Hot Shard in MongoDB: Diagnose and Fix

Q: Hot Shard in MongoDB: Diagnose and Fix

A **hot shard** means one shard gets disproportionate writes while others sit idle — root cause is almost always a **bad shard key** (monotonically increasing `_id`, low-cardinality fields like `status`, or one dominant tenant). **Diagnose** with `sh.status()`, `getShardDistribution()`, `db.currentOp()`, and the profiler. **Short-term mitigation**: vertical scale the hot shard, split jumbo chunks, throttle problem producers, pin whales to their own zone. **Long-term fix**: run `reshardCollection` (5.0+) with a hashed or compound key so writes spread evenly. The lasting lesson is that **shard-key choice is the single most important design decision in MongoDB sharding** — it defines your scaling ceiling.

Intermediate·2 min read·

MongodbAtlasMetaLyftEbay

Fast recall

Quick revision

Core answer

A hot shard means one shard gets disproportionate writes while others sit idle — root cause is almost always a bad shard key (monotonically increasing _id, low-cardinality fields like status, or one dominant tenant). Diagnose with sh.status(), getShardDistribution(), db.currentOp(), and the profiler. Short-term mitigation: vertical scale the hot shard, split jumbo chunks, throttle problem producers, pin whales to their own zone. Long-term fix: run reshardCollection (5.0+) with a hashed or compound key so writes spread evenly. The lasting lesson is that shard-key choice is the single most important design decision in MongoDB sharding — it defines your scaling ceiling.

Interview-ready guide

Interview answer

Name the pattern fast

Hot shard = one shard owns the hottest chunks; almost always a bad shard key
Monotonic keys, low-cardinality keys, or whale values are the classic causes

Walk through diagnosis

Per-shard metrics, sh.status(), getShardDistribution()
Profiler and currentOp() for hot ops
Check jumbo chunks and the actual shard key

Offer immediate mitigation

Vertical scale the hot shard, manual splits, throttle producers, zones for whales, Kafka buffer

Propose the long-term fix

reshardCollection (5.0+) with hashed or compound key
Before 5.0: dual-write migration
Good keys: high cardinality, low frequency, non-monotonic, query-aligned

Close with prevention

Shard-key review in design docs
Load test with representative data
Alert on chunk distribution imbalance and jumbo chunks

03 · Deep dive