With newer versions of orjson, users need to specify the json engine explicitly (bug?)

[xucian]

Hey

I found out I get an
AttributeError: partially initialized module 'orjson' has no attribute 'OPT_NON_STR_KEYS'
if I don't specify this
plotly.io.json.config.default_engine = 'orjson'
when using orjson v3.6.6 (latest as of 25jan2022)

Also, additional note for whoever might have this issue: you don't need to uninstall orjson if you don't want to use it. just set the engine to 'json' explicitly.

I'm using orjson because of the performance claims, although I ran some tests switching between the 2 engines and they seem to yield the same results: using go.Candlestick with 10000 candlesticks and some 4-5 indicators, getting ~0.8sec in each case for creating the plot. My purpose is to improve the dash server performace, but it seems it makes no difference (the web page still renders slower than the ticker even with 600 candles)

[HeddeCrisp]

+1

[marcelo-ventura]

+1. App is presenting problems after a while. Not sure what's going on.

[noamgot]

+1. Happens to me when I use streamlit's st.plotly_chart (plotly v5.18.0)

[C0deBeez]

File "/.venv/lib/python3.11/site-packages/plotly/io/_json.py", line 146, in to_json_plotly
opts = orjson.OPT_NON_STR_KEYS | orjson.OPT_SERIALIZE_NUMPY

This line of code may cause this issue

[ndrezn]

A fix may be using a regular import rather than using importlib to manage this package.

[ndrezn]

To reproduce using Dash:

1. pip install orjson
2. Make a Dash app
3. Trigger reload

If you look at network requests, it'll show this exact error.

[afeld]

Strangely, this error should have been fixed in Plotly v5.21.0, but I'm seeing it an environment with orjson v3.10.11. I'm not seeing it in other environments with the same version of Plotly with orjson v3.10.7 and v3.10.12. Neither the CHANGELOG nor diff suggest there was anything fixed in orjson around that setting from v3.10.11 to 3.10.12 🤨

[emilykl]

Hi @afeld, thanks for the heads-up. I just did a quick test and I wasn't able to reproduce -- could you share replication steps, including your OS, Python version, and pip freeze of packages in your Python environment? If you can share that info I will try again to reproduce the issue.