Defer record_node_startup_error in prestartup error path; add docstrings

Buffer prestartup failures into a module-level list inside main.py
instead of importing 'nodes' (and therefore 'torch') from within the
exception handler. After the normal 'import nodes' line, drain the
buffer via nodes.record_node_startup_error so bootstrap order stays
deterministic regardless of whether a prestartup script succeeded.

Also convert the explanatory '#' comment on the new
/node_startup_errors endpoint into a proper docstring and add a
docstring to execute_prestartup_script, addressing CodeRabbit's
docstring-coverage warning on this PR.

Addresses review feedback on PR #13184.

Amp-Thread-ID: https://ampcode.com/threads/T-019e2f90-26fe-7048-9855-5ff39d08a3e0
Co-authored-by: Amp <amp@ampcode.com>
This commit is contained in:
Jedrzej Kosinski
2026-05-21 14:08:23 -07:00
parent ae539cfa0a
commit 7259e664ef
2 changed files with 46 additions and 19 deletions

View File

@@ -767,17 +767,20 @@ class PromptServer():
@routes.get("/node_startup_errors")
async def get_node_startup_errors(request):
# Group errors by source so the frontend/Manager can render them
# in distinct sections. `source` is the same string as the
# module_parent used at load time (e.g. "custom_nodes",
# "comfy_extras", "comfy_api_nodes") and is left as a free-form
# string so the contract survives node-source layouts evolving.
# The response only contains source buckets that actually had a
# failure; consumers should not assume any particular set of keys
# is always present.
#
# `module_path` is stripped because the absolute on-disk path is
# internal detail that the frontend has no use for.
"""Return startup errors recorded during node loading, grouped by source.
Group errors by source so the frontend/Manager can render them in
distinct sections. ``source`` is the same string as the
``module_parent`` used at load time (e.g. ``"custom_nodes"``,
``"comfy_extras"``, ``"comfy_api_nodes"``) and is left as a
free-form string so the contract survives node-source layouts
evolving. The response only contains source buckets that actually
had a failure; consumers should not assume any particular set of
keys is always present.
``module_path`` is stripped because the absolute on-disk path is
internal detail that the frontend has no use for.
"""
grouped: dict[str, dict[str, dict]] = {}
for entry in nodes.NODE_STARTUP_ERRORS.values():
source = entry.get("source", "custom_nodes")