On Wed, Feb 7, 2018 at 1:47 PM, Jim Jagielski <[email protected]> wrote:
> Just for fun, what is the functional difference, if any, between this
> very large patch, that adds lots of code, and this extremely simple
> diff which, from what I can tell, handles the exact defined
> "problem" with the original code???

This oneliner is what I proposed to Mark in the PR to unblock his
situation (IPC-SysV leaks on main process crash)...
But unless one configures a different ServerName (or ServerAdmin??)
for each vhost, multiple servers may have the same ID.
AFAICT, this is neither a requirement nor an enforcement in httpd, and
there may even be multiple vhost configured on different ports with
the same ServerName (legitimately).
If this happens, mod_proxy_lb will possibly reuse slotmems with
incompatible sizes (two vhosts with the same ID but for example a
different number of balancer members), which is controlled by
mod_slotmem but leads to the previous SHM to be overwritten (i.e. all
bets are off).

We *really* need an unique ID for mod_proxy_lb to work correctly.

>
> Just curious if our current policy is to use a sledgehammer now
> to fix what can be handled with a pair of tweezers.

You just ignored anything related the ID (hence the issue) so far, at
least there is a proposal this time.
Please go ahead with a simpler fix if the sledgehammer hurts you.

> On Feb 6, 2018, at 12:59 PM, Yann Ylavic <[email protected]> wrote:
>> <mod_proxy_lb.diff>

This patch was bogus, I have fixed it but I'd better wait for yours now.


Regards,
Yann.

Reply via email to