Re: Re: Re: Re: Re: Re: Upgrading Perl in production environment

[chip]

Anything financial is almost by definition high value and should never involve the use of unsafe signals. You can call it "maintenance and information gathering" but that doesn't change anything.

There are libraries for asynchronous DNS lookup. There are fork and pipes. There are lots of things you can do. If any of them involve unsafe signals ... well, let's just say I'd like to know where you wrote such a system so I can make sure not to invest my money there.

[jfroebe]

I'm using "trading systems" as an example

unfortunately, even the gethostbyaddr_r series of c functions can hang as well. depending on the load of the box and a dozen other reasons, the response from the name server (or whatever else) may not actually be received.

If a signal comes in and perl core dumps with half finished work... that's fine, there are clean up routines that identify where the breakage occurred and repair whatever was broken.

Using unsafe signals IS okay, IF the proper precautions are taken for recovery and continuing where it broke.

What I'm saying is that unsafe signals do have their uses and should never be removed from perl entirely.. Perhaps perl should be fixed to handle them better?

[chip]

You just don't get it.

It is never safe to use unsafe signals. You don't know what will happen. You can't be sure that when things go wrong you'll get a SEGV and program death, allowing for cleanup. You don't know what will happen. Nobody knows. Nobody can predict what an unsafe signal will cause. Nobody! That's what makes them "unsafe"!

(What frame of mind is required for someone to think that it's safe to use a feature called "unsafe signals"? I can't fathom it, I just can't.)

[jfroebe]

Actually I do understand your point of view... and agree with you for the most part.

even if the process goes into lala land, a separate process will kill it, start a new process which will examine what was done, undo whatever was 'messed up', and continue with the task. The overall effect would be no worse than performing kill -SIGKILL (-9).

However, if I just wrote a routine to restart the system call without any other failsafe in place, then that would be bad and irresponsible.

For most tasks, the unsafe signals are undesirable.. on certain tasks, they are unavoidable and necessary. The fact that they are unsafe is a failure of the Perl engine... so we have to make do, for now, with what we have and build contingency plans for using the unsafe signals.