Replica sets re-enabled for local provider

Andrew Wilkins andrew.wilkins at canonical.com
Mon Jun 9 06:09:40 UTC 2014


On Mon, Jun 9, 2014 at 10:04 AM, Andrew Wilkins <
andrew.wilkins at canonical.com> wrote:

> On Sat, Jun 7, 2014 at 1:32 AM, Nate Finch <nate.finch at canonical.com>
> wrote:
>
>> That is awesome.   I presume that means we can re-re-re-enable HA on
>> local provider?  At least Michael has all the code to do so now, so it
>> should be quick this time.
>>
>
> Yes, I think we should re-enable unconditional HA now.
> I'd like to get to the bottom of this, but I concur with Curtis that we
> should not be too concerned about supporting ephemeral disks. It's not
> something we'd ever recommend.
>

I spent a little bit more time looking at this this afternoon. I had
previously thought that the I/O speed wouldn't be the issue, since the
ephemeral disk is faster than the EBS-backed one on those nodes. I
remembered earlier that the replication oplog's size is, by default, a
percentage (5%) of the available space on the disk that contains the db.

With a cleanish install of 14.04 I have ~5.3G free on /, and ~380G free on
/mnt. We don't preallocate the oplog files like we do for the journal, so
mongo creates the files and takes its time about it. It seems that this is
what is causing the problem. I set the oplog size to 1G (passing
"--oplogSize 1024" to mongod) and it all works.

I've created https://bugs.launchpad.net/juju-core/+bug/1327940

Very nice work, Andrew.  I'm not sure it ever would have occurred to me
>> that the storage type would matter.
>>
>>
>> On Fri, Jun 6, 2014 at 12:56 PM, Curtis Hovey-Canonical <
>> curtis at canonical.com> wrote:
>>
>>> On Fri, Jun 6, 2014 at 4:50 AM, Andrew Wilkins
>>> <andrew.wilkins at canonical.com> wrote:
>>> > I've gotten a bit closer to figuring out what the deal is. Even on the
>>> CI
>>> > machine, local provider works with HA enabled, but only if you use the
>>> EBS
>>> > storage. If you use that machine's ephemeral disk (as your jenkins
>>> setup
>>> > does), then mongo is not happy. The fact that it's running Precise has
>>> > nothing to do with it. I can reproduce it on Trusty.
>>>
>>> I bow to your mad skillz. I symlinked /var/lib/mongodb to the EBS
>>> volume. We did this for jenkins and LXC. The tests pass. We have a new
>>> blessed revision.
>>>
>>> I think this is an edge case for the local provider. I would expect
>>> special support for ephemeral disk
>>>
>>>
>>> --
>>> Curtis Hovey
>>> Canonical Cloud Development and Operations
>>> http://launchpad.net/~sinzui
>>>
>>> --
>>> Juju-dev mailing list
>>> Juju-dev at lists.ubuntu.com
>>> Modify settings or unsubscribe at:
>>> https://lists.ubuntu.com/mailman/listinfo/juju-dev
>>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/juju-dev/attachments/20140609/8f0db3b6/attachment-0001.html>


More information about the Juju-dev mailing list