Rev 692: dmapi got into a state on a test cluster where dm_get_dmattr returned in http://samba.org/~tridge/3_0-ctdb

[Alexander Bokovoy]

tridge at samba.org пишет:
> ------------------------------------------------------------
> revno: 692
> revision-id:tridge at samba.org-20080121122450-er192pvz90my47kp
> parent: tridge at samba.org-20080118202503-05j7288nbv9cbeup
> committer: Andrew Tridgell <tridge at samba.org>
> branch nick: s3-ctdb-tridge
> timestamp: Mon 2008-01-21 23:24:50 +1100
> message:
> dmapi got into a state on a test cluster where dm_get_dmattr returned
>  -1/EINVAL but it was impossible to delete the stale session, and the
>  session still showed up in dm_getall_sessions(). This patch copes
> with that by changing to a new session name when the error occurs.
Is this a GPFS bug?
As we have about 4000 DMAPI handles limit in GPFS what do you plan to do
with limiting DMAPI exhaustion from Samba side? Currently session_num is
growing unbound and if all sessions are used up and inaccessible like in
case above the loop below will be indefinite:

> +again:
>  	ret = dm_get_dmattr(*sessionp, dmhandle, dmhandle_len, 
> -			    DM_NO_TOKEN, &dmname, 0, NULL, &rlen);
> +			    DM_NO_TOKEN, &dmname, sizeof(buf), buf, &rlen);
> +
> +	if (ret == -1 && errno == EINVAL) {
> +		DEBUG(0,("Stale DMAPI session - recreating\n"));
> +		if (dmapi_new_session()) {
> +			sessionp = dmapi_get_current_session();
> +			goto again;
> +		}
> +	}
Wouldn't we need to do something with it in this case?

-- 
/ Alexander Bokovoy
Samba Team                      http://www.samba.org/
ALT Linux Team                  http://www.altlinux.org/
Midgard Project Ry              http://www.midgard-project.org/