default disk size for MongoDB is too small

[Matthew Mosesohn]

The / partition on a MongoDB host should be 60gb to support MongoDB logging in debug mode. The default 16gb size is simply inadequate if the system is deployed in Debug mode and then Rally is used for testing. It will fill up well before log rotation occurs and within a few days, even rotated logs will fill the disk.

I had a conversation with Nikolay M and Vitaly K and they seemed to think this was possible to implement, although a bulletproof workflow setup is near impossible. This affects the 100 node deploy scale project, so it deserves high priority.

[Dima Shulyak (dshulyak)]

Currently disk partition strategy for mongo is:

"min" for os - /
"all" for mongo - /var/lib/mongo

we can add another partition, log - /var/log, with "min" allocator - 60 gb, or "all" allocator,
in case of all - space between mongo and log will be splitted
also note that it is important not to add /var partition, because we install os on a plain partition in case of ubuntu

[Vladimir Kuklin (vkuklin)]

I would say that this bug is partially a duplicate of https://bugs.launchpad.net/fuel/+bug/1393400

[Roman Prykhodchenko (romcheg)]

Do we still need to increase the size after logging level was reduced in https://bugs.launchpad.net/fuel/+bug/1393400 ?

[Matthew Mosesohn (raytrac3r)]

I guess it can be lowered in priority because it hasn't come back after reducing verbosity and rotation is far more effective now. There was another discussion about setting proportional disk sizes that are relative to the total disk space available.

[Roman Prykhodchenko (romcheg)]

Discussed that with some other guys and we agreed to lower its priority.

[Ivan Berezovskiy (iberezovskiy)]

Logging level for Mongodb was reduced. Now it doesn't affect deploy on scale. We should discuss about setting proportional disk sizes, but it should be decided for 7.0 becuase for 6.1 it's too late (SCF is now).

[Aleksey Kasatkin (alekseyk-ru)]

Does not break deployment according to latest comments. Moving to 8.0.

[Dmitry Klenov (dklenov)]

@ Alexander Rozdymakha, it is the bad practice to just unassign bugs from yourself. Please always find the proper team and assign bugs to it.

[Ivan Berezovskiy (iberezovskiy)]

Mongo oplog_size Mongo is 10Gb (https://github.com/openstack/fuel-library/blob/master/deployment/puppet/openstack_tasks/manifests/roles/mongo.pp#L63). So, in default configuration for base tests 16 Gb is more than enough (10Gb for oplog is a lot of more that actually needed most of the time, and 6Gb for other needs). This other space (6 Gb) will be used for user data (in the majority it's ceilometer data).
But for long-living environments fuel user should take care of disk size for mongo before deployment and set it depending openstack cluster usage (and as result how much data will be collected by ceilometer).

So, I think that bug is invalid because users have an ability to configure disk for mongo and they are motly responsible for setting it in correct way.

[Ivan Ponomarev (ivanzipfer)]

Agree with @iberezovskiy