Category Archives: Uncategorized

DevOps in a Bi-Modal World

The business environment has never been more competitive and disruptive than it is today. Businesses need to come to terms with three realities:

  1. They need a continuous competitive advantage

Just ask Kodak who has seen the camera business transform from a standalone device to a feature on every mobile phone with new players like Snapfish, Shutterfly, and Chatbooks creating new ways of engaging with markets. If you don’t have a way of continually developing new competitive advantages you will not be relevant for long.

  1. They are a software company

Bank of America is not just a bank, they are a transaction processing company. Exxon Mobil, is not only an oil and gas company, they are a GIS company. With each passing day Walgreens business is more reliant on electronic health records.

  1. Their competition is everywhere

Ten years ago if I asked you who the biggest competitive threats were to Fedex names like UPS, and DHL might come to mind. Increasingly Fedex, UPS, and DHL face threats from Uber, Walmart, Amazon, and others who may enter their market of logistics with new ways of reaching customers.

What do businesses need to do given these three realities?

To quote Mark Zuckerberg, they need to “Move fast and be stable”. Moving fast and being stable can be translated to more quickly developing new services that could be scaled to meet fast growing demand if needed but also with an extremely low cost of failure should they not work. In other words, cheap experiments need to be able to become global successes.

The scientists conducting these cheap experiments are software developers. Lines of business naturally turn towards their development teams to request new services at an increasingly faster rate. The problem is, developers can’t obtain those environments fast enough from operations because traditional processes and non-flexible infrastructure and applications stand in the way. It’s no surprise then, according to a 2012 McKinsey and Company study [1], that software delivery in the enterprises surveyed was 45% over budget, 7% over on time, and has 56% less value than expected when delivered.

This is no secret to businesses and they are looking to new methods and designs to help improve these metrics. In fact:

  • Over 90 percent are running or experimenting with Infrastructure as a Service [2].
  • Greater than 70 percent expect to use Platform as a Service in their organization [3].
  • More than 90 percent expect new investments in DevOps enabling technologies in the next two years [4].
  • Over 70 are using or anticipate using containers for cloud applications.

Businesses are turning to new development and operations processes, new cloud infrastructures, and application methodologies that are conducive to these new processes and infrastructures. Looking at one of the leaders in public cloud, Amazon Web Services, we see they use these same principles and designs to achieve upwards of 10,000 releases per hour (as much as a release every 12 seconds) with a very low outage rate caused by these releases.

At first glance it would appear that enterprises could simply yell “DevOps and Cloud to the Rescue!” and solve their problem of deploying faster on scalable infrastructure, but the reality is far from that. Enterprises have existing assets and investments, and many of these are not going away anytime soon. In fact, the existing systems and processes most likely power the very core of the business and cannot simply be replaced over night nor would they fit the paradigm of moving quickly and experimenting. Gartner coined the term Bi-modal to describe this approach of two modes of delivery for IT – one focused on agility and speed and the other on stability and accuracy.

Gartner has also recognized an approach that enterprises can take that would allow them to maximize the use of their existing assets. In their research “DevOps in the Bimodal Bridge” [7] they suggest an approach where the patterns and practices of DevOps can be applied to existing assets (mode 1) to make it more agile and efficient.

I have observed this trend and I believe most organizations are trying to address four key problems across their emerging bi-modal world.

In mode-1 they are looking to increase relevance and reduce complexity. In order to increase relevance they need to deliver environments for developers in minutes instead of days or weeks. In order to reduce complexity they need to implement policy driven automation to reduce the need for manual tasks.

In mode-2 they are looking to improve agility and increase scalability. In order to improve agility they need to create more agile development and operations processes and embrace new application architectures that allow for greater rates of change through decreased dependencies. In order to increase scalability they need to implement infrastructure that utilizes an asynchronous design and is entirely API driven in order to change the admin to host ratio from a linear to an exponential model in order to increase scalability.

In order to make these examples more concrete, let’s look at each of them in more detail.

Increasing Relevance by Accelerating Service Delivery

Delivering development and test environments to developers in many enterprises generally starts with either a request to a service management system or a tap on the shoulder of a system administrator. This usually depends on the size of the organization and maturity of the IT department. Either way, once requests fall into a service management system there are often many teams that need to perform tasks to deliver the environment to the developer. These might include virtual infrastructure administrators, systems administrators, and security operations. In larger organizations you could expect to see disaster recovery teams, networking teams, and many others involved in this process too. Again, depending on the maturity of the organization how all of this is coordinated could range from taps on shoulders to passing tickets around in a service management system.

At best each team takes minutes or hours to respond and perform some manual tasks and often the person who requests the service must be asked follow up questions (“Are you sure you need 16GB of RAM?”, “What version of Java do you need for this?”). The result is lots of highly skilled people spending lots of time and very slow delivery of this environment to the developer. Multiply this by the number of developers in an organization and the number of requests for environments and you can understand why traditional IT processes and systems are struggling to maintain relevance.

A solution for this problem is to introduce a service designer into the process (you may be familiar with this from ITIL) that can enable self-service consumption of everything developers need. The designer works with all stakeholders including virtual infrastructure administrators, system administrators, and security operations to obtain requirements. Then, the designer builds the necessary configuration management content and couples it with a service catalog item. By invoking this catalog item the environment can be deployed automatically across any number of providers including virtualization providers, private, or public cloud.

The result of this solution is that all the teams responsible for delivering an environment are now free to do more valuable work (like working with development to design operations processes that work as part of development instead of being bolted on after). It also removes human error from the equation, and most importantly, it delivers the environment in significantly less time. We have seen upwards of a 95 percent improvement in delivery times in many of our customers [9].

Reducing Complexity by Optimizing IT

Speeding up delivery of environments to developers or end users is a great way to make IT more relevant, but a lot of what IT is spending their time on is the day-to-day management of those environments. If IT is spending so much time on day-to-day tasks how can we expect them to deploy the next generation of scalable and programmable infrastructure or have time to work with development teams during early stages of development to increase agility?

I have found that many virtual infrastructure administrators spend time on several common tasks that should be largely automated through policy.

First, are policies around workload placement. Often one virtual infrastructure cluster will be running hot while another one is completely cold. This leads to operations teams being inundated by calls from the owners of applications running on the hot cluster asking why response times are poor. Automating this balancing through control policies can alleviate this problem and keep virtual infrastructure administrators free to other things.

Next is the ability to quickly move workloads between different infrastructures. This has become increasingly important as organizations looks to adopt scale-out IaaS clouds. Operations leadership realizes if they can identify workloads that do not need to run on (typically) more expensive virtual infrastructure they could save money by moving those workloads to their IaaS private cloud. This migration is typically a manual process and it’s also difficult to even understand what workloads can be moved. By having a systematic and automated way of identifying and migrating workloads enterprises can save time and move workloads quickly to reduce costs.

Yet another issue is ensuring compliance and governance requirements are met, particularly with workloads running on new infrastructures, like an OpenStack based private cloud. Not knowing what users, groups, data, applications, and packages reside on systems running across a heterogeneous mix of infrastructure presents a large risk and operations teams often have the responsibility and obligation of ensuring this risk is minimized. By being able to introspect workloads across platforms operations teams can gain insight into exactly what users, data, and packages are running on systems and leverage the migration capabilities I mentioned previously to make sure systems are running on appropriate providers.

Finally, since IT has often become a broker of public cloud services it’s important that they can account for costs and place workloads on appropriate regions in the public cloud to control costs while also ensuring service levels for end users are maintained. If developers are based in Singapore then we should leverage public cloud infrastructure in that location instead of deploying to a more expensive and more latent public cloud infrastructure in Tokyo.

By implementing policy based automations our customers have seen large improvements in their resource utilization and a reduction in CapEx and OpEx per workload managed [10].

Improving Agility by Modernizing Development and Operations

With resources now free from handling each and every inbound request for an environment and being confident that those environments are running efficiently and securely on the right providers operations teams can begin to work with development teams to design new processes for their cloud native applications.

These newly designed processes and cross-functional team structure combined with a platforms that supports running the broadest amount of languages and frameworks within microservices based architectures will enable the development and operations teams to achieve higher release frequencies. By utilizing microservices and standardized platforms and configurations these new applications will allow for independent release and scaling of components of the application.

This results in an increased success rate of change, faster cycle time, and the ability to scale specific services independently, making the life of both development and operations teams easier and allowing them to meet the needs of the line of business. We have experience doing this with very large software development organizations [11].

Scalability with Programmable Infrastructure

As agility of development and operations processes is improved and release frequency increased, so to does the demand for more scalable infrastructure to run those releases on. Operations teams face the challenge of delivering infrastructure that will scale to meet the demand of this ever-growing number of applications. The last thing the head of operations would like to have to explain to the management team of a company is why an extremely successful new application was hitting a wall as to the maximum number of users it could support. This simply can’t happen. Unfortunately, the current infrastructure is not scalable, neither from a financial nor technical standpoint.

One option might be to build out a scale-out infrastructure, perhaps based on OpenStack, the leading open source project for infrastructure-as-a-service. However, the operations team doesn’t want to spend it’s time taking open source code and making it consumable and sustainable for the enterprise. It doesn’t have the resources to test and certify that OpenStack will work with each new piece of hardware it brings in. It also can’t afford to maintain the code base for long periods of time with the resources available. Finally, OpenStack is missing key features that operations needs and they don’t want to develop those in house as well.

What operations really needs is a way to minimize cost and increase scale through the use of commodity hardware and a massively scalable distributed architecture coupled with the enterprise management features required to operate that infrastructure and a stable, tested, certified way of consuming the open source projects that make up that infrastructure. By having this, operations can deploy scale-out infrastructure in multiple locations and still aggregate management functions like chargeback, utilization, governance, and workflows into a single logical location. Many of our customers have found this solution beneficial in reducing cost and ensuring stability at scale [12].

Introducing Red Hat Cloud Suite

Red Hat Cloud Suite is a family of suites from Red Hat that brings together all the award winning products from Red Hat in a consistent way to solve specific problems. It allows IT to accelerate service delivery and optimize their existing assets while allowing them to build their next generation infrastructure and application platforms to support massive salability and more agile development and operations processes. In other words, it meets them where they are and lays the foundation for where they want to go.

A Different Approach

It should come as no surprise to you that Red Hat is not the only company solving these problems. Red Hat is, however, one of the few companies that can solve all of these problems because of its broad portfolio of technologies and expertise. Most think of Red Hat as having the largest percentage of the paid Linux market share. That is true, but Red Hat has been adding to its portfolio and has grown acquired expertise and industry leading technology from Software Defined Storage [13][14] to Mobile Development Platforms [15]. These offerings place Red Hat alone with only Microsoft in terms of depth of capability.

An Important Difference

Along with this depth of expertise and capability comes an approach that sets Red Hat apart. Red Hat is the only vendor that uses an open source development model for all of the solutions it delivers. This is important for customers because the world of cloud infrastructure and applications and DevOps is built entirely on open source software. By having a strict open source only mentality customers can have access to the greatest amount of innovation and be ensured that as technologies change they could adopt them more easily because Red Hat can adopt and deliver these technologies. Two great examples of this are how Red Hat adopted the KVM hypervisor [16] and embraced and delivered it’s container platform with support for Docker and Kubernetes [17] – leading open source projects that become popular in a short amount of time. Red Hat is committed to the open source development model, so much so that it even creates communities when it acquires non-open source licensed technologies [18]. Customers should know that when they leverage a solution from Red Hat it is based entirely on open source, leading to greater access to innovation and lower exit costs.

Technical Capabilities are Important Too

While philosophical differences are important for ensuring that the right long term decisions have been made, Red Hat is also at the forefront of innovation in cloud infrastructure, applications, and DevOps tools.

True Hybrid Support

The term hybrid cloud has often been over used and abused, but it is important. Enterprises need to be able to run workloads across the four major deployment models that exist today: physical, virtual, private, and public cloud. Equally as important to the deployment model is the ability to support multiple service models, such infrastructure-as-a-service, platform-as-a-service, or even bare metal, virtual machines running on scale-up virtual infrastructure, and public cloud services. When most vendors claim they support “hybrid” cloud they are typically limited to only managing hybrid deployment models. Red Hat supports both hybrid deployment models and hybrid service models. This is important to both Development and Operations teams. For developers, it means being able to develop on the broadest choice of languages and frameworks. They could use an Oracle database running on bare-metal or virtual machine, JBoss EAP running on virtual machines on OpenStack, combined with Node.js and Ruby running in Containers on OpenShift. They are not constrained to a single service model that doesn’t give them everything they need.

Using Big Data to Optimize IT

Red Hat has been supporting Linux for a long time. In fact, we’ve been supporting Red Hat Enterprise Linux for over 13 years since RHEL AS 2.1’s release in 2002 [16]. There are over 700 Red Hat Certified Engineers in our support organization and they’ve documented over 30,000 solutions while resolving over 1 million technical issues. The Red Hat customer portal has won plenty of awards for helping connect customers searching for resolution to an issue to the right technical solution. With Red Hat Access Insights, Red Hat’s new predictive analytics service, connecting support data to recommendations is going to reach a new level of ease of use. Users can send small amounts of data about their environment back to Red Hat and it will be compared to optimal configurations to find opportunities to improve security, reliability, availability, and performance. This service is already available for Red Hat Enterprise Linux and will soon be available for all the technologies in Red Hat’s portfolio through Red Hat Cloud Suite.

An Easy On-ramp and Consistent Lifecycle

Deploying a private cloud is not an easy task. The list of platforms that need to come together from configuration management, to storage, to infrastructure-as-a-service, to platform-as-a-service is large. Each of these has dependencies on sub-components within each of these platforms. For example, to generate new docker images need secure content and that takes integration between the content management system and the image building services. Literally hundreds of these integrations are needed to build a fully functional private cloud. This usually results in one of two options:

  • Operations requiring lots and lots of time to deliver this private cloud.
  • An army of high priced consultants arriving to deliver and maintain a private cloud.

Neither of these options are an optimal results for IT. Red Hat Cloud Suite provides an easy on-ramp that allows a single person in operations to deploy a private cloud and it provides the path for ongoing management of that private cloud. This allows developers to begin using the private cloud more quickly and helps operations deliver a private cloud more quickly.

A Quick Summary

Here is a quick summary for those that just want the cliff notes.

The World is Changing

  • Businesses need a continuous competitive advantage
  • All businesses are software companies
  • Competition is everywhere

IT Needs

  • To increase relevance and reduce complexity
  • To create more agile processes and build programmable & scalable infrastructure and platforms

Red Hat Helps

  • Accelerate delivery
  • Optimize for efficiency
  • Modernize development and operations
  • Deliver scalable infrastructure

Only Red Hat Delivers

  • Innovation in the form of pure open source solutions
  • Integration with world class testing, support, and certification





[4] DevOps, Open Source, and Business Agility. Lessons Learned from Early Adopters. An IDC InfoBrief, sponsored by Red Hat | June 2015















Red Hat Cloud Suite for Applications

For those following our recent announcement, I put together a short blog post that explains why Red Hat Cloud Suite for Applications is the only on-premise complete and open source solution for accelerating application delivery at scale.

Docker all the OpenStack Services Presentation

Slides from the presentation Brent Holden and I gave at OpenStack Summit can be downloaded here.

A Demonstration of Kolla: Docker and Kubernetes based Deployment of OpenStack Services on Atomic

The Problem

Screen Shot 2014-10-22 at 8.39.48 AM

The Beauty of OpenStack

OpenStack is a thing of beauty, isn’t it? Just look at all those cleanly defined services, perfectly atomic, able to run standalone … it’s simply amazing. What more could developers and operators ask for in a cloud?

Screen Shot 2014-10-22 at 8.42.30 AM

The Reality of OpenStack

Except, that it’s not exactly like that. All those services heavily rely on each other and given the rate of change OpenStack is experiencing the degree of complexity only stands to increase. The problem is that OpenStack has many services that are dependent on one another and managing the lifecycle is difficult and inefficient because of this.

Let’s look at an example of updating the keystone service, OpenStack’s identity management service. It is difficult to know whether or not deploying a new version of Keystone into an existing OpenStack deployment will cause problems because of compatibility with others services. It’s also difficult to move backwards and expensive to roll back a deployment of a new keystone service with today’s tools. Operators don’t want to use extra racks of hardware to test an upgrade of a service if they can avoid it and no lifecycle management tools that try to imperatively deploy and roll back can do so as reliability as we’d like between OpenStack releases.

At this point you might conclude that I have a personal vendetta against OpenStack. Although this could be justified after the many nights I’ve spent installing, configuring, and upgrading OpenStack I can assure you that’s not the case. In truth, OpenStack is not a beautiful and unique snowflake. Lots of different infrastructure platforms face this same problem and so do many application platforms.

The Many Paths to OpenStack Lifecycle Management

Today, there are many ways to manage the lifecycle of OpenStack services, but the two most prevalent can be loosely grouped into two categories: build based and image based deployments.

Build based lifecycle management uses a build service, such as PXE, and is typically coupled together with a bunch of lifecycle management tools and  almost always uses some type of configuration management whether that’s Puppet, Chef, Ansible, or others.

Screen Shot 2014-10-22 at 9.40.38 AM

This approach is generally inefficient because each OpenStack service is placed onto a different physical piece of hardware or at least a different operating system.

Screen Shot 2014-10-22 at 9.45.48 AM

It is possible to combine multiple services on a single operating system, but this can get tricky. How does the lifecycle management tool know that OpenStack Service A in the image above won’t conflict with OpenStack Service B in terms of resources required, ports required, file systems, etc? It takes an awful lot of logic in a lifecycle management tool to know this and given the rate of change experienced in a community like OpenStack, lifecycle management tools have a hard time keeping up and delivering what users would like to deploy. Could virtual machines be used here? Possibly, but virtual machine are heavyweight and also lack rich metadata or require large infrastructures and agents loaded into those virtual machines to get metadata. In other words, VMs are too heavy and they also lack the concept of inheritance.

Screen Shot 2014-10-22 at 10.00.53 AM

Finally, build based deployments can be slow. Copying each package back and forth over the wire is not the most efficient way of deploying at scale.

Image based deployments solve the problem of slow performance that build based systems have by not requiring each package to be installed. Typically an image based system has some sort of image building tool that stores images in a repository and these images are then streamed down to physical hardware.

Screen Shot 2014-10-22 at 10.12.33 AM

However, even while using images, incremental updates can be slow due to the large size of images. Also, the expense of pushing a large image around for small incremental updates doesn’t seem appropriate.

Screen Shot 2014-10-22 at 10.12.42 AM

Even more importantly, image based deployments don’t solve the fundamental problem of complexity that understanding the relationships between OpenStack services presents. This problem is only moved earlier in the process and must be solved when building the images themselves instead of at run-time.

There is one other consideration that should be taken when looking at building a lifecycle management solution for OpenStack and that is that OpenStack doesn’t live alone. The last thing most operators want is yet another way to manage the lifecycle of a new platform. They’d like something that they can use across platforms from bare metal, to IaaS, and possibly even in a PaaS.


What Atomic, Docker, and Kubernetes Bring to the Party

Wouldn’t it be great if there was a solution for managing the lifecycle of Openstack services that was:

  • Isolated, lightweight, Portable, and Separated
  • Easily Described run-time relationships
  • Could run on something thin and easy to update
  • Worked to manage the lifecycle of services beyond OpenStack too

That’s exactly what the combination of Docker, Kubernetes, and Atomic can provide to the existing lifecycle management solutions.

Screen Shot 2014-10-22 at 10.32.57 AM

Docker provides a level of abstraction for Linux Containers through APIs and an “Engine”. It also provides an image format for sharing that supports a base and child image relationship allowing for layering. Finally, Docker provides a registry for sharing docker images. This is important because it allows developers to ship a portable image that operators can deploy on a different platform.

Screen Shot 2014-10-22 at 10.34.25 AM

Kubernetes is an open source container cluster manager. It provides scheduling of Linux Containers using a master/minion construct. It uses a declarative syntax to express desired state. This is important because it allows developers to provide a description of the relationships between different Linux Containers and let’s the cluster manager do the scheduling.

Screen Shot 2014-10-22 at 10.35.39 AM

Atomic provides just enough of an operating system to run containers in a secure, stable, and high performance manner. It includes Kubernetes and Docker and allows for users to update using newly developed update mechanisms such as OSTree. Here is a quick video that shows how easy it is to deploy atomic (in this case on OpenStack) and also how easy it is to upgrade Atomic. Watch OGG


Screen Shot 2014-10-23 at 2.20.49 PM

So when you put these pieces together what you end up with is something that looks (at a high level) like the diagram above. OpenStack developers are free to develop on a broad choice of platforms (Linux/Vagrant/Libvirt pictured) and can publish completed images to a registry. Operators on the other side would pull the kubernetes configurations into their lifecycle management tools and the tools would launch the pods and services. This would trigger Docker running on Atomic to pull the images locally and deploy containers with the OpenStack services. Services are isolated and (we are fairly certain given our experience with our OpenShift PaaS) lots and lots of containers could be run on a single operating system to maximize density of Openstack services. There are LOTS of other benefits including ease of rollback, deployment and update speed, etc, but this alone should be enough for anyone looking at running an OpenStack cloud at scale to be interested.

 Show me the Demo!

Screen Shot 2014-10-23 at 2.21.29 PM

Here are several demonstrations that illustrate the scenario above. These are a demonstration of the OpenStack Kolla project and were produced in 2 weeks time by a group of amazing developers who saw the potential these technologies had.

First there is building the images and pushing them to a registry.  Watch OGG

Second there is deploying a few pods and services manually to see how they connect and what Kubernetes and Docker are actually doing. Watch OGG

Finally, there is an example of deploying all the OpenStack services that were completed in milestone-1 all with a single command.  Watch OGG

After deploying OpenStack countless times I can say that when you see each schema automatically created in MariaDB and endpoints, services, etc automatically created all in under a minute it is an amazing feeling!

“I’m Sold, What’s Next?”

In the end, the combination of Docker, Atomic, and Kubernetes show the promise of alleviating some of the pain OpenStack developers and operators have experienced. There are still a lot of unanswered questions, but we feel that this combination of technologies shows promise and are excited that they have found a home in the TripleO project through Kolla.

If you are interested in learning more or participating please:

If you want to learn more about some of the other projects related to this post please check out the following:

Docker All The Things – OpenStack Keystone and Ceilometer Automated Builds on Docker Hub

Ok, I borrowed part of the title of this post from Nicola Paolucci at Atlassian’s blog post who likely borrowed it from a bunch of others, but it was just too good to pass up.

I decided to test Docker Hub’s automated build feature to see if I could have automated docker images created from a project relevant to Red Hat Cloud Infrastructure (RHCI), Red Hat’s private IaaS cloud solution. RHCI combines datacenter virtualization based on Red Hat Enterprise Virtualization (RHEV), scale out IaaS based on Red Hat Enterprise Linux OpenStack Platform (RHELOSP), and cloud management based on CloudForms. These come from the upstream communities of oVirt, OpenStack, and ManageIQ.

If you are interested in why containers could be so beneficial to an Infrastructure as a Service solution you could read my previous post, “Why containers for OpenStack Services?”. The bottom line is that moving more logic about the lifecycle of the IaaS services into the application layer (Think PaaS for IaaS) could solve many problems and help IaaS become much easier to manage.

Keystone Docker Image

The natural choice for the first service to attempt to containerize was the identity service, Keystone. Keystone has (relative to other openstack projects) few moving parts and is also required by most of the other services since it publishes a catalog of endpoints for the other services APIs.


Here is what I did:

1. I forked the openstack keystone project.

Screen Shot 2014-07-07 at 9.15.50 PM

2. I created a new automated build on Docker Hub.

Screen Shot 2014-07-01 at 12.34.24 PM

3. I cloned the git repository for my fork of keystone.

jlabocki@localhost# git clone
Cloning into 'keystone'...
remote: Counting objects: 26085, done.
remote: Compressing objects: 100% (9122/9112), done.
remote: Total 26085 (delta 16076), reused 26085 (delta 16076)
Receiving objects: 100% (1285/1285), 5.61 MiB | 2.14 MiB/s, done.
Resolving deltas: 100% (176/176), done.
Checking connectivity... done.

4. I created a Dockerfile in the root of the cloned keystone repository. Here is the contents of the Dockerfile.

FROM fedora

# This Dockerfile installs the components of Keystone in a docker image as a proof of concept

#Timestamps are always useful
RUN date > /root/date

#Install required packages
RUN yum install python-pbr git python-devel python-setuptools python-pip gcc gcc-devel libxml2-python libxslt-python python-lxml sqlite python-repoze-lru -y
#RUN yum install python-sqlite2 python-lxml python-greenlet-devel python-ldap sqlite-devel openldap-devel -y

#Clone Keystone and setup
RUN git clone
WORKDIR /opt/keystone
RUN python install

#Configure Keystone
RUN mkdir -p /etc/keystone
RUN cp etc/keystone.conf.sample /etc/keystone/keystone.conf
RUN cp etc/keystone-paste.ini /etc/keystone/
RUN sed -ri 's/#driver=keystone.identity.backends.sql.Identity/driver=keystone.identity.backends.sql.Identity/' /etc/keystone/keystone.conf 
RUN sed -ri 's/#connection=<None>/connection=sqlite:\/\/\/keystone.db/' /etc/keystone/keystone.conf
RUN sed -ri 's/#idle_timeout=3600/idle_timeout=200/' /etc/keystone/keystone.conf
RUN sed -ri 's/#admin_token=ADMIN/admin_token=ADMIN/' /etc/keystone/keystone.conf

# The following sections build a script that will be executed on launch via ENTRYPOINT

## Start Keystone
RUN echo "#!/bin/bash" > /root/
RUN echo "/usr/bin/keystone-manage db_sync" >> /root/
RUN echo "/usr/bin/keystone-all &" >> /root/

## Create Services
#I'm not sure if exporting works, so I just specify these environment variables on each command, but it might be cleaner to test this
#RUN export OS_SERVICE_ENDPOINT=http://localhost:35357/v2.0
#RUN export OS_AUTH_URL=
RUN echo '/usr/bin/keystone --os_auth_url --os-token ADMIN --os-endpoint service-create --name=ceilometer --type=metering --description="Ceilometer Service"' >> /root/
RUN echo '/usr/bin/keystone --os_auth_url --os-token ADMIN --os-endpoint service-create --name=keystone --type=identity --description="OpenStack Identity"' >> /root/
RUN chmod 755 /root/

#This you will need to substitute your values and run later - the values are:
# CEILOMETER_SERVICE = the id of the service created by the keystone service-create command
# KEYSTONE_SERVICE = the id of the service created by the keystone service-create command
# CEILOMETER_SERVICE_HOST = the host where the Ceilometer API is running
# KEYSTONE_SERVICE_HOST = the host where the Keystone API is running
RUN echo 'keystone --os_auth_url --os-token ADMIN --os-endpoint endpoint-create --region RegionOne --service_id $KEYSTONE_SERVER --publicurl "http://KEYSTONE_SERVICE_HOST:5000/v2.0" --internalurl "http://KEYSTONE_SERVICE_HOST:5000/v2.0" --adminurl "http://KEYSTONE_SERVICE_HOST:35357/v2.0"' > /root/
RUN echo 'keystone --os_auth_url --os-token ADMIN --os-endpoint endpoint-create --region RegionOne --service_id $CEILOMETER_SERVICE --publicurl "http://CEILOMETER_SERVICE_HOST:8777/"  --adminurl "http://CEILOMETER_SERVICE_HOST:8777/" --internalurl "http://CEILOMETER_SERVICE_HOST:8777/"' > /root/



5. I committed and pushed the change.

[root@localhost keystone]# git commit -m "testing" .
[master fe12eff] testing
1 file changed, 6 insertions(+), 7 deletions(-)
[root@localhost keystone]# git push
Username for ''
Password for '':
Counting objects: 14, done.
Delta compression using up to 2 threads.
Compressing objects: 100% (3/3), done.
Writing objects: 100% (3/3), 335 bytes | 0 bytes/s, done.
Total 3 (delta 2), reused 0 (delta 0)
a3de5a2..fe12eff  master -> master

6. The automated build finished successfully.

Screen Shot 2014-07-08 at 10.36.53 AM

7. I could check the details of the build, including the Dockerfile used and the output of the build.

Screen Shot 2014-07-08 at 10.38.11 AM

Assuming that the image was good and I could run it and setup Keystone rather quickly I decided to focus on another service and then attempt launching the two together (see the results section if you want to spoil the surprise).

Ceilometer Docker Image

I selected the OpenStack telemetry project, commonly known as Ceilometer, for my next test of an automated build of a docker image to take place on commit to my forked repository. Why Ceilometer? After looking at the OpenStack architecture diagram I thought it might be one of the easier services to run in a container (basically, I used a dart board), and since it only requires keystone I thought I might be able to make it happen next. Here are the components of OpenStack Ceilometer (Telemetry) at a glance taken from the OpenStack docs.

The telemetry system consists of the following basic components:

  • A compute agent (ceilometer-agent-compute). Runs on each compute node and polls for resource utilization statistics. There may be other types of agents in the future, but for now we will focus on creating the compute agent.
  • A central agent (ceilometer-agent-central). Runs on a central management server to poll for resource utilization statistics for resources not tied to instances or compute nodes.
  • A collector (ceilometer-collector). Runs on one or more central management servers to monitor the message queues (for notifications and for metering data coming from the agent). Notification messages are processed and turned into metering messages and sent back out onto the message bus using the appropriate topic. Telemetry messages are written to the data store without modification.
  • An alarm notifier (ceilometer-alarm-notifier). Runs on one or more central management servers to allow settting alarms based on threshold evaluation for a collection of samples.
  • A data store. A database capable of handling concurrent writes (from one or more collector instances) and reads (from the API server).
  • An API server (ceilometer-api). Runs on one or more central management servers to provide access to the data from the data store. These services communicate using the standard OpenStack messaging bus. Only the collector and API server have access to the data store.

Here it is in a diagram.


I decided to start with the database and ceilometer collector and then add the API. I went the route of placing all of these services in a single image. I’m aware there is a lot of debate as to whether Docker images should only run a single process or if multiple processes could be beneficial. My intention wasn’t to optimize the image for production, rather it was to test how easy or difficult it was to take a forked GitHub project and get it into an image build in an automated fashion that I could run on my Fedora 20 workstation. Also, I did not plan to add the evaluator, notifier, or any agents to this image. Since most of the agents require other components of OpenStack.

Here is what I did:

1. I forked the openstack ceilometer project.

Screen Shot 2014-07-01 at 9.21.07 AM

2. I created another new automated build on Docker Hub.

Screen Shot 2014-07-01 at 12.34.24 PM

3. I cloned the git repository for my fork of ceilometer.

jlabocki@localhost# git clone
Cloning into ‘ceilometer’…
remote: Counting objects: 26085, done.
remote: Compressing objects: 100% (7112/7112), done.
remote: Total 26085 (delta 16076), reused 26085 (delta 16076)
Receiving objects: 100% (26085/26085), 8.81 MiB | 3.14 MiB/s, done.
Resolving deltas: 100% (16076/16076), done.
Checking connectivity… done.

4. I created a Dockerfile in the root of the project following the manual installation of OpenStack Ceilometer. Here is the contents of the Dockerfile. Note I wasn’t able to run the mongod command during the build successfully. More on that later, I just created a post launch script that could be executed after the docker image is launched as a work around.

FROM fedora

# This Dockerfile installs some of the components of Ceilometer in a Docker Image as a proof of concept

#Timestamps are always useful
RUN date > /root/date

#Install required packages
RUN yum install mysql-devel openssl-devel wget unzip git mongodb mongodb-server python-devel mysql-server libmysqlclient-devel libffi-devel libxml2-devel libxslt-devel python-setuptools python-pip libffi libffi-devel gcc gcc-devel python-pip python-pbr mongodb python-pymongo rabbitmq-server -y

#RUN pip install tox
#Can't run the line above because, need to specify version 1.6.1
RUN pip install tox==1.6.1

#MongoDB Setup
RUN mkdir -p /data/db
RUN echo 'db.addUser("admin", "insecure", true);' > /root/mongosetup.js

#RabbitMQ Setup
RUN /usr/sbin/rabbitmq-server -detached

#Clone Ceilometer
RUN git clone /opt/stack/

#Ceilometer Collector Configuration
WORKDIR /opt/stack
RUN python install
RUN mkdir -p /etc/ceilometer
RUN tox -egenconfig
RUN cp /opt/stack/etc/ceilometer/*.json /etc/ceilometer
RUN cp /opt/stack/etc/ceilometer/*.yaml /etc/ceilometer
RUN cp /opt/stack/etc/ceilometer/ceilometer.conf.sample /etc/ceilometer/ceilometer.conf

#Ceilometer Collector Configuration changes
RUN sed -ri 's/#metering_secret=change this or be hacked/metering_secret=redhat/' /etc/ceilometer/ceilometer.conf
RUN sed -ri 's/#connection=<None>/connection = mongodb:\/\/admin:insecure@localhost:27017\/ceilometer/' /etc/ceilometer/ceilometer.conf

#Ceilometer API Configuration changes
RUN cp etc/ceilometer/api_paste.ini /etc/ceilometer/api_paste.ini

##Ceilometer Post Launch Configuration
RUN echo "#!/bin/bash" > /root/

#Add Authenticate against keystone to the post launch script
# KEYSTONE_HOST = the keystone host
RUN echo "sed -ri 's/#identity_uri=<None>/identity_uri=KEYSTONE_HOST/' /etc/ceilometer/ceilometer.conf" >> /root/

#Add starting services to the postlaunch script
RUN echo "/bin/mongod --dbpath /data/db --fork --logpath /root/mongo.log --noprealloc --smallfiles" >> /root/
RUN echo "/bin/mongo mydb /root/mongosetup.js" >> /root/
RUN echo "/usr/bin/ceilometer-collector" >> /root/
RUN echo "/usr/bin/ceilometer-api" >> /root/
RUN chmod 755 /root/

5. I added and committed the change and pushed.

[root@localhost ceilometer]# git commit -m "testing" .
[master fe12eff] testing
1 file changed, 6 insertions(+), 7 deletions(-)
[root@localhost keystone]# git push
Username for ''
Password for '':
Counting objects: 14, done.
Delta compression using up to 2 threads.
Compressing objects: 100% (3/3), done.
Writing objects: 100% (3/3), 335 bytes | 0 bytes/s, done.
Total 3 (delta 2), reused 0 (delta 0)
a3de5a2..fe12eff  master -> master

6. The automated build finished successfully.

7. I was again able to view the Dockerfile and the output of the docker build.


The Results

Voila! The automated build in Docker Hub triggers builds each time I push a change to GitHub and my images were now ready to be pulled down. Here is the pull…

[jameslabocki@localhost ~]$ docker pull jameslabocki/ceilometer
Pulling repository jameslabocki/ceilometer
c6f1a8880f25: Download complete
511136ea3c5a: Download complete
fd241224e9cf: Download complete
3f2fed40e4b0: Download complete
dba879135119: Download complete
18e0ba2cede2: Download complete
f95adbc11b72: Download complete
0816f2338632: Download complete
9d5ab1dabbc6: Download complete
b9699e746fd1: Download complete
f5089b261315: Download complete
16ddaee924ac: Download complete
aa364c587e7f: Download complete
c33d108a2ed1: Download complete
a131665ee3cb: Download complete
de2e2481c394: Download complete
220014c6f68a: Download complete
291a7a267101: Download complete
e2394bfd4a3f: Download complete
6f3c660adc44: Download complete
7e32f4d4b9b3: Download complete
48eb2ed1f711: Download complete
4bdd06d07972: Download complete
e3299cecd69d: Download complete
c6ccb763dee0: Download complete
2c2de8d95775: Download complete
5d6699390133: Download complete
c6b60ee39aa4: Download complete

[jameslabocki@localhost ~]$ docker pull jameslabocki/keystone
Pulling repository jameslabocki/keystone
2653e44d0420: Download complete
511136ea3c5a: Download complete
fd241224e9cf: Download complete
3f2fed40e4b0: Download complete
ba543dd23e14: Download complete
64f9a0c3a01e: Download complete
89475c70d6b2: Download complete
541ad4ae8739: Download complete
7497b01e4a38: Download complete
68eb2b0e770d: Download complete
a5b4e9a6299c: Download complete
69804e70b9be: Download complete
061825f3a29d: Download complete
f2789b8735b4: Download complete
f25b5709610e: Download complete
133b65e73ad2: Download complete
34152ccfe1a0: Download complete
334d431c2297: Download complete
b0c278d16459: Download complete
8b4eb357e27b: Download complete
85d002913148: Download complete
a0ff34bfbe61: Download complete
e2f611facb89: Download complete
92335c389d86: Download complete
0dd39fdaf260: Download complete

I can also run them and in relatively short order have keystone and ceilometer running side by side on the same host. These containers are relatively isolated, much smaller then virtual machines, and I don’t have to worry about my local machine getting foobar’d while working on keystone or ceilometer. Some great benefits to developers and (eventually) to ops teams.

[jameslabocki@localhost ~]$ sudo docker run -i -t jameslabocki/keystone /bin/bash

On the keystone container I can execute each of the steps in /root/ one by one to get keystone up and running and create the services and endpoints.

bash-4.2# /usr/bin/keystone-manage db_sync
bash-4.2# /usr/bin/keystone-all &
bash-4.2# /usr/bin/keystone-manage pki_setup --keystone-user root --keystone-group root
2014-07-10 02:12:47.284 413 WARNING keystone.cli [-] keystone-manage pki_setup is not recommended for production use.
Generating RSA private key, 2048 bit long modulus
e is 65537 (0x10001)
Generating RSA private key, 2048 bit long modulus
e is 65537 (0x10001)
Using configuration from /etc/keystone/ssl/certs/openssl.conf
Check that the request matches the signature
Signature ok
The Subject's Distinguished Name is as follows
countryName           :PRINTABLE:'US'
stateOrProvinceName   :ASN.1 12:'Unset'
localityName          :ASN.1 12:'Unset'
organizationName      :ASN.1 12:'Unset'
commonName            :ASN.1 12:''
Certificate is to be certified until Jul  7 02:12:47 2024 GMT (3650 days)

Write out database with 1 new entries
Data Base Updated

bash-4.2# /usr/bin/keystone –os_auth_url –os-token ADMIN –os-endpoint service-create –name=ceilometer –type=metering –description=”Ceilometer Service”
WARNING: Bypassing authentication using a token & endpoint (authentication credentials are being ignored).
|   Property  |              Value               |
| description |        Ceilometer Service        |
|   enabled   |               True               |
|      id     | 27b508729fd84be7994846873b6d7ab2 |
|     name    |            ceilometer            |
|     type    |             metering             |

bash-4.2# /usr/bin/keystone –os_auth_url –os-token ADMIN –os-endpoint service-create –name=keystone –type=identity –description=”OpenStack Identity”
WARNING: Bypassing authentication using a token & endpoint (authentication credentials are being ignored).

|   Property  |              Value               |
| description |        OpenStack Identity        |
|   enabled   |               True               |
|      id     | 8ccd55d9d6e04b5ca768a033e61db1a1 |
|     name    |             keystone             |
|     type    |             identity             |

Unfortunately, while I’m able to get tokens as a user that I created I’m not able to list users, so I am stuck. I think this might be a bug on the master branch and I’ll be digging into it.

Now, on to Ceilometer …

[jameslabocki@localhost ~]$ sudo docker run -i -t jameslabocki/ceilometer /bin/bash

bash-4.2# sed -ri 's/#identity_uri=<None>/identity_uri=' /etc/ceilometer/ceilometer.conf

bash-4.2# /bin/mongod --dbpath /data/db --fork --logpath /root/mongo.log --noprealloc --smallfiles
note: noprealloc may hurt performance in many applications
about to fork child process, waiting until server is ready for connections.
forked process: 14
all output going to: /root/mongo.log
child process started successfully, parent exiting

bash-4.2# /bin/mongo ceilometer /root/mongosetup.js
MongoDB shell version: 2.4.6
connecting to: ceilometer
"user" : "admin",
"readOnly" : true,
"pwd" : "46553e9fc2cdeada18e714cedbd05c9e",
"_id" : ObjectId("53bdff272da99d751819ff1d")

bash-4.2# /usr/bin/ceilometer-collector &
[1] 162

bash-4.2# /usr/bin/ceilometer-api &
[1] 192

I also had one more issue that I needed to work around since I was running from trunk. I had to patch a file to work around a relative pathing issue with the ceilometer API –

Since the keystone service was having issues I wasn’t able to run ceilometer meter-list or other commands (yet), but I do have the processes running in containers. I’ll continue to troubleshoot the keystone issue to see if I can tie these two services together.


A few thoughts came to mind while running through this exercise.

1. An area that would benefit from tooling is the ability to take an existing docker image and determine how it could be re-based on an existing parent image. For example, after I went through installing python, python-devel, mysql-devel, etc it would be nice if Docker Hub or another tool could tell me that I could save time on builds by using a parent image that already contained those components (no need to `RUN docker yum install` anything). This would save time during build processes. Call it deduplication for Docker!

2. If build times could be kept really short with such tooling it would be REALLY cool to attach an IDE to Docker Hub so that as you typed code into a project on GitHub you could instantly find out the build status. Of course syntax checking could solve some problems in a Dockerfile, but I am thinking along the lines of launching multiple docker builds and testing them with real data (system, UAT, or performance testing scenarios) and returning the result in near real-time. Building a truly integrated development experience into a continuous delivery pipeline could be really powerful (I’m imagining an IDE showing you that the line you just wrote caused a failure when run with 3-4 other docker image builds and launched on AWS, GCE, etc or that the performance was degraded).

3. Extending docker files to have pre-requisites on other docker images would allow users to reference other images required. For example, instead of installing MongoDB on the same docker image it would have been nice to be able to put some statements like this in the Dockerfile.

REFERENCE MONGODB=`docker pull mongo`
sed -ri ‘s/#connection=/connection=${MONGODB:IP_ADDR}/’

Perhaps this should live outside the Dockerfile in systemd, geard, heat, or some configuration language (puppet) and orchestration engine (Kubernetes). Whatever the case, once Docker Hub and other automated Docker build services have this functionality building images that depend on other services will be very powerful.

4. Some of the commands, such as running mongod and then adding a user during the docker hub build kept failing. I’m not sure if I am missing something, but it would seem that being able to run mongod during the build process to add users or seed data into the docker image is something that would be useful. Local docker builds also failed at this. Again, this might be something I am doing incorrectly.

One thing is certain in my mind, the future is bright for containerized IaaS services and sooner or later PaaS will drive the lifecycle of IaaS private cloud services and make the life of Ops much easier!

Here is a link to my docker hub builds for ceilometer and keystone if you want to look further.

Satellite 6

Satellite 6 Slides from the Red Hat Tech Exchange can be downloaded here


Get every new post delivered to your Inbox.

Join 1,004 other followers