Multiple IP Ranges per Zone

[Jed Danner]

Hello!

I have a multi-data center setup with two existing networks that fit into my two exisitng storage zones.

Physical location #1:

dc1 = 10.10.0.0/24

Physical location #2:

dc2 = 10.20.0.0/24

We have added a new set of networks, one at each data center location:

172.10.0.0/24 (dc1)

172.20.0.0/24 (dc2)

We are getting the message 'Cannot find current zone for local ip X.X.X.X' when !watching the console. Additionally we will likely be moving some MySQL slaves into the new IP space.

I tried defining dc1 as:

dc1 = 10.10.0.0/24,172.10.0.0/24

But when I do so my replicate jobs fail (pretty consistenly):

# Child 10938 (replicate) died: 256 (UNEXPECTED)

So where to go from here? I've thought of two solutions (not sure if both will work out properly), anyone have a similar situation or ideas?

Option #1 : NAT - pulling requests through an IP address that is in my original zone when coming from my new IP space should return local paths. NATing the MySQL slave in the new space will look like it's within the existing space.

Option #2 : New Zones - create new zones for the new space. MySQL slaves will work within the new zone and lookups from the new zone will match up. I'm currently using MultipleNetworks() I think this will still work since I won't have any storage nodes in the new space. This may become an issue if I add storage in here since I could get out of balance (files in only one data center).

Any other ideas or thoughts?

Thank you,

Jed

[Eric Wong]

Jed Danner <jed.d...@gmail.com> wrote:
> We are getting the message 'Cannot find current zone for local ip X.X.X.X'
> when !watching the console. Additionally we will likely be moving some
> MySQL slaves into the new IP space.
>
> I tried defining dc1 as:
>
> dc1 = 10.10.0.0/24,172.10.0.0/24

I wondered about this, but it looks like the check_cache subroutine
in MogileFS/Network.pm is doing the right thing with comma-delimited
zone_masks. I'm looking at the 0.06 version on:
git://github.com/mogilefs/MogileFS-Network.git

I assume you waited >= 60s for the cache to expire?

> But when I do so my replicate jobs fail (pretty consistenly):
>
> # Child 10938 (replicate) died: 256 (UNEXPECTED)

Any other error messages? (check syslogs).

Can you check if your trackers actually talk to the machines in the new
zone? Can you try doing some large transfers to/from the new network to
the trackers and see if things run at full speed? It could also be a
case where the normal monitor requests are too small to show a problem.

Also, did you deploy new trackers? The latest Sys::Syscall is/was
broken, so maybe going to an older version can fix it.

[okwap]

I apply two class C in a zone recently, and do not notice the issue you metioned.