Thoughts on Openstack London 2014

Image

 

I attended Openstack London this week in the beautiful surroundings of Bishopsgate 155, right in the centre of the ‘city’.

A few thoughts:

Ceph

Although I have not deployed Ceph yet it is something I am very interested in.  I love the idea of disrupted feature rich storage on commodity hardware.  John Spray from Inktank did an awesome job of taking us through an introduction to Ceph outlining why distributed storage is hard to do and how Ceph attacks those challenges.  He then talked about the most recent features in the latest release.  I particularily like Cache tiering:

  • Cache tiering: support for creating ‘cache pools’ that store hot, recently accessed objects with automatic demotion of colder data to a base tier. Typically the cache pool is backed by faster storage devices like SSDs.

This is great; have the reads on SSDs and the replicas on slow mechanical disks.  I can see so many use cases for that

  • Primary affinity: Ceph now has the ability to skew selection of OSDs as the “primary” copy, which allows the read workload to be cheaply skewed away from parts of the cluster without migrating any data.

   This is also very cool from an operational standpoint.

Finally John demo’d calamari – the previously enterprise only web interface and management tool for Ceph.  I am sure this will be very useful in production deploys.  Kudos to Redhat for opensourcing this as part of the Inktank aqusition.

Juju

Canonical were sponsors and presenters at the event.  Mark Shuttleworth had a somewhat shaky time doing his demo’s but overall it was pretty interesting.  I already knew about cloud init etc so nothing massively new there.  My takeaway was Juju which I had also dismissed as Canonicals take on config management. I know realise that it’s much more of an Orchestration tool, like Mcollective.  Will try and carve out some time to test some Juju deploys later this week.

Networking

Networking is obviously a core component of Openstack and a few talks covered aspects, with Mark McClain doing the deepest dive.  Of all the components in Openstack, Neutron and networking is the one I think is the most challenging.  The loadbalanacing on demand was not really covered but was interested to hear it’s based on Haproxy but does not support SSL yet.

 

 

 

Advertisements

Checking for Postgres streaming replication delay

Postgres streaming replication is very cool but lacks the simplicity of show slave status; as in Mysql to determine how far behind a slave is from the master

Here is a simple query to check the delay approximately

we can also make a simple bash script to make execution easier:

Just a Pingdom Dashboard in 26 lines of python

We like to have various stats on the health of our websites on large 37in TV screens so that everyone can see what is going on.

Pingdom is a service we use for monitoring web page uptime and speed, I wrote a quick dashboard on Python the Parse the Pingdom API and display on the big screens.

Read more of this post

Watching US Netflix in the UK

Netflix is a great service which has just recently launched in the UK.  Unfortunately the UK selection  of content is poor compared to what is available in the US.  Fortunately there is an easy solution Read more of this post

Testing Akamai with curl

Sometimes with Akamai we need to test various things such as, if the asset we are caching is being cached or if the origin server is faster than the Akamai Edge Servers (useful for talking to Akamai support) Read more of this post

Integrating ifttt and Pingdom

Ifttt is an amazing service that allows carious web services and social network APIs to integrate really nicely.  For example, if I post a picture to instagram then post the same picture to flickr or if I change my Facebook avatar also update my Twitter avatar. These integrations take the form of recipes. Read more of this post