Re: [Sqlrelay-discussion] sqlrsh hanging on openbsd

[Jolan L. <jo...@pr...>]

On Thu, Jun 09, 2005 at 02:14:49PM -0400, Firstworks/4access wrote:
> On Fri, 2005-06-03 at 00:05 -0500, Jolan Luff wrote:
> > On Wed, Jun 01, 2005 at 11:53:25PM -0400, David Muse wrote:
> > > I just made a maintenance release, give it a try.  It should detect
> > > readline properly now and work even if you disable readline support.
> > > 
> > > I had some strange problems when I tested it on openbsd though.  All of
> > > my queries failed with authentication errors and 2 sqlr-listener
> > > processes were getting spawned off for each client connection.  It's
> > > possible that it won't happen on real hardware though.  I was running
> > > OpenBSD-3.7 in vmware and vmware doesn't officially support openbsd.
> > > 
> > > Give it a try and let me know.
> > 
> > readline support is working now, thank you.
> > 
> > i noticed during the build i get a stream of 'rm: : Invalid argument'
> > after every g++ invocation which did not happen before.
> > 
> > as far as running it, i am seeing the same problem you are with real
> > hardware:
> > 
> > $ sudo sqlrsh -id sqlitetest
> > SQLRShell - Version 0.22
> >         Connected to: localhost:9000 as sqlitetest
> > 
> >         type help; for a help.
> > 
> > 0> debug on;
> > 0> SELECT sqlite_version();
> > Result Set Buffer Size: 100
> > Connecting to listener...
> > Unix socket: /tmp/sqlitetest.socket
> > Authenticating with listener : sqlitetest:sqlitetest
> > No authentication error.
> > Must Not Reconnect.
> > Authenticating with connection : sqlitetest:sqlitetest
> > Broken pipe
> > Exit 141
> > 
> > and if I run sqlrsh -id sqlitetest without sudo sqlrsh crashes right
> > away.  maybe this is due to failing to drop privileges?
> 
> I'll look into this a little more.  So far, it appears to only be a
> problem under OpenBSD.  It looks like there's an error passing file
> descriptors between processes.  The most recent release of OpenBSD did a
> really strange thing with a macro related to passing file descriptors.
> The macro contained function calls and thus couldn't be used at compile
> time to determine the size of a struct.  I had to fudge the code to make
> it work.  Maybe it doesn't really work :)
> 
> Try setting handoff="reconnect" in the sqlrelay.conf file and see if
> that doesn't make it work.  It's less efficient, but it might work.

using handoff="reconnect" let's me execute queries, thanks.