Skip to content

Nexus error chain workaround #944

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Jul 7, 2025
Merged

Nexus error chain workaround #944

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
3ba4ce3
Always move first error message to top-level
dandavison Jul 7, 2025
b1d2ff8
Test
dandavison Jul 7, 2025
a0ff986
Inject ApplicationError
dandavison Jul 7, 2025
eb76cd1
Drop the type
dandavison Jul 7, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
54 changes: 27 additions & 27 deletions temporalio/worker/_nexus.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -337,38 +337,24 @@ async def _nexus_error_to_nexus_failure_proto(

See https://github.com/nexus-rpc/api/blob/main/SPEC.md#failure
"""
message = str(error)
if cause := error.__cause__:
try:
failure = temporalio.api.failure.v1.Failure()
await self._data_converter.encode_failure(cause, failure)
# nexusrpc.HandlerError and nexusrpc.OperationError have their
# own error messages and stack traces, independent of any cause
# exception they may have, and it would be reasonable to expect
# these to be propagated to the caller.
#
# In the case of OperationError (UnsuccessfulOperationError
# proto), the server takes the message from the top-level
# UnsuccessfulOperationError and replace the message of the
# first entry in the details chain with it. Presumably the
# server is anticipating that we've hoisted the message to that
# position and is undoing the hoist. Therefore in that case, we
# put the message from the first entry of the details chain at
# the top level and accept that the message of the
# OperationError itself will be lost.
#
# Note that other SDKs (e.g. Java) remove the message from the
# first item in the details chain, since constructors are
# controlled such that the nexus exception itself does not have
# its own message.
#
# Following other SDKs, we move the message from the first item
# in the details chain to the top level nexus.v1.Failure
# message. In Go and Java this particularly makes sense since
# their constructors are controlled such that the nexus
# exception itself does not have its own message. However, in
# Python, nexusrpc.HandlerError and nexusrpc.OperationError have
# their own error messages and stack traces, independent of any
# cause exception they may have, and this must be propagated to
# the caller. See _exception_to_handler_error for how we address
# this by injecting an additional error into the cause chain
# before the current function is called.
failure_dict = google.protobuf.json_format.MessageToDict(failure)
if isinstance(error, nexusrpc.OperationError):
message = failure_dict.pop("message", str(error))
else:
message = str(error)
return temporalio.api.nexus.v1.Failure(
message=message,
message=failure_dict.pop("message", str(error)),
metadata={"type": _TEMPORAL_FAILURE_PROTO_TYPE},
details=json.dumps(
failure_dict,
Expand Down Expand Up @@ -446,7 +432,21 @@ def _exception_to_handler_error(err: BaseException) -> nexusrpc.HandlerError:
# Based on sdk-typescript's convertKnownErrors:
# https://github.com/temporalio/sdk-typescript/blob/nexus/packages/worker/src/nexus.ts
if isinstance(err, nexusrpc.HandlerError):
return err
# Insert an ApplicationError at the head of the cause chain to hold the
# HandlerError's message and traceback. We do this because
# _nexus_error_to_nexus_failure_proto moves the message at the head of
# the cause chain to be the top-level nexus.Failure message. Therefore,
# if we did not do this, then the HandlerError's own message and
# traceback would be lost. (This hoisting behavior makes sense for Go
# and Java since they control construction of HandlerError such that it
# does not have its own message or stack trace.)
handler_err = err
err = ApplicationError(
message=str(handler_err),
non_retryable=not handler_err.retryable,
)
err.__traceback__ = handler_err.__traceback__
err.__cause__ = handler_err.__cause__
elif isinstance(err, ApplicationError):
handler_err = nexusrpc.HandlerError(
# TODO(nexus-preview): confirm what we want as message here
Expand Down
13 changes: 9 additions & 4 deletions tests/nexus/test_workflow_caller_error_chains.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -174,9 +174,14 @@ def action_in_nexus_operation():
(
ApplicationError,
{
# TODO(nexus-preview): empirically, this is "handler-error-message",
# but it should be "runtime-error-message"
# "message": "runtime-error-message",
"message": "handler-error-message",
"non_retryable": True,
},
),
(
ApplicationError,
{
"message": "runtime-error-message",
"type": "RuntimeError",
"non_retryable": False,
},
Expand Down Expand Up @@ -234,7 +239,7 @@ def action_in_nexus_operation():
(
nexusrpc.HandlerError,
{
"message": "handler-error-message",
"message": "handler-error-message-2",
"type": nexusrpc.HandlerErrorType.UNAVAILABLE,
"retryable": True,
},
Expand Down
Loading