[Kea-users] Segmentation fault using REST API in DHCPv4 HA setting

[Caciano dos Santos Machado]

Sure,

I'll open an issue with the additional data you asked.

Thanks
Caciano Machado

On 9/22/22 14:15, Andrei Pavel wrote:
> Hi there,
>
> This is a very detailed report. I appreciate it.
>
> I've tried to replicate the crash, but was unsuccessful so far. Things 
> I've done to match your setup:
> * Used the python script to send the config-set.
> * Used the same config you posted below, with the exception of the 
> interface and address in "interfaces-config" which had to be changed 
> because the interface did not exist on my system and the configuration 
> was being rejected (possibly before the crash could have happened).
> * Sent the config-set command to the standby server, just as you did.
>
> On top of that, as exploratory measures, I've tried:
> * sending the config-set to the primary server instead
> * having DHCP traffic sent to the HA nodes while sending the command
> * configuring reclamation to delete leases as frequently as possible, 
> and as many leases as possible so that it takes a longer time. This is 
> after noticing that the handling of the config-set command and the 
> reclamation seem to be intermingling in the logs you shared. So I 
> figured that maybe replicating it requires the config-set to be 
> received during a reclamation.
> * sending config-set repeatedly at various rates
>
> The result was always that Kea received the config-set and the config 
> was accepted and Kea was reconfigured. Needless to say, there was no 
> crash.
>
> We would like to get to the bottom of it. Would you be willing to 
> create an issue on Kea's Gitlab[1] with the details you provided here 
> so that we can track it more easily? We could also use a bit more 
> information, if it can be easily provided:
> * a stack trace for the signaled thread or for all threads would be 
> most helpful. If using a debugger, this could be "thread apply all bt" 
> or "thread backtrace all" or something else.
I have a gdb output with "thread apply all bt" using the core dump.
> * the core dump if you still have it
Yes, I have.
> * the starting configuration - do I understand correctly that it is 
> the same configuration you sent via config-set, except it has some 
> additional shared networks?

OK, I'll just obsfucate MAC and IP information before sending it.

>
>
> Thank you!
> Andrei Pavel,
> Software Engineer,
> Internet Systems Consortium
>
> [1] https://gitlab.isc.org/isc-projects/kea/-/issues/new
>