Reference · Cheat sheet

Shard sizing

The compressed essence. Print it, pin it.

From lesson 0008See also glossary

A shard costs (even when empty)

Heap — segment metadata + a little per mapped field, per segment.
One search thread per shard per query.
Cluster state entry gossiped to every node + file handles.

A small set of large shards uses fewer resources than many small shards.

Target numbers

Guideline	Value
Shard size	10–50 GB (50 GB rollover, 10 GB floor)
Docs per shard	< 200M
Shards per data node	≤ ~1,000 (non-frozen)
Master heap	~1 GB / 3,000 indices

Retired: the “20 shards per GB of heap” rule — deprecated in ES 8.3.

Size it

primaries ≈ ceil(total_GB / 40) → 300 GB ≈ 8 primaries. Then check per-node cap & docs headroom. Benchmark before committing.

Over-sharding = too many tiny shards

Symptoms: heap pressure, search thread-pool exhaustion, cluster-state bloat, slow searches, instability.
Common cause: many small indices (per-tenant / low-volume daily).
Fixes: _reindex to consolidate · _shrink to cut primaries · ILM rollover at size, not per-day.

Diagnose

GET /_cluster/health                       // shard counts
GET /_cat/nodes?v&h=name,heap.percent,node.role
GET /_cat/shards?v&s=index                 // size + node per shard

← lesson 0008 · Source: Size your shards