How to Troubleshoot Bokeh Server Periodic Callback Error

Community Support
1

So I have a Bokeh plot, server and a periodic callback function. The periodic callback function was working just fine yesterday, but today I am all of a sudden getting the error:

TypeError: 'NoneType' object is not callable

every time the callback is called. In order to initially populate some of the data on the plot, I call the callback function 10 times. The callback function works just fine when it is called while initializing the plot, it just breaks during the actual callback after it is loaded in the browser.

First, is there any reason this might be happening? I haven’t changed anything in my code since yesterday when it was working. Second, how can I troubleshoot this error? I have tried putting print statements into the code, but none of the print statements execute, it just throws the error. Additionally, the error displayed doesn’t show the actual line of code that is not working, which makes troubleshooting very difficult. How can I get around this? Is there a way to step through and debug the bokeh callback while it is actually running on the bokeh server? The callback works fine in all other cases.

Here is the full error thrown:

Error thrown from periodic callback:
Traceback (most recent call last):
File "/usr/local/lib/python3.6/site-packages/tornado/gen.py", line 526, in callback
        result_list.append(f.result())
File "/usr/local/lib/python3.6/site-packages/bokeh/server/session.py", line 67, in 
        _needs_document_lock_wrapper
        result = func(self, *args, **kwargs)
File "/usr/local/lib/python3.6/site-packages/bokeh/server/session.py", line 195, in 
        with_document_locked
        return func(*args, **kwargs)
File "/usr/local/lib/python3.6/site-packages/bokeh/document/document.py", line 1212, 
        in wrapper
        return doc._with_self_as_curdoc(invoke)
File "/usr/local/lib/python3.6/site-packages/bokeh/document/document.py", line 1198, 
        in _with_self_as_curdoc
        return f()
File "/usr/local/lib/python3.6/site-packages/bokeh/document/document.py", line 1211, 
        in invoke
        return f(*args, **kwargs)
        TypeError: 'NoneType' object is not callable

Here is the actual code from the callback function. It updates two plots and two data tables with ship positions and a scrolling spectrogram.

rollover = 185127
def update_spectrogram():
    # try to grab the next segment, if at the end of the wc object, then create a new one with new times
    try:
        segment = next(wc)
    except StopIteration:
        print("Stop iteration reached")
        #Change this to query the latest time from the predictions database, then user that time to query the
        #wavcrawler
        # compare old time to new time, if no difference, then return, continue checking but don't update
        # spectrogram data if no new data
        
        app_db = os.environ.get('DATABASE_URL') #or 'sqlite:///../app.db'

        app_engine = sqlalchemy.create_engine(app_db)
        app_connection = app_engine.connect()

        latest_time_query = "SELECT MAX(end_time) FROM PREDICTIONS"
        t2 = app_connection.execute(latest_time_query)

        t2 = int(t2.all()[0][0])
        t1 = t2-1800

        app_connection.close()
        app_engine.dispose()

        # if no new predictions, then continue without updating data
        # check if the current segments first timestamp is greater than the t1 of the new timestamp from the predictions
        # If it is not, then that means there are new predictions, so create a new wavcrawler
        # If it is, there are no new predictions, and the plot will stay paused at current values
        if int(spectrogram_source.data['Time'][0]) >= t1:
            print("No new predictions available, waiting...")
            return
        print("Creating new Wavcrawler object for new predictions")
        new_wc = WavCrawler(file,t1, t2, segment_length=8000, overlap=0.25)
        segment = next(new_wc)

    print("Creating spectrogram")
    signal = segment.samples[0, :]

    f, t, Sxx = spectrogram(signal, sample_rate)
    i=0
    df_length = f.shape[0] * t.shape[0]
    new_df_spectrogram = pd.DataFrame(np.nan, index=range(0,df_length), columns=['Frequency', 'Time', 'Sxx'])
    for freq in range(f.shape[0]):
        for time in range(t.shape[0]):
            new_df_spectrogram.loc[i] = [f[freq],t[time],Sxx[freq][time]]
            i = i+1

    new_df_spectrogram['Time'] = new_df_spectrogram['Time'] + float(segment.time_stamp)
    new_df_spectrogram['Time'] = pd.to_datetime(new_df_spectrogram['Time'], unit='s')

    # Data to keep in frame, should be desired number of seconds * 8000 (sample rate)

    spectrogram_source.stream(new_df_spectrogram, rollover=rollover)

    new_df_spectrogram_time = new_df_spectrogram.iloc[[new_df_spectrogram['Time'].idxmax()]]
    new_df_spectrogram_time = new_df_spectrogram_time.copy()
    
    new_df_spectrogram_time.loc[:,'str_time'] = new_df_spectrogram_time.loc[:,'Time'].dt.strftime('%Y-%b-%d %H:%M')
    text_source.data = ColumnDataSource.from_df(new_df_spectrogram_time)

    t1 = int(segment.time_stamp)

    app_db = os.environ.get('DATABASE_URL') #or 'sqlite:///../app.db'
    app_engine = sqlalchemy.create_engine(app_db)
    app_connection = app_engine.connect()
    print(app_engine)
    print(app_connection)
    print("Creating model predictions")
    #---------------------Update model predictions--------------------------------------
    query = "SELECT * FROM PREDICTIONS WHERE START_TIME <= " + str(t1) + " AND END_TIME >=" + str(t1)
    predictions = pd.read_sql_query(query, app_engine)

    if not predictions.empty:
        pred_json = json.loads(predictions['model_predictions'][0])
        pred_df = pd.DataFrame(pred_json).T
        pred_df.index = pred_df.index.rename("Model ID")
        prediction_df = pred_df.sort_values(by=['Model ID'])
        prediction_df = prediction_df.reset_index()
        prediction_df['Model Name'] = prediction_df.apply(lambda x: get_model_info(x['Model ID'], 'model_name', app_connection),axis=1)
        prediction_df['Model Type'] = prediction_df.apply(lambda x: get_model_info(x['Model ID'], 'model_type', app_connection),axis=1)
        prediction_df['Channels'] = prediction_df.apply(lambda x: get_model_info(x['Model ID'], 'channels', app_connection),axis=1)
    else:
        prediction_df = pd.DataFrame(data={"Model ID":[None],"Model Name":[None],"Model Type":[None],"Channels":[None],\
                                        "pred":[None],"pred_max_p":[None],"pred_vi_mean_max":[None],"entropy":[None],"nll":[None],"pred_std":[None],"var":[None],\
                                        "norm_entropy":[None],"epistemic":[None],"aleatoric":[None]})

    prediction_source.data = prediction_df

    t1_ais = t1 - (60*60*1) #(60*60*24)
    t2 = t1 + 1
    radius = 40

    query = 'SELECT * FROM AIS WHERE "timeOfFix" >= ' + str(t1_ais) + ' AND "timeOfFix" <= ' + str(t2) + \
            ' AND dist_from_sensor_km <=' + str(radius)

    new_ship_pos = pd.read_sql_query(query, app_engine)

    if new_ship_pos.empty:
        print("No ship positions for this time, check AIS stream")

    # Check if any ship exited, delete if they did
    allowed_mmsis = new_ship_pos.groupby('mmsi').agg({'timeOfFix':'max'})

    allowed_mmsis = allowed_mmsis.reset_index()

    allowed_mmsis = allowed_mmsis[allowed_mmsis.apply(lambda x: check_range(x['timeOfFix'], t1, radius, x['mmsi'], app_connection), axis=1)]['mmsi']

    new_ship_pos = new_ship_pos[new_ship_pos['mmsi'].isin(allowed_mmsis)]

    new_ship_pos['timeOfFix'] = pd.to_datetime(new_ship_pos['timeOfFix'], unit='s')
    
    source.data = ColumnDataSource.from_df(new_ship_pos)

    # Update source for lines
    new_ships_grouped_df = new_ship_pos.groupby('mmsi')
    colors_list = []
    class_list = []

    for key, data in new_ships_grouped_df:
        colors_list.append(color_dictionary[data['ship_class'].unique()[0]])
        class_list.append(data['ship_class'].unique()[0])

    new_ships_data = dict(
                    xs=[list(x[1]) for x in new_ships_grouped_df.merc_longitude],
                    ys=[list(y[1]) for y in new_ships_grouped_df.merc_latitude],
                    classes=class_list,
                    color=colors_list
                    )

    idx = new_ship_pos.groupby(['mmsi'])['timeOfFix'].transform(max) == new_ship_pos['timeOfFix']

    new_circle_ship_df = new_ship_pos[idx].copy()
    new_circle_ship_df['color'] =  new_circle_ship_df['ship_class'].map(color_dictionary)

    line_source.data = new_ships_data
    circle_source.data = ColumnDataSource.from_df(new_circle_ship_df)

    # Update ship classes present
    classes = []
    for mmsi, group in new_ships_grouped_df:
        classes.append(group['ship_class'].iloc[0])

    pred_labels = {"AIS Labels":classes}

    label_source.data = pred_labels

    app_connection.close()
    app_engine.dispose()

    return

EDIT
Ok so now I know that my code is not causing this issue. I commented out all of the code in the callback so that the only thing getting executed is this:

def update_spectrogram():
    print("Beginning")
    print("End")
    return

and I am still receiving this error. What would be the cause of this?

2

There is both too much and too little information above to speculate. Too much, because there is clearly parts that have no bearing on this question and are a distraction, and too little because the example code is not actually complete (i.e. runnable). For example:

Ok so now I know that my code is not causing this issue.

It’s entirely possible the problem is in your usage outside the callback. As things are, it’s impossible to say.

Please update the post to include a complete Minimal Reproducible Example.

To be clear: I am specifically asking you to take the effort to construct a toy example that is as small as possible, but still self-contained and runnable (e.g. with synthetic or dummy data). There’s a decent chance that this exercise in and of itself will lead you to the solution, but if not, then a providing a proper MRE is by far the most efficacious way to leverage the expertise in this forum.

3

Ok, so I attempted to create a reproducible example, but this issue is odd. I am running the bokeh server inside a Docker container. It was working before, but now it’s not, so I completely rebuilt the Docker environment and still got the same issue. This is the code that I’ve made as a test.

test_plot = figure(title="Spectrogram",x_axis_location="below", plot_width=650, plot_height=400)

test_plot.circle([1, 2, 3, 4, 5], [6, 7, 2, 4, 5], size=20, color="navy", alpha=0.5)

def callback():
    print("Start")
    print("End")
    return

curdoc().add_root(layout(test_plot))
curdoc().add_periodic_callback(callback(), 400)

This is how I start the server

bokeh serve test_plot.py --port 5011 --allow-websocket-origin '*'

When I run the server, I get the Nonetype error again on the callback function. I guess I am looking for recommendations on where to start troubleshooting this? Or what a potential origin of the problem could be? I am running Bokeh 2.3.0

1 Like
4

This is the problem:

curdoc().add_periodic_callback(callback(), 400)

The return value of excuting callback() is None, so you are passing None to add_periodic_callback.

You need to pass in the callback function, without actually executing it yourself (it is Bokeh’s job to execute callbacks):

curdoc().add_periodic_callback(callback, 400)  #  no ()
1 Like
5

Thank you, that fixed the issue!

6

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.