Debugging

VCR.py has a few log messages you can turn on to help you figure out if HTTP requests are hitting a real server or not. You can turn them on like this:

import vcr
import requests
import logging

logging.basicConfig() # you need to initialize logging, otherwise you will not see anything from vcrpy
vcr_log = logging.getLogger("vcr")
vcr_log.setLevel(logging.INFO)

with vcr.use_cassette('headers.yml'):
    requests.get('http://httpbin.org/headers')

The first time you run this, you will see:

INFO:vcr.stubs:<Request (GET) http://httpbin.org/headers> not in cassette, sending to real server

The second time, you will see:

INFO:vcr.stubs:Playing response for <Request (GET) http://httpbin.org/headers> from cassette

If you set the loglevel to DEBUG, you will also get information about which matchers didn’t match. This can help you with debugging custom matchers.

CannotOverwriteExistingCassetteException

VCR.py raises CannotOverwriteExistingCassetteException when a request isn’t in the cassette and the record mode doesn’t allow recording it. The message starts with what went wrong, then shows the request, the cassette’s settings and, when it helps, the recorded requests closest to the one being made, with how each of them differs. It ends with suggestions for fixing it.

For example, when the query string of a request changed:

CannotOverwriteExistingCassetteException: Can't play back this request: no recorded interaction matches it (the closest differ in query).
record_mode 'once' doesn't record new requests into an existing cassette.

  GET https://www.googleapis.com/?alt=json&maxResults=200

Cassette: cassette.yaml
  record_mode: once
  match_on: method, scheme, host, port, path, query
  interactions: 1 recorded, 0 played

Closest recorded requests:
  1. GET https://www.googleapis.com/?alt=json&maxResults=500
     query differs: [('alt', 'json'), ('maxResults', '200')] != [('alt', 'json'), ('maxResults', '500')]

How to fix:
  * If the request changed on purpose, record it again: run once with
    record_mode='new_episodes' to add it to the cassette, or record_mode='all'
    with drop_unused_requests=True to re-record the cassette from scratch [1].
  * If the difference changes on every run (a timestamp, a random ID), make
    matching ignore it: loosen match_on, or normalize the request with
    before_record_request [2].

[1] https://vcrpy.readthedocs.io/en/latest/usage.html#record-modes
[2] https://vcrpy.readthedocs.io/en/latest/configuration.html#request-matching

When a request body differs, the message shows a short diff of the bodies. At most three of the closest recorded requests are shown in full; the rest are summarized by how they differ.

The exception also has these attributes, for tools that want to report the problem themselves:

• cassette and failed_request: the cassette and the request that couldn’t be handled.

• reason: why it couldn’t be handled. "cassette_not_found" when the cassette file doesn’t exist (or couldn’t be loaded) and the record mode can’t create it, "already_played" when the matching recorded interactions have all been played back already, "no_match" when recorded requests are similar but none matches, and "nothing_similar" when nothing in the cassette is similar.

• similar_requests: the closest recorded requests, as (request, succeeded_matchers, failed_matchers) tuples, where failed_matchers is a list of (matcher_name, message) tuples.