ValueError(“Invalid value”) in basedatatypes.py

After upgrading to Plotly 5 (either 5.1.0 or 5.2.2, or 5.3.1, 5.4.0), I am seeing the error (given below) for various plots, at random. The issue goes away if Plotly is downgraded to 4.14.3.

File "plotly/basedatatypes.py", line 3952, in _index_is
raise ValueError("Invalid value")

Here are relevant reports from me and other users

I have a Dash app that plots several graphs. When the Dash app starts, some plots do not get displayed, and I see the error. This only occurs on the initial startup of the app. When the webpage is refreshed, the error does not re-appear, and all plots get displayed without errors.

If I downgrade to Plotly 4.14.3, the issue goes away entirely.

Here is a reproducible (almost reliably) example. Start the script, wait a few sec, access the page at localhost:5000. The failures happen at random on either of the four graphs or several of them at once.

Plotly 5.4.0
Dash 1.21.0 (or 2.0.0 does not matter)

Running the code with debug=False seems to trigger the error more reliably than with debug=True.

Note that this only occurs on the initial startup of the app and the first visit to the page. When the webpage is refreshed, the error does not re-appear, and all plots get displayed without errors.

The error message and the screenshot are given below.

import dash
import dash_core_components as dcc
import dash_html_components as html
import plotly.express as px
from dash.dependencies import Input, Output

def gen_plot():
    return px.bar([1, 2, 3], [4, 5, 4])

def init_callbacks(app):
    @app.callback(
        Output('graph1', 'figure'),
        Input('dropdown', 'value'),
    )
    def update1(_):
        return gen_plot()

    @app.callback(
        Output('graph2', 'figure'),
        Input('dropdown', 'value'),
    )
    def update2(_):
        return gen_plot()

    @app.callback(
        Output('graph3', 'figure'),
        Input('dropdown', 'value'),
    )
    def update3(_):
        return gen_plot()

    @app.callback(
        Output('graph4', 'figure'),
        Input('dropdown', 'value'),
    )
    def update4(_):
        return gen_plot()

def init_layout():
    return html.Div([
        dcc.Dropdown(
            id='dropdown',
            value=1,
        ),
        dcc.Graph(id='graph1'),
        dcc.Graph(id='graph2'),
        dcc.Graph(id='graph3'),
        dcc.Graph(id='graph4'),
    ])

app = dash.Dash(__name__)
app.layout = init_layout
init_callbacks(app)

app.run_server(debug=False, port=5000)

The error:

127.0.0.1 - - [29/Oct/2021 19:53:20] "POST /_dash-update-component HTTP/1.1" 500 -
[2021-10-29 19:53:20,832] ERROR in app: Exception on /_dash-update-component [POST]
Traceback (most recent call last):
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/flask/app.py", line 2070, in wsgi_app
    response = self.full_dispatch_request()
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/flask/app.py", line 1515, in full_dispatch_request
    rv = self.handle_user_exception(e)
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/flask/app.py", line 1513, in full_dispatch_request
    rv = self.dispatch_request()
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/flask/app.py", line 1499, in dispatch_request
    return self.ensure_sync(self.view_functions[rule.endpoint])(**req.view_args)
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/dash/dash.py", line 1096, in dispatch
    response.set_data(func(*args, outputs_list=outputs_list))
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/dash/dash.py", line 1017, in add_context
    output_value = func(*args, **kwargs)  # %% callback invoked %%
  File "/home/skhan/notebooks/scripts/dash_dev.py", line 37, in update4
    return gen_plot()
  File "/home/skhan/notebooks/scripts/dash_dev.py", line 8, in gen_plot
    return px.bar([1, 2, 3], [4, 5, 4])
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/express/_chart_types.py", line 365, in bar
    return make_figure(
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/express/_core.py", line 1931, in make_figure
    apply_default_cascade(args)
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/express/_core.py", line 969, in apply_default_cascade
    args["pattern_shape_sequence"] = [
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/express/_core.py", line 970, in <listcomp>
    bar.marker.pattern.shape for bar in args["template"].data.bar
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/graph_objs/bar/marker/_pattern.py", line 252, in shape
    return self["shape"]
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/basedatatypes.py", line 4693, in __getitem__
    elif self._props is not None and prop in self._props:
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/basedatatypes.py", line 4388, in _props
    return self.parent._get_child_props(self)
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/basedatatypes.py", line 4402, in _get_child_props
    if self._props is None:
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/basedatatypes.py", line 4388, in _props
    return self.parent._get_child_props(self)
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/basedatatypes.py", line 4402, in _get_child_props
    if self._props is None:
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/basedatatypes.py", line 4388, in _props
    return self.parent._get_child_props(self)
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/basedatatypes.py", line 4422, in _get_child_props
    child_ind = BaseFigure._index_is(children, child)
  File "/home/skhan/.virtualenvs/penv-dant-dash/lib/python3.8/site-packages/plotly/basedatatypes.py", line 3952, in _index_is
    raise ValueError("Invalid value")
ValueError: Invalid value

The screenshot

I am also hit by this. Is there any progress?

I have not heard anything yet, besides a workaround by LeoWY here https://community.plotly.com/t/valueerror-invalid-value-in-basedatatypes-py/55993/6

I found the same problem. Any progress apart from the workaround with the try?

Also facing the same issue with a similar setup (Plotly 5.6.0 and Dash 2.3.0) and use case. Exact same behaviour observed : dash app plotting several graphs and a random number of them are not being displayed at startup, with the same error thrown as stated above.
Would be glad to hear some news about it, at least some insights on how to tackle this issue.

Same issues here with Plotly 5.6.0 and Dash 2.1.0. For some reason the workaround with the try clause avoids errors.

Fixed by adding:
fig = go.Figure(layout=dict(template='plotly'))
before instantiating the figure.

e.g.
fig = go.Figure(layout=dict(template='plotly'))

fig = px.bar(temp_df, ...

@KnopKirrie

Thanks, @KnopKirrie. That helped.

I have 11 graphs in 4 tabs in a Dash app, which consistently produced the ValueError(“Invalid value”) error on random graphs. Until now, I was using the workaround with the try.

So, to implement your approach, I removed the try logic, and added figure=go.Figure(layout=dict(template='plotly')) for each dcc.Graph in the layout.

Note that previously, I did not have the value for the figure set in dcc.Graph at all. It was returned by a callback on the startup.

As an experiment, I also tried using figure=go.Figure(), but it did not help.

FYI - This same error is still occurring when using plotly/dash, in 2024. Luckily, the above fix by "KnopKirrie" still works.

In fact, you only have to include this line one time before any of your other plotly or dash code:

go.Figure(layout=dict(template='plotly'))

@plan-9-from-outer-space

@plan-9-from-outer-space , Can you please share the code the reproduces this error?
I just the code posted by the original poster, but I didn't get an error.

Original poster is here. The code sample I submitted in 2021 produces the error just fine. I have just tested it again in Dash 2.15.0 and got the same error.

As many other reporters clearly indicated, the nature of this error seems random. Try running the code a few times, and opening a fresh browser tab perphaps for each try?

Below is a link to a self-contained file of code that shows the error. I am using plotly 5.18.0 and dash 2.15.0 (both are the latest versions on pypi.org). I am using python 3.11.7. BTW, I also get an annoying warning from plotly, which gets printed to the terminal whenever I click on one of the dash tabs:

C:...\envs\PY1\Lib\site-packages\plotly\express_core.py:2065: FutureWarning: When grouping with a length-1 list-like, you will need to pass a length-1 tuple to get_group in a future version of pandas. Pass (name,) instead of name to silence this warning.

Below is the link to a file that shows the error. Just save locally and run "python dashboard.py".

I forgot to mention -- you will need to comment out line 26, which is the bug fix.

dashboard.py

@sergiykhan

The FutureWarning is being addressed in #4519

@sergiykhan and @plan-9-from-outer-space thanks for the self-contained apps. But I'm afraid I'm still unable to reproduce this error using either, no matter how much I restart or reload. Looks like it's happening both on Windows and on Linux, any other info those seeing this can give to help us reproduce?

I created a new python environment, installed only python 3.11.7, dash, and pandas, and the same error occurs. I am running Windows 11. Can you try doing that?

That's exactly what I did before, but upon doing it again I did see the error! Seems like there's some sort of a race condition when you fire multiple figure callbacks simultaneously, I couldn't get it to happen with only one graph callback. But also after one or two times showing the error it never happens again, no matter how many times I stop Python and restart it, and/or close and reopen the page, so it's like plotly.py has sorted itself out, maybe generated __pycache__ files or some such, and the problem never recurs.

Thanks for your patience - this is still going to be tricky to figure out but at least I've seen it now!

@alexcjohnson

@alexcjohnson Not sure if it helps, but here is a bit more info about my test (original post).

I keep the Chrome tab open at http://localhost:5000/, and hit F5 to refresh the page after the app is re-launched.

The algorithm: launch the app, hit F5, check if there is an error, close the app. Repeat.

Right now, I did about 15 attempts. Three attempts produced an error. The number of unrendered plots was between 1 and 3.

I do not recreate the Python virtual environment nor do I clean any files.

Still reproducible on 6.9.0. I tracked down the mechanism and have a candidate fix — details below in case they're useful.

Cause

BaseLayoutHierarchyType._get_prop does a check-then-act on a shared dict, with no synchronisation:

if self._compound_array_props.get(prop, None) is None:      # CHECK
    self._compound_array_props[prop] = [                    # ACT — not atomic
        validator.data_class(_parent=self)
        for _ in self._props.get(prop, [])
    ]
return validator.present(self._compound_array_props[prop])

Two threads can both find the entry empty, both build the list, and the later assignment replaces the earlier one. A thread that already received the first list is now holding children that are no longer in their parent's tuple. When it reads a property on one of them, _get_child_props locates the child via BaseFigure._index_is, which searches by object identity:

index_list = [i for i, curr_val in enumerate(iterable) if curr_val is val]
if not index_list:
    raise ValueError("Invalid value")

Its object isn't there, so ValueError("Invalid value").

This bites servers in particular: the process-global default template (pio.templates) is still cold when the first concurrent requests arrive, and it's plotly's own code (apply_default_cascade) that touches it during an ordinary px.bar(...). There's no seam where a caller could take a lock.

File "plotly/express/_core.py", line 994, in apply_default_cascade
    bar.marker.pattern.shape for bar in args["template"].data.bar
File "plotly/graph_objs/bar/marker/_pattern.py", line 252, in shape
    return self["shape"]
File "plotly/basedatatypes.py", line 4437, in _props
    return self.parent._get_child_props(self)
File "plotly/basedatatypes.py", line 4471, in _get_child_props
    child_ind = BaseFigure._index_is(children, child)
File "plotly/basedatatypes.py", line 3965, in _index_is
    raise ValueError("Invalid value")
ValueError: Invalid value

Reproduction

Every built-in template is loaded lazily, so each gives one independent cold start to race on. Reproduces in 9–10 out of 10 runs.

import sys
import threading

import pandas  # noqa: F401  -- import eagerly; narwhals' lazy import has its own race
import plotly
import plotly.express as px
import plotly.io as pio

# Not required to trigger the bug, but shortening the GIL switch interval makes
# it reproduce reliably instead of ~1% of the time.
sys.setswitchinterval(1e-6)

THREADS = 16
failures = {}

for template_name in pio.templates:
    pio.templates.default = template_name
    errors = []
    barrier = threading.Barrier(THREADS)

    def build_first_figure():
        barrier.wait()          # line all threads up on the still-cold template
        try:
            px.bar(x=[0], y=[0])
        except Exception as exc:
            errors.append(exc)

    workers = [threading.Thread(target=build_first_figure) for _ in range(THREADS)]
    for worker in workers:
        worker.start()
    for worker in workers:
        worker.join()

    if errors:
        failures[template_name] = errors

print(f"plotly {plotly.__version__} on python {sys.version.split()[0]}")
print(f"{len(failures)}/{len(pio.templates)} templates raced")
for name, errs in failures.items():
    print(f"  template {name!r}: {len(errs)}/{THREADS} threads -> "
          f"{type(errs[0]).__name__}: {errs[0]}")

plotly 5.24.1 on python 3.12.11
3/11 templates raced
  template 'ggplot2': 1/16 threads -> ValueError: Invalid value
  template 'seaborn': 7/16 threads -> ValueError: Invalid value
  template 'plotly_white': 2/16 threads -> ValueError: Invalid value

Affected versions I measured (16 threads, fresh process, rate over 10 runs): 5.24.1, 6.0.1, 6.3.1, 6.6.0, 6.9.0 — all 9–10/10.

Proposed fix

Build first, then publish with setdefault, so the entry is written exactly once and every thread converges on the same object. Threads may still duplicate the work, but no thread can be left holding an orphaned child — which is the invariant _index_is depends on. dict.setdefault is atomic in CPython, and dicts are internally locked under free-threading, so no explicit lock is needed.

--- a/plotly/basedatatypes.py
+++ b/plotly/basedatatypes.py
@@
             if isinstance(validator, CompoundValidator):
                 if self._compound_props.get(prop, None) is None:
                     # Init compound objects
-                    self._compound_props[prop] = validator.data_class(
-                        _parent=self, plotly_name=prop
-                    )
+                    child = validator.data_class(_parent=self, plotly_name=prop)
                     # Update plotly_name value in case the validator applies
                     # non-standard name (e.g. imagedefaults instead of image)
-                    self._compound_props[prop]._plotly_name = prop
+                    child._plotly_name = prop
+                    # Publish with setdefault, not assignment: two threads can
+                    # both find the entry empty and both build a child. A plain
+                    # assignment lets the late one replace an object the other
+                    # already handed out, orphaning it from this parent so that
+                    # _get_child_props' identity lookup fails. setdefault makes
+                    # the first published child win for everyone.
+                    self._compound_props.setdefault(prop, child)
 
                 return validator.present(self._compound_props[prop])
             elif isinstance(validator, (CompoundArrayValidator, BaseDataValidator)):
                 if self._compound_array_props.get(prop, None) is None:
                     # Init list of compound objects
                     if self._props is not None:
-                        self._compound_array_props[prop] = [
+                        children = [
                             validator.data_class(_parent=self)
                             for _ in self._props.get(prop, [])
                         ]
                     else:
-                        self._compound_array_props[prop] = []
+                        children = []
+                    # See the setdefault note above.
+                    self._compound_array_props.setdefault(prop, children)
 
                 return validator.present(self._compound_array_props[prop])

On 6.9.0, same version, same script, 15 runs each:

	reproduced
unpatched	14/15
patched	0/15

Figures still render identically, and template values are unchanged.

Happy to open a PR with this plus a regression test if the approach looks reasonable — or if maintainers would rather have an explicit lock or eager initialisation, that's an easy change.

Possibly separate: a second race in 6.x

While isolating the above I hit a different failure on concurrent first calls, which survives the fix above:

File "plotly/express/_core.py", line 1203, in to_named_series
    x = nw.from_native(x, series_only=True, pass_through=True)
File "narwhals/_native.py", line 408, in is_native_pandas
    (pd := get_pandas()) is not None and isinstance(obj, (pd.DataFrame, pd.Series))
AttributeError: partially initialized module 'pandas' has no attribute 'DataFrame'
    (most likely due to a circular import)

That's narwhals' lazy import pandas racing across threads, unrelated to templates. It's why the repro above imports pandas eagerly. Happy to file separately if it's not already known.

@hb1915

@hb1915 could you open a PR with the change? If it resolves the issue we would love to review it!

@robertclaus

@robertclaus should be there now!

Warm plotly templates once per process to avoid a startup race

Navigation Menu

Repository navigation

ValueError(“Invalid value”) in basedatatypes.py #3441

Cause

Reproduction

Proposed fix

Possibly separate: a second race in 6.x

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

ValueError(“Invalid value”) in basedatatypes.py #3441

Description

Activity

sergiykhan commented on Oct 29, 2021

mrolinek commented on Dec 13, 2021

sergiykhan commented on Dec 13, 2021

joseortunomacias commented on Feb 21, 2022

benjaminh commented on Mar 29, 2022

henriquegrabin commented on Mar 30, 2022

KnopKirrie commented on Oct 7, 2022

sergiykhan commented on Apr 7, 2023

plan-9-from-outer-space commented on Feb 12, 2024

Coding-with-Adam commented on Feb 14, 2024

sergiykhan commented on Feb 14, 2024

plan-9-from-outer-space commented on Feb 14, 2024

alexcjohnson commented on Feb 14, 2024

plan-9-from-outer-space commented on Feb 14, 2024

alexcjohnson commented on Feb 15, 2024

sergiykhan commented on Feb 15, 2024

hb1915 commented on Aug 3, 2026

Cause

Reproduction

Proposed fix

Possibly separate: a second race in 6.x

robertclaus commented on Aug 3, 2026

hb1915 commented on Aug 4, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions