Fix pandas read_parquet: Unable to Find a Usable Engine

The error

ImportError: Unable to find a usable engine; tried using: 'pyarrow', 'fastparquet'.
A suitable version of pyarrow or fastparquet is required for parquet support.
Trying to import the above resulted in these errors:
 - Missing optional dependency 'pyarrow'. pyarrow is required for parquet support.
 - Missing optional dependency 'fastparquet'. fastparquet is required for parquet support.

pandas cannot read Parquet by itself. pd.read_parquet and df.to_parquet are thin wrappers that hand the work to pyarrow or fastparquet. This error means neither import succeeded — it says nothing about your file.

Read the lines after "resulted in these errors" before doing anything else. They contain the real failure for each engine, and that is what decides which fix below applies.

Fix 1: install into the right interpreter

The common case is a plain missing dependency:

python -m pip install pyarrow
python -c "import pandas, pyarrow; print(pandas.__version__, pyarrow.__version__)"

Use python -m pip, not bare pip. A bare pip on PATH frequently belongs to a different environment than the python that runs your script, which is why "I already installed it" is the most common follow-up. With uv:

uv add pyarrow          # project
uv pip install pyarrow  # active venv

In a notebook, confirm the kernel first:

import sys; print(sys.executable)

If that path is not your project venv, switch the kernel — installing again will not help.

Fix 2: the import failed for a different reason

If the error tail shows something other than Missing optional dependency — a DLL load failed, a version conflict, or a build error — pyarrow is present but broken. Two frequent causes:

• No wheel for your Python version or platform. Just-released Python minors, musl-based images (Alpine), and unusual architectures may have no prebuilt wheel, so pip tries a source build and fails. Force the issue to surface with pip install --only-binary=:all: pyarrow, then either pin to a Python version that has wheels or switch base image.
• A pyarrow too old for your pandas. pandas enforces a minimum version and reports the same "usable engine" message. python -m pip install -U pyarrow resolves it.

Mixing conda and pip installs of pyarrow in one environment also produces broken imports; keep to one installer.

Fix 3: pick the engine explicitly

Since pandas 2.0, pyarrow is the default when installed. If you must use the other engine — typically because no pyarrow wheel exists — install it and say so:

df = pd.read_parquet("data.parquet", engine="fastparquet")

Expect gaps: fastparquet handles plain columnar data well but struggles with nested structs and newer logical types. The trade-offs are covered in fastparquet vs pyarrow.

Skip the engine entirely

If you only need to inspect the file, no Python engine is required at all. DuckDB reads Parquet natively:

duckdb -c "SELECT * FROM 'data.parquet' LIMIT 20"

Or open the file in the Parquet Viewer to see schema, types and a row sample, run aggregates in the SQL Workbench, or export it with the Parquet to CSV converter. These run locally in the browser on WebAssembly — no install, no upload, and no dependency on the environment that just broke.