Plotly renderer limits for headless container deployment

Direct answer

No renderer setting gives you static SVG files without an export engine. plotly.io.renderers.default = 'svg' only changes how figures are displayed interactively (for example, what fig.show() emits in a notebook). It does not power fig.write_image() or fig.to_image(). Plotly's Python library has no pure-Python SVG exporter, so headless file export requires Kaleido (the current, lightweight option) or the deprecated Orca. The good news: Kaleido needs no display server or GUI libraries, so it fits your container constraints.

What each piece actually does

• renderers.default = 'svg' — controls interactive rendering output only. Safe to set globally in dev and production; it has no effect on file generation and nothing to do with Kaleido/Orca.
• plotly.offline.plot() — writes a standalone HTML file with embedded plotly.js. It bypasses the renderer system entirely, works headless with zero extra dependencies, and is unaffected by the default renderer.
• plotly.plot() (online mode) — uploads to Plotly Cloud. Also unaffected by renderer defaults. (Note: this legacy API has been deprecated in the 4.x/5.x line in favor of Chart Studio; treat that as a separate migration concern.)
• fig.write_image('out.svg') / fig.to_image(format='svg') — the only paths to a real SVG file, and both require Kaleido. Without it they raise ValueError about no supported image export engine.

Recommended setup

Install Kaleido in the container image — it ships as a self-contained binary wheel, no X11, no browser packages:

pip install "plotly>=5" kaleido

Then export explicitly where you need files:

import plotly.graph_objects as go

fig = go.Figure(go.Scatter(y=[1, 2, 3]))
fig.write_image("/tmp/chart.svg")   # or fig.to_image(format="svg") for bytes

On Plotly 5.x, Kaleido is picked up automatically once installed. On 4.x you may need engine="kaleido" explicitly in write_image — verify against your pinned version. Make sure the Kaleido wheel matches your container's OS/architecture (linux x86_64 vs arm64); a mismatched wheel is a common silent failure in multi-arch builds.

For Dash callbacks

Dash's interactive figures are rendered client-side in the user's browser, so headless export doesn't apply to normal dcc.Graph output. If a callback must return an image or file, call fig.to_image(format="svg") (or "png") server-side with Kaleido installed and return the bytes via dcc.send_bytes or a download component. If a callback returns HTML, use fig.to_html() — pure Python, no export engine needed.

Verification in the target container

python -c "import kaleido; print(kaleido.__version__)"
python -c "import plotly.graph_objects as go; go.Figure(go.Scatter(y=[1,2])).write_image('/tmp/t.svg')"
head -c 100 /tmp/t.svg   # should start with <?xml or <svg

Bottom line

Setting pio.renderers.default = 'svg' globally is harmless but solves a different problem. For consistent headless SVG output across scripts and Dash, install Kaleido and use write_image/to_image; use offline.plot/to_html when HTML output is acceptable and you want zero extra dependencies. Version-specific behavior (4.x engine selection, Kaleido's own version compatibility) should be confirmed against your pinned releases before rollout.