Errors from the manhole go to the logfile rather than the manhole

[richvdh]

This is only happening on matrix.org, and I'm sure it didn't previously:

in the manhole:

>>> slkghdklg
>>> 

meanwhile, in the log:

2019-02-28 17:14:03,867 - twisted - 240 - ERROR -  - Traceback (most recent call last):
2019-02-28 17:14:03,884 - twisted - 240 - ERROR -  -   File /usr/local/lib/python3.6/code.py, line 91, in runcode
2019-02-28 17:14:03,896 - twisted - 240 - ERROR -  -     exec(code, self.locals)
2019-02-28 17:14:03,908 - twisted - 240 - ERROR -  -   File <console>, line 1, in <module>
2019-02-28 17:14:03,917 - twisted - 240 - ERROR -  - NameError: name 'slkghdklg' is not defined

[richvdh]

I can't figure out what is different between matrix.org and sw1v.org that could cause this.

[erikjohnston]

Is this because we're capturing stderr?

[richvdh]

hrm, could be. I think I've disabled that on sw1v. Yet another reason we need to fix #1539 if so.

[richvdh]

hrm it seems to work ok on sw1v.org even if I allow it to redirect stdio.

[richvdh]

neil this is pretty painful for ops things. I'm going to bump it up.

[richvdh]

turns out it was being swallowed by the sentry exception hook. hopefully fixed by #5035.