Welcome! Log In Create A New Profile

Advanced

BUG/MEDIUM: incompatibility between DNS SRV records and server-state

Posted by Baptiste 
Hi all,

Jude (and FrancisL) reported an issue when using DNS SRV resolution coupled
with server-state: the state of the servers were not properly re-applied to
the new process.
This patch improve the server-state file to fix this issue: the srv record
used to manage this server is now saved by the previous process and changes
can be re-applied by the new one (unless the SRV record has changed, of
course)

Thx to Jude who provided time and resources to be able to reproduce a last
bug.

There are 2 patches, one for -dev and one for 1.8 (since server.c content
has changed a bit between those 2 versions).

Baptiste
On Tue, Sep 04, 2018 at 10:02:09AM +0200, Baptiste wrote:
> This patch improve the server-state file to fix this issue: the srv record
> used to manage this server is now saved by the previous process and changes
> can be re-applied by the new one (unless the SRV record has changed, of
> course)
(...)

applied, thanks Baptiste!

Willy
On Tue, Sep 4, 2018 at 5:46 PM, Willy Tarreau <[email protected]> wrote:

> On Tue, Sep 04, 2018 at 10:02:09AM +0200, Baptiste wrote:
> > This patch improve the server-state file to fix this issue: the srv
> record
> > used to manage this server is now saved by the previous process and
> changes
> > can be re-applied by the new one (unless the SRV record has changed, of
> > course)
> (...)
>
> applied, thanks Baptiste!
>
> Willy
>


I did not see the patch in 1.8 (I sent a specific one in my mail). I think
it's safer to wait a bit to ensure there is no regressions.
Note that 2 people at least are currently using it successfully in prod.

Baptiste
On Wed, Sep 05, 2018 at 11:21:08AM +0200, Baptiste wrote:
> I did not see the patch in 1.8 (I sent a specific one in my mail).

It's expected, we usually run into some backporting sessions taking a
whole day instead of doing them one at a time in random order. It helps
us maintain the ordering (which is important sometimes) and to be sure
we didn't miss any. Your commit message mentions the presence of the
1.8-specific patch so it will be picked at this moment.

> I think it's safer to wait a bit to ensure there is no regressions.

Noted, thanks.

> Note that 2 people at least are currently using it successfully in prod.

Great!

Willy
Sorry, only registered users may post in this forum.

Click here to login