On 20 April 2016 at 10:16, Eyal Edri <eedri@redhat.com> wrote:
>
>
> On Wed, Apr 20, 2016 at 9:28 AM, Barak Korren <bkorren@redhat.com> wrote:
>>
>> > I'd try that approach first, though the mirror is a good idea that will
>> > probably have to be implemented anyhow once we start adding slaves,
>> > having real
>> > info on the network usage/errors will give us insight to actually
>> > determine
>> > what's the issue, and thus, what's the best solution.
>> >
>> > @infra what do you think?
>> >
>>
>> The issue with mirroring is how can you make sure that you mirror fast
>> enough to enable CI. Even if Gerrit can push to the mirror on patch
>> submission, there will still be some time delta between the submission
>> happening (and the patch event showing up in Jenins) and the mirror
>> being synced. This looks like a nasty race condition.
>> What the mirror essentially does is make sure that bits are copied
>> from Amazom to PHX just once. I wonder if we can get the same benefit
>> with a simple HTTP proxy, how proxy-able is the Git HTTP protocol?
>>
>
> I think we should prioritize mirroring the GIT (not gerrit) repos to PHX,
> this will help:
>
> Speed up all post merge jobs and reduce potential of errors from git clone
> (they will be in the same network)
> Reduce load (?) from the gerrit server and perhaps reduce errors of the per
> patch jobs that will still run from gerrit.ovirt.org (AMAZON)
> A longer goal will be either to migrate the gerrit server to PHX or to find
> away to properly mirror the gerrit server (but then i fear there might be
> race/problem as mentioned)
>

Please look at my comment about possible race conditions caused by
mirroring. Simple mirroring may cause more trouble then its worth. We
need to consider proxying instead.