[kam-validmind]

Problem

On Databricks serverless, Kaleido 1.x can't start Chrome (no root, missing system libs), so run_documentation_tests() fails as soon as a test produces a Plotly figure. Retries can also leave the cell hanging.

Change

• If fig.to_image fails, log one warning and upload a placeholder PNG plus the Plotly JSON (json_image), so the platform can render the figure interactively.
• After the first failure, skip Kaleido for later figures (no repeated retries).
• VALIDMIND_PLOTLY_STATIC_EXPORT=false skips the attempt up front.
• Placeholder images aren't sent to the LLM when generating test descriptions.

Notes

• The backend /log_figure requires the image file, so a placeholder is uploaded rather than omitting the PNG. Anything using the PNG instead of the JSON (e.g. document export) shows the placeholder.
• The fallback latches on any to_image exception, not only "Chrome is missing". If Chrome works but one figure fails for a one-off reason (Kaleido's 90-second render timeout on a very large figure, a Chrome crash), the rest of the session also gets placeholders, where before only that one test failed. This is deliberate: it keeps the Databricks case to a single attempt, and the one-off case is rare and comes with a warning naming the error. Alternatives if it bites: latch only on Kaleido/Chrome start-up errors, or after two or three consecutive failures.
• Not tested on Databricks serverless; covered by unit tests with a mocked Kaleido failure.

Tests

tests/test_figure_static_export.py (4 tests).

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}