Re: parallel/sge 6.x

[Mark Davies <mark%mcs.vuw.ac.nz@localhost>]

On Wednesday 09 August 2006 14:11, George Georgalis wrote:
> Hey! Thanks! There is a lot of work in there :)
>
> I made some progress, but don't know much about the last error...

I've updated http://www.mcs.vuw.ac.nz/~mark/sge6.tgz to incorporate your 
points below and to update to the 6.0.8 tarball.

> First in the Makefile, I added -no-jni to to the lines that have
> it below, so JAVA_HOME doesn't need define (sigh).

Added.

> The MESSAGE file could include
>
> sge_qmaster     8701/tcp        # grid engine 6 qmaster
> sge_execd       8702/tcp        # grid engine 6 execd
>
> (thoes ports are from another install, may (or not) be a good port
> choice.)

I used 536 and 537 as they were the examples used in some Sun document I found 
(and they follow on from 535).  Not sure the sge_commd line is still needed 
but I've left it there for the moment.

> doesn't seem to depend on database/db4, I guess it should by default

? It does depend on it and the dependency is there.

> Then, the DB errors...
>         cd /usr/pkg/sge && ./install_qmaster -csp
> invokes
>
> ./utilbin/nbsd-x86_64/spoolinit berkeleydb libspoolb
> /usr/pkg/sge/grid/spool/spooldb init
>
> which looks at /usr/pkg/lib/libspoolb.so and /usr/lib/libspoolb.so
> but fails to check where it was actually installed, to quick fix:
> ln -s /usr/pkg/sge/lib/nbsd-x86_64/libspoolb.so /usr/pkg/lib/libspoolb.so

The way spoolinit is called from within inst_sge it does look in the correct 
place and load this.

> Similar problem, libdb4-4.4.so.4 is only looked for in one
> location, quick fix:
> ln -s /usr/pkg/lib/libdb4-4.4.so.4 /usr/lib/libdb4-4.4.so.4

This was broken and now fixed in the updated patch-am.

> Then I get:
        [...]
> Initializing spooling database
> critical error: Profiling array is not initialized!

> Looks like a sge software problem. Maybe ./install_qmaster is not
> properly setting env? The other stuff is cleanup, but this one
> looks like some effort is needed, any idea what's wrong?

Thats the initializing databases problem I mentioned.  The person I asked to 
look at it never got anywhere, and I haven't really poked at it myself.  The 
relevant code is in gridengine/source/libs/uti/sge_profiling.c.

cheers
mark