Can't figure out how to use extendData when my y-axis is a numpy array

Dash Python
1

I have a pandas dataframe with the x column being timestamps and y column being integers. I create a line chart using Plotly Express. So far so good.

Now I want to use extendData. I am sure my return value is in the correct format, but it throws the following error in browser:

cannot extend missing or non-array attribute: y

On a closer inspection of the Figure object I found out that the y axis object looks like

{
        "dtype": "i2",
        "bdata": "AgAAAAEAAAAAAAAAAAAAAAA...",
        "_inputArray": {
          "0": 2, // my actual data
          "1": 0,
          "2": 1,
          ...
        }
}

suggesting that it was a numpy array, which works with plotting but likely not with extendTraces which expects an array of numbers.

To confirm my hypothesis I converted the y column to strings of numbers. The error went away but now the plot behaves erratically. So is there a way to force the y column to be a vanilla array?

For reference, my update function looks like this:

@app.callback(
    Output("time-series", "extendData"),
    Input("time-series", "figure"),
    Input("time-series-interval", component_property="n_intervals")
)
def update_time_series(f, n):
    if n == 0:
        return None
    now_time = datetime.now(timezone.utc).replace(second=0, microsecond=0)
    print(f"update_time_series: {n} {now_time}")
    df = ... # get new dataframe since now_time
    if df.empty:
        return (dict(x=[], y=[]), [])
    df = df.tail(1)
    x = []
    y = []
    traces = []
    for i, trace in enumerate(f['data']):
        traces.append(i)
        x.append(df[df['type'] == trace['name']]['date'].dt.strftime('%Y-%m-%dT%H:%M:%S.%f%z').tolist())
        y.append(df[df['type'] == trace['name']]['count'].tolist())
    
    return (dict(x=x, y=y), traces)