Question

I have a long-running program on a remote machine and want to be sure that (1) I have a record of any exception that causes it to terminate and (2) someone is notified if it terminates. Does anyone see drawbacks to the method I am using? (or have recommendations for a better one?)

I’ve read the Python docs and many exception-related posts here, and understand that blanket except clauses are usually a bad idea. Within subroutines and modules I always
use except to handle specific expected exceptions, but it seems useful to have a “catch-all” except clause at the highest level of the program to ensure I can log the exception before the program exits.

What do you think?

import traceback
try:
    # main program code here
except BaseException:
    tb = traceback.format_exc()
    msg = "Exiting program due to exception:" + tb
    LogToFile(msg)         # custom logging function
    SendAlertEmail(msg)    # warn admin that program terminated
    raise                  # program exits with the existing exception

Note that I was using BaseException instead of Exception because if someone at the terminal presses Ctrl-C, I would like to log that as the reason for exiting the program (and alert an admin that the program was exited). But I suppose I could also use:

except Exception, KeyboardInterrupt:

Answer 1

There is no specific drawback, but there is an excellent alternative — sys.excepthook.

In your specific version, consider using a bare except:, and sys.exc_info() to get the exception information; that will ensure you do catch everything — even in the weird case where some module raises something else than an instance of a subclass of BaseException. E.g.:

>>> class X: pass
... 
>>> raise X
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
__main__.X: <__main__.X instance at 0xc9ad0>

As you see, it is still possible to raise something a except BaseException: would not catch — that’s why bare-except except: still exists (specifically for very special uses such as yours!).

Whether you use the hook, or build your own, consider (perhaps depending on configuration flags or environment settings) not burdening the end-user with all the details (just as a neat touch of improved user experience!), just a meaningful summary (reassuring the user that all details of the problem have been recorded, etc, etc).