Web Search Engine — 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 web search engine 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 100B documents · 200K queries/s · p99 < 500 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 web search engine 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 100B documents · 200K queries/s · p99 < 500 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	100B documents · 200K queries/s · p99 < 500 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

GET /v1/search?q=&cursor=ranked result page

POST /internal/documentsindex version

POST /internal/removalsurgent tombstone

POST /v1/clicksexposure/click batch

Authoritative records and indexes

document(doc_id PK, canonical_url, content_hash, version)

posting(term, shard, doc_id, positions, field_weights)

index_segment(segment_id, range, version, object_uri)

query_event(request_id, candidates, ranks, clicks)

Command and event envelopes

# Command envelope
operation: GET /v1/search?q=&cursor=
request_id: req_01J...
idempotency_key: idem_7f2...
aggregate_key: web-search-engine_123
expected_version: 17
payload: {...}

# Durable event
event_id: evt_01J...
aggregate_key: web-search-engine_123
event_type: web_search_engine_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
Search Client	Owns client protocol, request identity, retry, and resume behavior. In this design it participates in: Normalize language, spelling, intent, and policy before retrieval.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents the external caller from also owning the contract and failure behavior of Global Query Edge.	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 “Popular query overloads one cache key” 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 Search Client boundary before extracting or scaling it.
Global Query Edge	Owns authentication, validation, quotas, deadlines, and routing. In this design it participates in: Route terms to index shards and retrieve a bounded union of candidates.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Search Client from also owning the contract and failure behavior of Query Understanding.	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 “Index segment versions differ” to Search 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 Query Edge boundary before extracting or scaling it.
Query Understanding	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Fetch document and learned features only for the candidate set.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Global Query Edge from also owning the contract and failure behavior of Spell / Intent.	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 “Spam document manipulates ranking” to Global Query Edge.	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 Query Understanding boundary before extracting or scaling it.
Spell / Intent	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Rank, dedupe hosts, apply safety, and generate snippets.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Query Understanding from also owning the contract and failure behavior of Index Router.	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 “Removal must beat stale cache” to Query Understanding.	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 Spell / Intent boundary before extracting or scaling it.
Index Router	Serves a low-latency, disposable projection with authoritative fallback. In this design it participates in: Log exposures/clicks and publish immutable index segments plus freshness overlays.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Spell / Intent from also owning the contract and failure behavior of Inverted Index Shards.	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 “Ranker exceeds latency budget” to Spell / Intent.	Costs invalidation, staleness, stampede protection, and memory. Add or split it only after the workload (100B documents · 200K queries/s · p99 < 500 ms) proves repeated reads dominate. Measure the latency, saturation, lag, or mismatch at the Index Router boundary before extracting or scaling it.
Inverted Index Shards	Serves a low-latency, disposable projection with authoritative fallback. In this design it participates in: Normalize language, spelling, intent, and policy before retrieval.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Index Router from also owning the contract and failure behavior of Freshness Overlay.	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 “Popular query overloads one cache key” to Index Router.	Costs invalidation, staleness, stampede protection, and memory. Add or split it only after the workload (100B documents · 200K queries/s · p99 < 500 ms) proves repeated reads dominate. Measure the latency, saturation, lag, or mismatch at the Inverted Index Shards boundary before extracting or scaling it.
Freshness Overlay	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Route terms to index shards and retrieve a bounded union of candidates.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Inverted Index Shards from also owning the contract and failure behavior of Document Store.	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 “Index segment versions differ” to Inverted Index Shards.	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 Freshness Overlay boundary before extracting or scaling it.
Document Store	Owns authoritative versioned state or immutable evidence. In this design it participates in: Fetch document and learned features only for the candidate set.	Defines the source of truth and commit boundary. It prevents Freshness Overlay from also owning the contract and failure behavior of Feature Service.	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 “Spam document manipulates ranking” to Freshness Overlay.	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 Document Store boundary before extracting or scaling it.
Feature Service	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Rank, dedupe hosts, apply safety, and generate snippets.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Document Store from also owning the contract and failure behavior of Ranker Fleet.	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 “Removal must beat stale cache” to Document Store.	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 Feature Service boundary before extracting or scaling it.
Ranker Fleet	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Log exposures/clicks and publish immutable index segments plus freshness overlays.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Feature Service from also owning the contract and failure behavior of Snippet 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 “Ranker exceeds latency budget” to Feature Service.	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 Ranker Fleet boundary before extracting or scaling it.
Snippet Service	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Normalize language, spelling, intent, and policy before retrieval.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Ranker Fleet from also owning the contract and failure behavior of Safety / Spam.	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 “Popular query overloads one cache key” to Ranker Fleet.	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 Snippet Service boundary before extracting or scaling it.
Safety / Spam	Owns one bounded domain decision, transition, or operational capability. In this design it participates in: Route terms to index shards and retrieve a bounded union of candidates.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Snippet Service from also owning the contract and failure behavior of Query Cache.	Embed policy in every caller. Rules then drift by deployment version and emergency changes require fleet-wide redeployment. Here that simpler choice would couple “Index segment versions differ” to Snippet Service.	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 Safety / Spam boundary before extracting or scaling it.
Query Cache	Serves a low-latency, disposable projection with authoritative fallback. In this design it participates in: Fetch document and learned features only for the candidate set.	Removes repeated work from the latency-critical path. It prevents Safety / Spam from also owning the contract and failure behavior of Click Stream.	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 “Spam document manipulates ranking” to Safety / Spam.	Costs invalidation, staleness, stampede protection, and memory. Add or split it only after the workload (100B documents · 200K queries/s · p99 < 500 ms) proves repeated reads dominate. Measure the latency, saturation, lag, or mismatch at the Query Cache boundary before extracting or scaling it.
Click Stream	Preserves durable ordering, replay, and consumer checkpoints. In this design it participates in: Rank, dedupe hosts, apply safety, and generate snippets.	Decouples bursts and failure from user acknowledgment. It prevents Query Cache from also owning the contract and failure behavior of Index 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 “Removal must beat stale cache” to Query 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 Click Stream boundary before extracting or scaling it.
Index Builder	Serves a low-latency, disposable projection with authoritative fallback. In this design it participates in: Log exposures/clicks and publish immutable index segments plus freshness overlays.	Separates a distinct scale, ownership, policy, or failure boundary. It prevents Click Stream from also owning the contract and failure behavior of operators and repair tooling.	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 “Ranker exceeds latency budget” to Click Stream.	Costs invalidation, staleness, stampede protection, and memory. Add or split it only after the workload (100B documents · 200K queries/s · p99 < 500 ms) proves repeated reads dominate. Measure the latency, saturation, lag, or mismatch at the Index Builder boundary before extracting or scaling it.

Decision 01

Chosen boundary vs simpler monolith

Use immutable inverted-index segments plus a small mutable freshness layer; merging at query time is cheaper than rewriting the corpus for every crawl.

A single deployable with one transactional store is the right starting point for a small web search engine workload. The proposed boundary becomes justified when use immutable inverted-index segments plus a small mutable freshness layer; merging at query time is cheaper than rewriting the corpus for every crawl. Separating it contains the named failure “Popular query overloads one cache key” without placing unrelated work on the critical path.

Decision 02

Correctness vs availability

Separate retrieval from ranking so broad recall remains cheap while expensive features run on a bounded candidate set.

The design coordinates only where each accepted web search engine 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 term/document strategy based on fan-out and skew; replicas absorb reads, not indexing correctness.

This choice adds dashboards, ownership, schema compatibility, capacity planning, and recovery runbooks. Start with the minimum topology that supports “Fetch document and learned features only for the candidate set.” 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 web search engine 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

Use immutable inverted-index segments plus a small mutable freshness layer; merging at query time is cheaper than rewriting the corpus for every crawl.

Scale boundary

Separate retrieval from ranking so broad recall remains cheap while expensive features run on a bounded candidate set.

Operational boundary

Shard by term/document strategy based on fan-out and skew; replicas absorb reads, not indexing correctness.

↑ top

5. Critical flow and algorithm

Normalize language, spelling, intent, and policy before retrieval.

Route terms to index shards and retrieve a bounded union of candidates.

Fetch document and learned features only for the candidate set.

Rank, dedupe hosts, apply safety, and generate snippets.

Log exposures/clicks and publish immutable index segments plus freshness overlays.

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 web search engine 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
Popular query overloads one cache key	Isolate 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.
Index segment versions differ	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.
Spam document manipulates ranking	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.
Removal must beat stale cache	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.
Ranker exceeds latency budget	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: Popular query overloads one cache key
operation_id: op_01J...
aggregate_key: web-search-engine_123
last_durable_version: 18
projection_checkpoint: partition-07@offset-482901
retry_policy: same_identity / bounded_backoff
reconciliation_status: pending
owner: Index Builder

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: Popular query overloads one cache key — 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 you rank documents?

Use immutable inverted-index segments plus a small mutable freshness layer; merging at query time is cheaper than rewriting the corpus for every crawl. Walk the concrete lifecycle step “Normalize language, spelling, intent, and policy before retrieval.” Then test it against “Popular query overloads one cache key”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How are new pages searchable quickly?

Separate retrieval from ranking so broad recall remains cheap while expensive features run on a bounded candidate set. Walk the concrete lifecycle step “Route terms to index shards and retrieve a bounded union of candidates.” Then test it against “Index segment versions differ”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do you remove spam?

Shard by term/document strategy based on fan-out and skew; replicas absorb reads, not indexing correctness. Walk the concrete lifecycle step “Fetch document and learned features only for the candidate set.” Then test it against “Spam document manipulates ranking”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do index shards rebalance?

Use immutable inverted-index segments plus a small mutable freshness layer; merging at query time is cheaper than rewriting the corpus for every crawl. Walk the concrete lifecycle step “Rank, dedupe hosts, apply safety, and generate snippets.” Then test it against “Removal must beat stale cache”. State the retry identity, durable checkpoint, recovery signal, and cost of the alternative.

How do snippets get generated?

Separate retrieval from ranking so broad recall remains cheap while expensive features run on a bounded candidate set. Walk the concrete lifecycle step “Log exposures/clicks and publish immutable index segments plus freshness overlays.” Then test it against “Ranker exceeds latency budget”. 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 web search engine 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 web search engine 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

Search Client

Global Query Edge

Query Understanding

Spell / Intent

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