# HOW

## Recommendations
### Re-Indexing
* Set index settings and revert to original values after.
* set `number_of_replicas` to `0`
* set `refresh_interval` to `-1`
* set `version_type` to destination index to `internal`, unless we care about the versions from the old index?
* Use manual slicing when re-indexing to parallelize the process.
* Keep `shard_size` between 30 to 50 GB.

### Document Mapping
* For `geo_shape` types use `quadtree` strategy.

### Documents
* Convert `geo_shape` coordinate values from `string` to `float`.
* Use **Bulk API** when loading documents.

### Hardware
* Use memory optimized AWS instances.
* Use at least 3 master nodes to keep the cluster stable.
* Keep node space storage consumption under `85%`. If this goes over, the cluster stops assigning shards to that node.