Linux

This is comment history imported from bugzilla:

Comment 1 Hartmut Holzgraefe 2014-08-15 01:08:51 UTC
Problem is that there are only signal() handers for SIGINT, SIGHUP, SIGTERM

There are no handlers for SIGABRT (raised by failed assert()) or SIGSEGV/SIGBUS, SIGILL, SIGFPE, so assertions and hard crashes due to segmentation faults, bus errors, illegal instructions floating point errors terminate the maxscale process without putting anything into the logs ...

Comment 2 Hartmut Holzgraefe 2014-08-15 12:08:36 UTC
Ignore the previous comment on signal handlers, they are an issue as they prevent reason for termniantion being loged, but that is a different story ...

What's needed here is a redefinition of assert() that uses the log_manager instead of simply printing to stderr indeed ...

Also assert()s brute force termination of the whole process by calling abort() may not necessarily be the right way here, calling shutdown_server() may be better in most assertion failure situations?

(the ndbcluster storage engine e.g. has a ndb_require() call that is used throughout the code instead of assert() ...)

Comment 3 Mark Riddoch 2014-09-12 09:34:09 UTC
Please check this is only an issue with debug builds

Added assert information to log manager. Debug asserts now also flush all logs to disk before doing the assertion.

verified by manual checking. It is hard to implement automatic test for it and it is not really needed for minor logging issue.