Bug 762253 (GLUSTER-521)

Setup is CIFS reexport from Gluster Storage Platform 

fully loaded configuration using "distribute + replicate" 

Samba server version 3.4.2 

GlusterFS version 3.0.0

cifs module from client under use with kernel version 2.6.18-164.6.1 

Please find the attachment with the error during runs. 

No Logs found on backend glusterfs server/client and no logs reported 
in samba logs too. 

SPECFS command used for validation is  

"cd /opt/benchmarks/SFS2008/spec-sfs2008/manager"
"java SfsManager -r sfs_config_cifs_replicate -s avail_media_test -v 2" 

sfs_config_cifs_replicate contains configuration for replicate validation run. 

NOTE: Validation test passes for pure distribute without replicate fully loaded
configuration. Working distribute share use the below command. 

"cd /opt/benchmarks/SFS2008/spec-sfs2008/manager"
"java SfsManager -r sfs_config_cifs_distribute -s avail_media_test -v 2"

What is the CIFS server here? Samba over booster or samba over FUSE?

We've never tested the CIFS part so first you should run SFS with just storage/posix as the volume.

CIFS over booster never go off the initial testing so its not working at all. CIFS
is native glusterfs reexported. 

Apart from that we have a customer requirement/deployment at stake so we have to 
do it with fully loaded configuration and its part of the acceptance test. So if 
this is never tested i am going to tick off this as not working for now.

NOTE: We have CIFS reexport running at Partners since approx 2yrs. so i don't 
think this is something new for us.

(In reply to comment #3)
> CIFS over booster never go off the initial testing so its not working at all.
> CIFS
> is native glusterfs reexported. 
> 

OK.

> Apart from that we have a customer requirement/deployment at stake so we have
> to 
> do it with fully loaded configuration and its part of the acceptance test. So
> if 
> this is never tested i am going to tick off this as not working for now.
> 

What the deployment needs is different from how we need to test this. We need to test by starting small. I can tell by experience of compliance testing on NFS that if certain things fail with posix-only, they'll certainly fail with other configurations. Plus, if the tests fail in posix, you know where to look while in a fully loaded config it'll be hard to tell where to find the problem.

Say the test succeeds on posix and then you start adding protocol/client and protocol/server, and the test fails at this point, you'll know these two are to blame and you can diagnose accordingly. This is again from experience of posix compliance testing on NFS xlator where even if posix passes the tests, it does not guarantee that it'll pass on replicate and distribute.

> NOTE: We have CIFS reexport running at Partners since approx 2yrs. so i don't 
> think this is something new for us.
Thats fine too. Even if the server is working it does not mean it is working exactly in compliance. Most of the time users wont notice or care about compliance since their tools may not be sensitive to compliance related problems. This is not so for SFS-like tools which test specifically for compliance.

(In reply to comment #4)
> 
> > NOTE: We have CIFS reexport running at Partners since approx 2yrs. so i don't 
> > think this is something new for us.
> Thats fine too. Even if the server is working it does not mean it is working
> exactly in compliance. Most of the time users wont notice or care about
> compliance since their tools may not be sensitive to compliance related
> problems. This is not so for SFS-like tools which test specifically for
> compliance.

Shehjar, below quoted were the paragraph from bug report.

| NOTE: Validation test passes for pure distribute without replicate fully 
| loaded
| configuration. Working distribute share use the below command. 
| 
| "cd /opt/benchmarks/SFS2008/spec-sfs2008/manager"
| "java SfsManager -r sfs_config_cifs_distribute -s avail_media_test -v 2"

Harsha did start small, and in his tests, he achieved successful tests upto distribute only tests. Hence the bug report on replicate + distribute scenario.

-Amar

My big bad! bad! bad!

I don't agree. Test runs are running fine with "Distribute" which was stated in this bug already (even with fully loaded config). So issues with the other translators are ruled out already. Its just the replicate+distribute which is making the test runs fail. 

Idea of Nit picking each translators is perfect with coding perspective and a developer point of view which i am not really concerned at. Requirement is just a working setup with testcases completed successfully.

PATCH: http://patches.gluster.com/patch/2670 in master (cluster/afr: Send the struct flock returned by the server to the user.)

PATCH: http://patches.gluster.com/patch/2689 in release-2.0 (cluster/afr: Send flock returned by locks xlator to user.)

PATCH: http://patches.gluster.com/patch/3171 in release-2.0 (cluster/afr: Send the struct flock returned by the server to the user only in the UNWIND path.)

*** Bug 863 has been marked as a duplicate of this bug. ***