...in fork mode stderr is not available (Bad file descriptor error) and thus wallace traceback when it tries to send emails
Details
- Reviewers
vanmeeuwen machniak petersen - Maniphest Tasks
- T3751: Emails stuck in wallace queue until wallace is restarted
T2163: wallace fails to connect to ldap then running in debug 9 mode
T2498: Wallace logs errors when run in debug 9 mode - Commits
- rP10f1c173941d: With this patch I'm trying to introduce a file-type object in logger, which…
none
Diff Detail
- Repository
- rP pykolab
- Lint
Lint Not Applicable - Unit
Tests Not Applicable
Event Timeline
This is using the new file-like object from pykolab.logger. With this patch smtplib debug output gets redirected to logger rather than to stderr, which is not available in fork mode
This amendment tries to avoid cyruslib printing to non-available file descriptor, which causes traceback of wallace.
pykolab/imap/cyrus.py | ||
---|---|---|
147–153 | Looks like indentation here is broken. |
In addition LDAP trace_file needs to be specified. It is a file-like object define in logger.py Now we get ldap debug logs in pykolab.log file
I think we need process id (PID) logged for any level log record, that makes it easy to associate and track wallace and other pykolab modules work flow.
It turns out smtplib debug logging redirected to logger produces excessive new lines. This is attempt to mitigate the problem, make logs more pretty.
Make invitationpolicy module use smtplib sendmail wrapper _sendmail from modules, which now supports debug level 9 logging.
Make all pykolab log messages appear on level <8. This will isolate pykolab debug messages from 3rd partty libraries debug messages. -d debug -l 8 will print pykolab debug messages only, while -l debug -d 9 will print 3rd party libraries messages in addition to level 8 pykolab messages.
Looking good, except for the corner-case of an errmsg over an un-"captured" exception.
pykolab/imap/cyrus.py | ||
---|---|---|
152 | I think you may need to; except Exception, errmsg: before you can refer to errmsg in the call to log.exception(). |
In send_update_notification() we send mail in a loop trying up to 5 times if an error occurrs. But this is the only place where we do this. I think it would make sense to do this in all cases, i.e. in _sendmail(). I see also we don't call smtp.quit() consistently. Once we do this on success only, once we do this on error too.
The calls to modules._sendmail() are themselves looped, by the caller. Does that address the concern?
Eliminating inconsistensies of sending mail from wallace. Now all the logic is in modules._sendmail() wrapper. Invitationpolicy module is using that wrapper and makes it responsible for re-submistion logic.
wallace/module_invitationpolicy.py | ||
---|---|---|
1350 | I'd put this log.debug() after modules._sendmail(). |
yet another attempt to create a better wrapper for smtplib.sendmail(). Now instantiation of smtplib object and connect are separated.