On Sat, Mar 17, 2018 at 8:06 AM, xsong682@xxxxxxxx <xsong682@xxxxxxxx> wrote:
> In current code, before PG peered, PG can’t do any operations including reading in following code
>
> void PrimaryLogPG::do_request( OpRequestRef& op, ThreadPool::TPHandle &handle)
> {
>
> if (!is_peered()) {
> // Delay unless PGBackend says it's ok
> if (pgbackend->can_handle_while_inactive(op)) {
> bool handled = pgbackend->handle_message(op);
> assert(handled);
> return;
> } else {
> waiting_for_peered.push_back(op);
> op->mark_delayed("waiting for peered");
> return;
> }
> }
>
> that will impact performance, I think the primary OSD control the data writing, if it haven’t got changed, the data on the primary OSD is always correct except silent data error, so the data can be read safely before peered.
> If it's ok, another question is how to know the OSD haven’t changed? Checking “primary of acting_set == primary of up_set” is enough or not? Should we check interval for the situation primary OSD change from “A-->B-->A”?
Knowing if the OSDs changed is what peering is for, and is why we
don’t allow reads on unpeered PGs. If we did allow you to read without
peering, you might read the data from an old and stale OSD that was
fenced off from the rest of the cluster and didn’t know it!
-Greg
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
