Re: [HPDD-discuss] HSM questions
On Mar 6, 2014, at 8:50 AM, Frank Zago <fzago(a)cray.com> wrote:
Hello,
We, at Cray, are writing another copytool, and we've got some questions.
1/ While no copytool is running, if I run "lfs hsm_archive" on a file,
and then delete that file, the action will still be on the MDS (lctl
get_param -n mdt.lustre-MDT0000.hsm.actions).
Also, if I start a copytool, it won't get an event for it. Thus the
action stays in the system until I manually "purge" it. Is that normal
behavior?
This is something we can improve - to purge pending HSM requests when a file is being
removed.
2/ What is the way, or is there a way, to archive/restore a single file
from different nodes? For instance, I have a 1TB file, and to hasten the
transfer, I want to spread the load between 3 nodes to do the transfer
between Lustre and the backend, with each node copying 1/3 of the data.
llapi_hsm_action_get_fd() doesn't seem to have been conceived to work
that way. How would the progress reporting work in that case?
In that case, you will need Master + Slaves configuration. When the/a master receives a
request it can distribute the workload to slaves, and after all slaves get their job done,
master can notify the corresponding MDT that the request has been finished. This model can
be used for restore case as well, BTW.
3/ What is the case where hai_extent.length is different than -1 for
archiving? The posix copytool accounts for it, but I haven't seen it in
my testings.
Partial archive and restore are out of the scope in HSM v1. We will address it in HSM v2 -
if there will be a HSM V2.
Jinshan
Regards,
Frank.
_______________________________________________
HPDD-discuss mailing list
HPDD-discuss(a)lists.01.org
https://lists.01.org/mailman/listinfo/hpdd-discuss