fuel-dev team mailing list archive

[Sergii Golovatiuk <sgolovatiuk@xxxxxxxxxxxx>]

Hi Fuelers,

While working on memcache HA issues I dug the problem, thus want to bring
up to your attention.

Currently, MOS services (keystone, horizon) uses python-memcached plugin.
The plugin itself is not bad but lacks of features for High Availability.

Firstly, it's not possible to specify how much time need to wait for an
answer before marking server as down. Also there are no way to specify how
often to check the server before bringing it back online.

However, it's possible to use python-pylibmc which is wrapper over
limemcached that has many options for tuning. So it requires slight
modification
1. We need to package libmemcached and python-pylibmc and include to MOS
2. there should be additional option in keystone.conf > [cache] >
backend_behavior which will allow operators to specify options for memcache
backend. This requires a slight modification in keystone.
3. We should package django-pylibmc for horizon. That will allow horizon to
use python-pylibmc
4. Modify Puppet manifest in order to configure keystone and horizon as
well as to install all required packages.

Once done, memcached related services will be more robust, allowing
operators to tune settings for their needs.

Feel free to share your thoughts how to make memcached related services HA
ready. Thank you in advance.
--
Best regards,
Sergii Golovatiuk,
Skype #golserge
IRC #holser