Service Discovery — 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 service discovery 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 10M instances · updates in seconds · lookup < 5 ms.
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 service discovery 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 10M instances · updates in seconds · lookup < 5 ms. Convert it into peak throughput, retained records, bandwidth, and failure budget; then connect each number to a component.

Dimension	Working estimate	Design implication
Traffic	10M instances · updates in seconds · lookup < 5 ms	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/instances/{id}register/renew lease

DELETE /v1/instances/{id}

GET /v1/services/{name}?generation=

WATCH /v1/services/{name}?after=

Authoritative records and indexes

instance(id PK, service, address, zone, metadata, lease)

service(name, policy, generation)

health(instance_id, status, observed_at, source)

change(revision, service, instance_id, operation)

Command and event envelopes

# Command envelope
operation: PUT /v1/instances/{id}
request_id: req_01J...
idempotency_key: idem_7f2...
aggregate_key: service-discovery_123
expected_version: 17
payload: {...}

# Durable event
event_id: evt_01J...
aggregate_key: service-discovery_123
event_type: service_discovery_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
Service Instance	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Authenticate instance and commit registration/lease under catalog generation.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents the external caller from also owning the contract and failure behavior of Registration Agent.	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 “Instance partitions from catalog but still serves” 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 Service Instance boundary before extracting or scaling it.
Registration Agent	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Probe health independently and transition availability only after policy thresholds.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Service Instance from also owning the contract and failure behavior of Identity / ACL.	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 “Health probes create false outage” to Service Instance.	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 Registration Agent boundary before extracting or scaling it.
Identity / ACL	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Append ordered membership change and build complete signed snapshots.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Registration Agent from also owning the contract and failure behavior of Health Prober.	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 “Client misses compacted revision” to Registration Agent.	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 Identity / ACL boundary before extracting or scaling it.
Health Prober	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Relays distribute snapshots/deltas; clients atomically swap local endpoint caches.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Identity / ACL from also owning the contract and failure behavior of Discovery API.	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 “Two regions disagree on endpoint” to Identity / ACL.	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 Health Prober boundary before extracting or scaling it.
Discovery API	Owns authentication, validation, quotas, deadlines, and routing. In this design it participates in: Clients load-balance locally and resume watch by revision or full snapshot after compaction.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Health Prober from also owning the contract and failure behavior of Consensus Catalog.	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 “Catalog unavailable during mass restart” to Health Prober.	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 Discovery API boundary before extracting or scaling it.
Consensus Catalog	Preserves durable ordering, replay, and consumer checkpoints. In this design it participates in: Authenticate instance and commit registration/lease under catalog generation.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Discovery API from also owning the contract and failure behavior of Lease Table.	Call every downstream consumer synchronously. That couples user latency and availability to analytics, notification, indexing, and audit dependencies. Here that simpler choice would couple “Instance partitions from catalog but still serves” to Discovery API.	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 Consensus Catalog boundary before extracting or scaling it.
Lease Table	Owns authoritative versioned state or immutable evidence. In this design it participates in: Probe health independently and transition availability only after policy thresholds.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Consensus Catalog from also owning the contract and failure behavior of Change Log.	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 “Health probes create false outage” to Consensus Catalog.	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 Lease Table boundary before extracting or scaling it.
Change Log	Preserves durable ordering, replay, and consumer checkpoints. In this design it participates in: Append ordered membership change and build complete signed snapshots.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Lease Table from also owning the contract and failure behavior of Snapshot Builder.	Call every downstream consumer synchronously. That couples user latency and availability to analytics, notification, indexing, and audit dependencies. Here that simpler choice would couple “Client misses compacted revision” to Lease Table.	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 Change Log boundary before extracting or scaling it.
Snapshot Builder	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Relays distribute snapshots/deltas; clients atomically swap local endpoint caches.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Change Log from also owning the contract and failure behavior of Regional Relay.	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 “Two regions disagree on endpoint” to Change 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 Snapshot Builder boundary before extracting or scaling it.
Regional Relay	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Clients load-balance locally and resume watch by revision or full snapshot after compaction.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Snapshot Builder from also owning the contract and failure behavior of Client Resolver.	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 “Catalog unavailable during mass restart” to Snapshot Builder.	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 Regional Relay boundary before extracting or scaling it.
Client Resolver	Owns client protocol, request identity, retry, and resume behavior. In this design it participates in: Authenticate instance and commit registration/lease under catalog generation.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Regional Relay from also owning the contract and failure behavior of Local Endpoint 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 “Instance partitions from catalog but still serves” to Regional Relay.	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 Client Resolver boundary before extracting or scaling it.
Local Endpoint Cache	Serves a low-latency, disposable projection with authoritative fallback. In this design it participates in: Probe health independently and transition availability only after policy thresholds.	Removes repeated work from the latency-critical path. It prevents Client Resolver from also owning the contract and failure behavior of Load Balancer.	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 “Health probes create false outage” to Client Resolver.	Costs invalidation, staleness, stampede protection, and memory. Add or split it only after the workload (10M instances · updates in seconds · lookup < 5 ms) proves repeated reads dominate. Measure the latency, saturation, lag, or mismatch at the Local Endpoint Cache boundary before extracting or scaling it.
Load Balancer	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Append ordered membership change and build complete signed snapshots.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Local Endpoint Cache from also owning the contract and failure behavior of Failure Detector.	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 “Client misses compacted revision” to Local Endpoint 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 Load Balancer boundary before extracting or scaling it.
Failure Detector	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Relays distribute snapshots/deltas; clients atomically swap local endpoint caches.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Load Balancer from also owning the contract and failure behavior of Audit / Metrics.	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 “Two regions disagree on endpoint” to Load Balancer.	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 Failure Detector boundary before extracting or scaling it.
Audit / Metrics	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Clients load-balance locally and resume watch by revision or full snapshot after compaction.	Proves correctness and supports controlled recovery. It prevents Failure Detector from also owning the contract and failure behavior of operators and repair tooling.	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 “Catalog unavailable during mass restart” to Failure Detector.	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 Audit / Metrics boundary before extracting or scaling it.

Decision 01

Chosen boundary vs simpler monolith

The catalog requires a consistent membership generation, but clients should resolve from local caches to keep discovery off the request hot path.

A single deployable with one transactional store is the right starting point for a small service discovery workload. The proposed boundary becomes justified when the catalog requires a consistent membership generation, but clients should resolve from local caches to keep discovery off the request hot path. Separating it contains the named failure “Instance partitions from catalog but still serves” without placing unrelated work on the critical path.

Decision 02

Correctness vs availability

Lease-based registration removes dead instances without trusting explicit deregistration; health status and ownership generation prevent stale resurrection.

The design coordinates only where each accepted service discovery 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

Push watches reduce staleness but require resume cursors and snapshot fallback; pure polling is simpler at small fleets.

This choice adds dashboards, ownership, schema compatibility, capacity planning, and recovery runbooks. Start with the minimum topology that supports “Append ordered membership change and build complete signed snapshots.” 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 service discovery 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

The catalog requires a consistent membership generation, but clients should resolve from local caches to keep discovery off the request hot path.

Scale boundary

Lease-based registration removes dead instances without trusting explicit deregistration; health status and ownership generation prevent stale resurrection.

Operational boundary

Push watches reduce staleness but require resume cursors and snapshot fallback; pure polling is simpler at small fleets.

↑ top

5. Critical flow and algorithm

Authenticate instance and commit registration/lease under catalog generation.

Probe health independently and transition availability only after policy thresholds.

Append ordered membership change and build complete signed snapshots.

Relays distribute snapshots/deltas; clients atomically swap local endpoint caches.

Clients load-balance locally and resume watch by revision or full snapshot after compaction.

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 service discovery 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
Instance partitions from catalog but still serves	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.
Health probes create false outage	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.
Client misses compacted revision	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.
Two regions disagree on endpoint	Resolve 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.
Catalog unavailable during mass restart	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: Instance partitions from catalog but still serves
operation_id: op_01J...
aggregate_key: service-discovery_123
last_durable_version: 18
projection_checkpoint: partition-07@offset-482901
retry_policy: same_identity / bounded_backoff
reconciliation_status: pending
owner: Audit / Metrics

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: Instance partitions from catalog but still serves — 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 are stale instances removed?

The catalog requires a consistent membership generation, but clients should resolve from local caches to keep discovery off the request hot path. Walk the concrete lifecycle step “Authenticate instance and commit registration/lease under catalog generation.” Then test it against “Instance partitions from catalog but still serves”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

Push or pull discovery?

Lease-based registration removes dead instances without trusting explicit deregistration; health status and ownership generation prevent stale resurrection. Walk the concrete lifecycle step “Probe health independently and transition availability only after policy thresholds.” Then test it against “Health probes create false outage”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do clients survive catalog outage?

Push watches reduce staleness but require resume cursors and snapshot fallback; pure polling is simpler at small fleets. Walk the concrete lifecycle step “Append ordered membership change and build complete signed snapshots.” Then test it against “Client misses compacted revision”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How is multi-region discovery handled?

The catalog requires a consistent membership generation, but clients should resolve from local caches to keep discovery off the request hot path. Walk the concrete lifecycle step “Relays distribute snapshots/deltas; clients atomically swap local endpoint caches.” Then test it against “Two regions disagree on endpoint”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How does DNS-based discovery compare?

Lease-based registration removes dead instances without trusting explicit deregistration; health status and ownership generation prevent stale resurrection. Walk the concrete lifecycle step “Clients load-balance locally and resume watch by revision or full snapshot after compaction.” Then test it against “Catalog unavailable during mass restart”. 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 service discovery 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 service discovery 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

Service Instance

Registration Agent

Identity / ACL

Health Prober

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