---
name: servicex
description: >-
  Use when delivering analysis-ready columns from remote HEP datasets with
  ServiceX: defining FileList, XRootD, Rucio, or CERN Open Data inputs, choosing
  an uproot query, limiting branches and events, and diagnosing delivery or
  authorization failures.
---

# ServiceX

ServiceX performs server-side selection and transformation so an analysis
downloads only needed events and columns. Use the query type supported by the
deployed ServiceX instance and the actual input format.

## Minimal uproot-raw request

```python
from servicex import dataset, deliver, query

request = {
    "Sample": [
        {
            "Name": "sample",
            "Dataset": dataset.FileList(
                ["root://storage.example.org//path/to/file.root"]
            ),
            "Query": query.UprootRaw(
                [
                    {
                        "treename": "Events",
                        "filter_name": ["mass", "weight"],
                        "cut": "mass > 0",
                    }
                ]
            ),
        }
    ]
}

delivered = deliver(request)
print(delivered["sample"])
```

Choose the dataset source deliberately:

- `dataset.FileList([...])` for explicit HTTP or XRootD URLs.
- `dataset.XRootD("root://.../*")` for an endpoint-side wildcard.
- `dataset.Rucio(...)` only when the deployment can resolve and read that DID.
- `dataset.CERNOpenData(record_id)` for a supported public record.

Start with one file, a few branches, and a restrictive cut. Inspect the input
tree with uproot first, then confirm that the ServiceX deployment supports the
chosen query transformer. The service—not the local client—must be able to read
every source file. Do not treat delivery success as validation of physics
units, selections, or weights.

Use the current [ServiceX user guide](https://tryservicex.org/guide/) for query,
delivery, caching, and deployment-specific configuration.