socket_helper.transient_internet helper fails when FTP server returns 500
#122999
Comments
|
I tried to fix this with the following change: diff --git a/Lib/test/support/socket_helper.py b/Lib/test/support/socket_helper.py
index 87941ee1791..dcbd7e3ebd4 100644
--- a/Lib/test/support/socket_helper.py
+++ b/Lib/test/support/socket_helper.py
@@ -234,9 +234,12 @@ def filter_error(err):
(isinstance(err, urllib.error.HTTPError) and
500 <= err.code <= 599) or
(isinstance(err, urllib.error.URLError) and
- (("ConnectionRefusedError" in err.reason) or
- ("TimeoutError" in err.reason) or
- ("EOFError" in err.reason))) or
+ ((isinstance(err.reason, BaseException) and
+ isinstance(err.reason.__cause__,
+ (ConnectionRefusedError, TimeoutError, EOFError))) or
+ ("ConnectionRefusedError" in str(err.reason)) or
+ ("TimeoutError" in str(err.reason)) or
+ ("EOFError" in str(err.reason)))) or
n in captured_errnos):
if not support.verbose:
sys.stderr.write(denied.args[0] + "\n")But then the underlying issue was solved, and I do not know how to reproduce the failure. This may only fix some symptoms, and I am not sure that it is the best fix. There may be a regression in the stdlib code after making an argument of URLError an exception instead of a string. Or the helper take this into account and unroll these exceptions like it already do with other exception chains. Without reproducer I cannot do experiments. |
|
#32074 may be one of culprits. Before, the error message was formatted as |
|
@serhiy-storchaka I was thinking taking of a similar approach, but hit the same issue - how do you validate the code is actually working? However - @zware raised an interesting point on Discord: the fact this failed could be considered a good thing, because it drew our attention to the fact the So - maybe not fixing this is the better approach? |
|
Looks like @jeremyhylton reported this separately a few days ago (#122909), and has a prototype patch that goes a bit deeper into url lib (#122913). Closing as a duplicate. |
Bug report
Bug description:
We've seen a bunch of CI failures over the last few hours in the urllib2 FTP tests. These errors have been observed on iOS (1 2), macOS (1) and Linux (1). [edit: and Windows too (1) – @encukou]
The affected tests are:
It's not yet clear how to reliably reproduce the error - it appears to be a transient issue caused by the upstream FTP server raising a 500 error (possibly as a result of #122795).
From a code perspective, the issue is that the error being raised internally is an instance of
error_perm; this is used as theerr.reasonfor the URLError raised by urllib2, which breaks the exception filter.The error manifests as:
CPython versions tested on:
CPython main branch
Operating systems tested on:
macOS
The text was updated successfully, but these errors were encountered: