bird-1.5.0-1.x86_64 crasches quite often on reconfiguration including renaming BGP protocols due to bad set of CFLAGS passed on compilation

[Tomasz Pala]

On Sat, Mar 26, 2016 at 13:46:32 +0100, Bartek Radwan wrote:

>> Yes, but not current. Nevertheless, so variable LA in 15-minutes window
>> is weird -
> 1min, 5min and 15min have been quite similar since upgrade:
> http://www.sq9mev.info/bird_cflags_loadavg_change_8h.png

OK, this looks like some 'classic' choking. If it's so long and
regular, you can catch this in action and attach debugger. Just install
bird-debuginfo.

> PLD before 1.5.0-2:
> 
> -fwrapv 

Try disabling this one (if they were affected by previous two, this is similar).

> -fno-debug-types-section -fvar-tracking-assignments
> -Wp,-D_FORTIFY_SOURCE=2 -fstack-protector-strong 
> --param=ssp-buffer-size=4

These shouldn't make such effect.

>> using binaries they provide and compare?
> Do they provide compiled binaries at all? Or you just meant binaries 

http://bird.network.cz/?download&tdir=redhat/ ?

> day. I do not observe such a behavior in test enviroment, but in 
> production enviroment there are several full view and IX sessions and 
> it's rather hard to reproduce such enviroment on my desktop.

Why? On desktop - use private ASn, on BGP router - use tables not
imported/exported anywhere else. I.e. simulate some customer. This is
not quagga, where everything is either two-way synchronized or not at
all (when put into separate view preventing RIB from going into FIB).

> Here's forked PLD bird repo with these changes:
> https://github.com/sq9mev/bird/tree/rename_segfault

Some day I need to learn using git efficiently... ;)

-- 
Tomasz Pala <gotar at pld-linux.org>