Elasticsearch Python client

Use this skill when a task uses the official elasticsearch Python package or asks for Elasticsearch connections, indexing/search APIs, async client code, bulk helpers, the high-level DSL, ES|QL, vector-store helpers, or client troubleshooting. The source snapshot is package version 9.5.0; read repo-provenance.md before deciding whether the skill is stale for a different client checkout or version.

Route the request

• Connect, authenticate, configure transport, use generated API namespaces, or manage sync/async lifecycle: read client-operations.
• Bulk index/update/delete, streaming, scan, reindex, or helper failures: read helpers-ingest.
• Q, Search, aggregations, mappings, Document, typed persistence, or high-level query composition: read dsl-search.
• ES|QL source/processing builders, safe parameters, or tabular results: read esql-query-builder.
• Shared install, serialization, version, or security failures: read troubleshooting.md in addition to the route.

For a cross-cutting task, load client-operations first, then the workflow route; keep the client/auth configuration separate from request-body construction.

Install and smoke check

The package supports Python 3.10+. Install only the extras used by the task:

python -m pip install elasticsearch
python -m pip install "elasticsearch[async]"       # aiohttp async transport
python -m pip install "elasticsearch[requests]"    # requests transport
python -m pip install "elasticsearch[orjson]"      # optional serializer
python -m pip install "elasticsearch[pyarrow]"     # Arrow/ES|QL workflows
python -m pip install "elasticsearch[vectorstore_mmr]"  # MMR helper support

from elasticsearch import AsyncElasticsearch, Elasticsearch
print(Elasticsearch, AsyncElasticsearch)

An import proves package availability only. client.info() is a live service probe and requires a reachable endpoint, credentials, and correct TLS.

Guardrails

• Keep API keys, passwords, bearer tokens, CA files, and cluster URLs in runtime configuration or environment variables, not in generated guidance or logs.
• Prefer TLS verification with ca_certs or a supported fingerprint. Do not use verify_certs=False as a production solution.
• Generated client modules are specification output; do not edit files under elasticsearch/_sync/client/ or elasticsearch/_async/client/ to change an API. Upstream API changes belong in the Elasticsearch specification project.
• A bulk HTTP response can contain item failures. Inspect helper results and use bounded retries only for operations whose retry semantics are understood.
• A rendered DSL/ES|QL request is not server validation. Check target mappings, server version, privileges, response shape, and partial-result flags.
• Close clients: client.close() for sync and await client.close() for async, preferably with an async context manager.

Verification entry points

Use the bundled offline checks when no cluster is available:

• python sub-skills/helpers-ingest/scripts/bulk_actions_smoke.py
• python sub-skills/dsl-search/scripts/dsl_query_smoke.py
• python sub-skills/esql-query-builder/scripts/esql_query_smoke.py

Read routing metadata only when the managed repo-skill router is being rebuilt; it is structured metadata, not a user workflow reference.