Why does sqlite (read-only) work running in Docker locally but not in Google Cloud Run?

I have a substantial sqlite database file (~2 million rows) which I query in a Rust application using rusqlite on Google Cloud Run, using a bundled version of sqlite.

I do not add or write data, and open the database connection in read-only mode. Running the container in Docker locally, queries work perfectly. In Cloud Run, the database is successfully connected to, but attempting to prepare a query panics with "Error preparing search statement: SqliteFailure(Error { code: NotADatabase, extended_code: 26 }, Some("file is not a database"))".

This is a little bit mystifying because the database file has already been connected to! I had a few hypotheses:

1. Something to do with a shortage of memory or attempting to write to Cloud Run's read-only file system? However, running in Docker with 256mb of memory and read-only mode enabled still worked.
2. A syscall prevented by gvisor? Running strace in docker showed no problems though, and there are no unsupported syscall logs.

Any suggestions as to what might be happening?

EDIT: just to add to the mystery: I wrote a quick Flask application to do the same thing using Python's built in sqlite, and get the same error...

EDIT 2: Using a database with a single row instead of 2 million, everything works as expected.

The original database is around 435MB in size.

To have a minimally reproducible example, the Python version simply receives a request, opens a connection, and queries:

@app.route('/detail/<int:rowid>/', methods=["GET"])
def get_individual_address(rowid):
    conn = sqlite3.connect(DB_NAME)
    cur = conn.cursor()
    cur.execute("""SELECT 
                postcode,
                town_city_or_rural_location,
                suburb,
                line_one
            FROM addresses WHERE id = ?""", (rowid,))
    if result := cur.fetchone():
        conn.close()
        return {
            "postcode": result[0],
            "town_city_or_rural_location": result[1],
            "suburb": result[2],
            "line_one": result[3]
        }
    else:
        conn.close()
        abort(404)

Similarly a simplified Dockerfile which still reproduces the problem is below:

FROM python:3.9-slim-buster
COPY requirements.txt /
RUN pip install -r requirements.txt
COPY . /app
EXPOSE 8080
ENV PORT 8080
WORKDIR /app
CMD exec gunicorn --bind :$PORT address_lookup:app --workers 1 --threads 1 --timeout 60