

Application SDK

In-Process Evaluator

Local Snapshot Cache

Streaming Relay

Complete System Design Tutorial

1. Requirements and scope

Functional requirements

Non-functional requirements

2. Capacity assumptions

3. API and data model

Authoritative records and indexes

Command and event envelopes

4. Why this architecture and what it costs

Chosen boundary vs simpler monolith

Correctness vs availability

Operational cost and evolution

5. Critical flow and algorithm

Retry-safe mutation skeleton

6. Caching, hot keys, and backpressure

7. Partitioning and consistency

Strong boundary

Eventual boundary

8. Failure drills and recovery

Recovery checkpoint

9. Security, regions, and observability

Security and abuse

Signals that prove correctness

Regional evolution

10. Interviewer follow-ups

11. How to present this in 45 minutes

Final decision summary

Closing summary