# Document CRUD and the Bulk API — Elasticsearch

Source: https://www.skillbyai.com/en/elasticsearch/m-crud

> Index, get, update, delete, in batches.

## Writing documents

`PUT /index/_doc/<id>` creates or replaces a document with a known id; `POST /index/_doc` lets Elasticsearch generate one; `PUT /index/_create/<id>` fails if the id already exists. `GET /index/_doc/<id>` reads a document by id in real time, while searches only see changes after a refresh. `POST /index/_update/<id>` applies a partial update (internally Elasticsearch re-indexes the whole document, since segments are immutable), and `DELETE` removes one. Every write returns a `_seq_no` and `_primary_term`, which you can pass back as `if_seq_no`/`if_primary_term` for **optimistic concurrency control**. For many documents use the **bulk API**: a newline-delimited JSON (NDJSON) body of action lines and source lines, which must end with a newline.

## Getting data in with the right shape

Documents are written with CRUD and bulk APIs, and the mapping decides how every field is indexed.

![Three ideas: document APIs, field types, explicit versus dynamic mapping.](assets/figures/elasticsearch/section-2-map.svg) — Figure 2.1 — Documents flow through the mapping into indexed fields.

## Single writes and a bulk request

Kibana Dev Tools console syntax; send the same requests with curl or a client library.

```http
PUT /products/_doc/42
{ "name": "Rain jacket", "price": 120, "tags": ["outdoor"] }

POST /products/_update/42
{ "doc": { "price": 99 } }

GET /products/_doc/42

DELETE /products/_doc/42

# bulk: one action line, then (for index/create/update) one source line
POST /_bulk
{ "index":  { "_index": "products", "_id": "1" } }
{ "name": "Trail shoe", "price": 89.9 }
{ "create": { "_index": "products", "_id": "2" } }
{ "name": "Wool socks", "price": 12 }
{ "update": { "_index": "products", "_id": "1" } }
{ "doc": { "price": 79.9 } }
{ "delete": { "_index": "products", "_id": "3" } }

```

## Always check the bulk errors flag

A bulk call can return HTTP 200 while individual items failed. Check the top-level "errors" field and inspect each item's status, then retry only the failed items (for example 429 rejections) with backoff.

**Quiz:** What format does the bulk API body use?

- [ ] A single JSON array of documents
- [x] Newline-delimited JSON with action and source lines
- [ ] CSV with a header row
- [ ] XML with one element per document

*Answer:* Newline-delimited JSON with action and source lines. NDJSON lets the coordinating node split the request without parsing it as one big document.