Strange problems with GCC compile farm nodes gcc210 and gcc211

[Jacob Bachmeyer]

Dagobert Michelsen wrote:
> Hi Jacob,
>
> Am 19.05.2021 um 05:33 schrieb Jacob Bachmeyer via buildfarm <buildfarm at lists.opencsw.org>:
>   
>> 1.  A "git push" to a repository on gcc210 (gcc-solaris10) fails with "sh: git-receive-pack: not found" and "ssh gcc210 env" gives a clue: PATH=/usr/bin:/bin
>>
>> Shell sessions there have "PATH=/usr/sbin:/usr/bin:/opt/csw/bin" and (of course, according to Murphy) `which git-receive-pack` returns "/opt/csw/bin/git-receive-pack".  I would like to request that /opt/csw/bin be added to the PATH set by sshd if that is possible.
>>     
>
> I have now added
>   PATH=/usr/bin:/opt/csw/bin
> to /etc/default/login and it works for me now on gcc210 and gcc211:
>   dmichelsen at Andromeda-45 ~ % ssh dam at 213.178.77.185 'echo $PATH‘
>   /usr/bin:/opt/csw/bin
>   dmichelsen at Andromeda-45 ~ % ssh dam at 213.178.77.186 'echo $PATH‘
>   /usr/bin:/opt/csw/bin
>   

Thank you; git push to gcc210 works now.

>> 2.  In a first effort at a workaround, I decided to try pushing to the same repository on gcc211, only to find that sshd on that zone does not accept the same public key that works at gcc210.
>>
>> [...]
>>
>> I have no idea why gcc210 accepts my key while gcc211 does not.
>>     
>
> I checked your SSH key configuration against mine and compared key formats.
> From what I can say I use the same key format (RSA) and length (4096) and it
> works for me:
>
> gcc-solaris11# pwd
> /export/home/jcb/.ssh
> gcc-solaris11# ssh-keygen -lf authorized_keys               
> 4096 e1:ab:01:77:ea:b9:53:32:e1:b8:c0:16:08:fc:7a:04 authorized_keys
>
> As /export/home is a loopback-mounted filesystem for both gcc210 and gcc211
> the key is indeed absolutely identical on both machines. If the problem persists
> we need to make a quick debug session where I run sshd in full debug mode and
> probably trusslog the I/O to see whats going wrong.
>   

I have no idea what changed, but it works now.  It was consistently 
broken before -- those failures were repeatable but now I get a 
successful login.  Whatever you did to fix it, thanks.


-- Jacob