Re: Remote Build-Job nodes fail AND Build-Target stops in/after stage 1

Am Samstag 11 Februar 2006 23:28 schrieb Terry Mackintosh:
> Hi all

Hi Terry,

>
> The earlier mentioned patch worked great for the 'junk at end of line'
> problem.
>
> I'm doing a cluster build for PIII.
> Current trunk as of yesterday? day before? version 7067 which I noticed
> had a lot of new changes.
>
> The Build-Job that runs on the same box as Build-Target works fine.
>
> But the Build-Job(s) that run on the other nodes all fail.
> Mostly they fail during configure with a complaint that the compiler can
> not create executables.

It is often helpful to investigate the config.log that each run of configure 
creates (in the same directory that configure is run from), or configure 
itself, to find out why the test fails.

...

> I deleted the .err logs and the jobs re-queued and ran fine on the
> parent box.
>

I've started a ROCK cluster build, with two nodes. Master and nodes are set up 
as described in http://www.rocklinux.org/wiki/Cluster_Build:

	- create a new config, set  "Make a parallel (cluster) build". I used 
defaults for everything else (generic x86 target).

	- the master directory is /usr/src/ROCK-TRUNK (svn rev. 7067 with security 
patches from SubMaster)
	- the nodes are in /scratch/rock-node and scratch/rock-node2

	- set up the nodes:

/usr/src/ROCK-TRUNK#  ./scripts/Create-Links -config -build /scratch/rock-node
/usr/src/ROCK-TRUNK#  ./scripts/Create-Links -config 
-build /scratch/rock-node2

	- start the build:

/usr/src/ROCK-TRUNK#  ./scripts/Build-Target
/scratch/rock-node#  ./scripts/Build-Job -daemon
/scratch/rock-node2#  ./scripts/Build-Job -daemon

I started both Build-Jobs right after Build-Target, both nodes are busy when 
possible and the build has reached stage 5 so far.

> All boxes are PIII's.
> All boxes are newly installed with ROCK 2.0.3 PII generic and all have a
> newly build gcc 3.4.5 installed in /usr/local/
> All have /usr/local/(s)bin/ at the head of $PATH
> All have a link /usr/local/bin/cc that points to the new gcc.

Maybe this is causing problems somehow? Please take a look at the config.logs
and the configure output in the build logs (see above).

> And I did remember to run ldconfig on all machines after installing the
> new gcc.
> I have no_root_squash on the exported directory, so it should not be a
> permissions issue.
>
> I did not try to start a remote node until stage 1 after there was more
> then one job to do.
>

In my build that's after dietlibc finished:

02:14 firefox:      building job '1-dietlibc' (14562) ..
02:18 firefox:      finished job '1-dietlibc' (ok)
02:18 2006-02-12:   creating new job '1-binutils'
02:18 2006-02-12:   creating new job '1-gzip'

> At the moment only the parent box is running and the build is still in
> stage 1 with no errors so far and the build has passed where it was when
> the other nodes were in use.
>
> Ahhh... I just looked and Build-Target has just ended in or just after
> stage one claiming that all packages are built! I don't think so.
> Reran it, it just said that all packages are built:(
> There are 49 log files for stage 1, I'm not sure how many there should be.

I've got 6 packages built in stage 0 and 33 in stage 1...

With best regards,
	Stefan Fiedler

>
> Thanks