# Shard imbalance

Symptom: uneven disk or CPU across shards; the balancer never catches up.

## Diagnose

Run `sh.status()` (or the Atlas sharding views). Look at chunk counts per shard and the balancer state. Chunks that cannot split or move are flagged jumbo.

## Confirm

- Jumbo chunks: usually a low-cardinality shard key value that outgrew the chunk size. They cannot be moved or split automatically.
- Balancer off or throttled: chunks pile up on one shard.
- Monotonic shard key: all new writes land on the max-key chunk of one shard.

## Fix

- Jumbo chunks need manual intervention: `clearJumboFlag` (requires atlasAdmin) after splitting manually, or reshard the collection with a better key.
- Turn the balancer on and let it drain; for big imbalances it takes hours, not minutes.
- A monotonic key is a design bug: plan a reshard with a hashed or compound key.

## Verify

Chunk counts converge across shards, no jumbo flags remain, and write load spreads evenly in the metrics.