Search Engine 2 - Ranking, Autocomplete & Global Scale

Interview-Ready Approach

1) Clarify Scope and SLOs

• •Problem statement: Scale to 1 B pages with PageRank, autocomplete, personalized results, and multi-region serving.
• •Design for a peak load target around 20,000 RPS (including burst headroom).
• •Indexed pages: ~1,000,000,000
• •Index size: ~100 TB
• •PageRank computation: Runs daily (batch)
• •Query throughput: ~100,000 QPS
• •Search latency (P99): < 300 ms

2) Capacity Planning Method

• •Convert traffic and growth constraints into request rate, storage growth, and concurrency budgets.
• •Keep at least 2-3x safety margin per tier (ingress, compute, storage, async workers).
• •Reserve explicit latency budgets per hop so p95 can be defended in review.

3) Architecture Decisions

• •Databases: Define a clear system-of-record and design read/write paths separately before adding optimizations.
• •Caching: Put cache on hot read paths first and pick cache-aside or write-through explicitly.
• •Sharding: Choose shard keys around access patterns and growth hotspots, not just data size.
• •Analytics: Maintain separate OLTP and analytics paths; stream events into a warehouse/time-series layer.
• •Geo Distribution: Route users to nearest region/edge while keeping write-consistency boundaries explicit.
• •Search: Use primary store for writes and async index updates for search relevance + scale.

4) Reliability and Failure Strategy

• •Use strong write constraints (transactions or conditional writes) and explicit backup/restore strategy.
• •Bound staleness with TTL + invalidation hooks for critical entities.
• •Support rebalancing and hotspot detection from day one.
• •Version event schemas and monitor drop/late-event rates.
• •Design region failover and data residency controls as first-class requirements.

5) Validation Plan

• •Run one peak-load test, one dependency-degradation test, and one failover test.
• •Verify idempotency for all retried writes and async consumers.
• •Track user-facing SLOs first: p95 latency, error rate, and successful throughput.

6) Trade-offs to Call Out in Interviews

• •Databases: SQL gives stronger transactional guarantees; NoSQL often gives better write scaling and flexibility.
• •Caching: Higher hit rate cuts latency/cost, but stale data and invalidation bugs become primary risks.
• •Sharding: Sharding improves horizontal scale but makes cross-shard queries and transactions harder.
• •Analytics: Analytics pipeline unlocks insights, but adds eventual consistency and governance overhead.
• •Geo Distribution: Global latency improves, but cross-region consistency and operations become harder.

Practical Notes

• •PageRank is computed offline using a distributed computation framework (MapReduce / Spark) over the link graph.
• •Autocomplete can use a prefix trie stored in-memory (or Redis) with the top-k completions pre-computed per prefix.
• •Partition the index by document so each query fans out to all shards in parallel, then merge-sort results.