langgenius/dify · error · CompletionRequestError

completion_request_error

completion_request_error

Error message

Completion request failed.

What it means

HTTP 400 with error_code completion_request_error, raised when generate_more_like_this raises an InvokeError from the model runtime. The provider was reached and credentials are valid, but the actual completion invocation failed.

Solutions

1. Retry once with backoff for transient provider errors.
2. Reduce input length (trim the source message/context) to fit the model window.
3. Inspect ex.description (logged server-side) to distinguish content-policy vs. token-limit vs. provider-5xx.
4. Switch to a fallback model/provider if the primary keeps failing.
5. Verify the provider's own dashboard shows the request reaching it.

Example fix

# before: single shot, no retry
resp = get(more_like_this_url(id, mid))

# after: bounded retry with backoff, then fallback model
for attempt in range(3):
    try:
        resp = get(more_like_this_url(id, mid)); break
    except CompletionRequestError:
        sleep(2 ** attempt)
else:
    switch_model(); resp = get(more_like_this_url(id, mid))

// No purely client-side validation prevents provider invoke failures;
// preflight only checks credentials/quota/model (errors 755-757)
if (!providerConfigured(app, providers)) return; // avoids 755, not 758

for (let attempt = 0; attempt < 3; attempt++) {
  try {
    return await get(moreLikeThisUrl(id, mid));
  } catch (e) {
    if (e.code === 'completion_request_error' && attempt < 2) {
      await sleep(2 ** attempt * 500); // backoff for transient provider errors
      continue;
    }
    if (e.code === 'completion_request_error') {
      await switchToFallbackModel(id);
      return get(moreLikeThisUrl(id, mid));
    }
    throw e;
  }
}

When it happens

Trigger: GET more-like-this where the underlying provider invoke() raised InvokeError — network error mid-call, provider 4xx/5xx, malformed request, content-policy rejection, or context-length exceeded. The controller maps InvokeError.description to CompletionRequestError(400).

Common situations: Provider outage or transient error; prompt+context exceeding the model's token limit; content filter rejection; rate limit (provider-side, distinct from hosted quota); model deprecation.

Related errors

• model_currently_not_support
• provider_not_initialize
• provider_quota_exceeded
• app_more_like_this_disabled
• Internal Server Error

AI-assisted analysis of langgenius/dify@ef8544b173 (2026-08-12). Data as JSON: /api/errors/222c16c382b1c662. Report an issue: GitHub.

Appendix: source

Thrown at api/controllers/console/explore/message.py:191

                user=current_user,
                message_id=message_id_str,
                invoke_from=InvokeFrom.EXPLORE,
                streaming=streaming,
            )
            # response-contract:ignore compact_generate_response
            return helper.compact_generate_response(response)
        except MessageNotExistsError:
            raise NotFound("Message Not Exists.")
        except MoreLikeThisDisabledError:
            raise AppMoreLikeThisDisabledError()
        except ProviderTokenNotInitError as ex:
            raise ProviderNotInitializeError(ex.description)
        except QuotaExceededError:
            raise ProviderQuotaExceededError()
        except ModelCurrentlyNotSupportError:
            raise ProviderModelCurrentlyNotSupportError()
        except InvokeError as e:
            raise CompletionRequestError(e.description)
        except ValueError as e:
            raise e
        except Exception:
            logger.exception("internal server error.")
            raise InternalServerError()


@console_ns.route(
    "/installed-apps/<uuid:installed_app_id>/messages/<uuid:message_id>/suggested-questions",
    endpoint="installed_app_suggested_question",
)
class MessageSuggestedQuestionApi(InstalledAppResource):
    @console_ns.response(200, "Success", console_ns.models[SuggestedQuestionsResponse.__name__])
    @with_current_user
    def get(self, current_user: Account, installed_app: InstalledApp, message_id: UUID):
        app_model = installed_app.app_with_session(session=db.session())
        if app_model is None:
            raise AppUnavailableError()

View on GitHub (pinned to ef8544b173)