Writing Parquet timestamps for BigQuery

Some things to keep in mind when working with pandas DatetimeIndex and looking to persist data using Parquet to back a table in BigQuery:

• timestamps must be physical Parquet columns (index metadata is ignored)
• store UTC instants (convert to UTC, then drop timezone metadata)
• BigQuery TIMESTAMP only supports microsecond precision

df.index = df.index.tz_convert(None)
df = df.reset_index()

df.to_parquet(
    buffer,
    engine="pyarrow",
    compression="snappy",
    index=False,
    coerce_timestamps="us",
    allow_truncated_timestamps=True,
    use_deprecated_int96_timestamps=False,
)

Without this, timestamps were loaded with the wrong type / incorrect values.