On Mon, 28 Feb 2005, Gergely Debreczeni wrote:
> Hi *,
>
> Here at BUDAPEST we are preparing for a shut down.
> We would like the next:
> - accepting no more jobs
> - waiting till already running jobs will be finished and
> their owner retrived the output
>
> Is it enough to shut down globus-gatekeeper and wait, or is there
> an other/better way of handling this problem ?
No, you cannot shut down the gatekeeper, because it would prevent
jobs from getting cleaned up when they finish, i.e. the RB will
consider them as having failed.
In principle it would be sufficient to change this attribute in
the LDIF record for each queue:
GlueCEStateStatus: Production
to
GlueCEStateStatus: Draining
or
GlueCEStateStatus: Closed
or anything else, because in /opt/edg/etc/edg_wl_ui_cmd_var.conf
there is this line:
requirements = other.GlueCEStateStatus == "Production";
You can also remove all of the GlueCEAccessControlBaseRule lines
in your queue definitions (maybe put in a dummy VO placeholder).
You cannot simply switch off the site GIIS, because then one would
also lose the SE, which may be needed by running jobs.
|