Day 2 was much the same as day 1, with similar quality speakers and content (but a shorter day). Overall there was a great amount of material, and I'm very glad I made it along. I'm looking forward to checking out and making use of some of the preview tooling and features that'll be released in the next few months.
DevOps best practices for Azure and VSTS
Simon Lamb started off by stating that we shouldn't need dev-ops teams, that everyone should be dev-ops, and that Microsoft believes in services in any language, any platform, for all developers.
Microsoft defines dev-ops as the "union of people, process and products to enable continuous delivery of value to our end users", highlighting that it's measurable and deliverable. Great reasons to implement dev-ops are: competition is already doing it, increase velocity, reduce downtime, reduce human error through automation. Your options provided by MS are: TFS (on prem, upgrade yourself), or VSTS (cloud, automatically upgraded). VSTS also has release documentation with the features which have been added and updated, and you can switch on or off 'preview' features in your profile
The demo was a new MVC project (with tests) pushed to a git repo in VSTS. The 'continuous delivery tools for Visual Studio' extension creates a web app, build definition and release definition. Simon went through the Release templates and tasks (as well as failing the performance test if response time was 5sec+), and swapping slots. He ran the build, viewed build logs, highlighted that you can link the build/release back to code changes, view test results, and see release that was executed; you can also send a release summary email from release.
You can run a hosted build/release agent which is handled by MS, runs in a clean environment every time, and you can see the state (variables etc.) after each build. You can also run a private agent, which is on your own build machine. Deployment groups, which are a new addition, provide you with registration scripts (available for Windows or Linux) which will auto-download and install on your machine, and run phases in parallel on all machines in the group
Simon then demonstrated how easy it was to also work with Java and VSTS, through either Eclipse or IntelliJ. He ran through the usual process of viewing work item, commiting to feature branch (a policy was implemented which meant no committing to master); he went into VSTS to create a PR which he then approved to run the build and release (which was set to auto-run when master successfully built). He highlighted the policy which meant that a commit must correspond to a work item, must have a comment, and can't break build.
The final demo was a Node.js project using VS Code with with Azure/VSTS extension. You can create a web app straight from VS code. The continuous deployment Azure blade on the web app creates your build and release definitions in VSTS.
Drive Azure governance with Policy and Cost Management
This was another session by Alistair Speirs, who began by acknowledging that in the past costs were very complicated and painful to manage (bills, 3rd party tooling, APIs). These days Azure has integrated Cloudyn, which can be used over more than 1 subscription and makes life easier for those who have to manage lots of projects (and heaps of resources). It's not just for Azure, but "all 3 cloud providers" (the other 2 being AWS and Google), and you can have a different policy per platform if you want. Cloudyn also uses the RBAC model that Azure employs. Typically it's been extremely difficult to split costs for shared resources or costs like ExpressRoute, security, ingress / egress, but with Cloudyn it's very easy and customisable. Different teams also have a different dashboard. On-prem we typically over-provision to leave room for growth, but in the cloud it's all about cost-saving by not over-provisioning. Cloudyn has tools to help you optimise your resources, as well as reports, and you can schedule this to be before your budget.
He demonstrated the Cloudyn interface, which did look very simple to use: you open the dashboard from cost management in portal. The cost by service, cost by region, ability to create monthly reports were all demo'd.
Cost management is typically "someone else's problem", but Azure wants to prevent this facilitate things by bringing visibility, which brings accountability (and allows you to set budgets and set forecasts). You can make changes in the Portal and see resulting cost changes practically straight away. You can also set alerts, and now get notified of anomalies. Tagging is also a great way to aggregate your costs, so don't forget to tag.
Azure also now has Reserved Instances where you can pre-purchase 1 or 3 years worth of compute for a massive discount.
Alistair concluded with the promise that the policy and cost tooling will come together, and Azure Advisor will improve.
Azure Infrastructure and application monitoring
John Pritchard and Rebecca Lyons both presented this session, which I think was one of my favourites.
John kicked things off by outlining that monitoring means different things to different people; that monitoring in the cloud means getting an overall picture of all your resources, which can be spun up/down at will, and that some resources might not even exist by the time you're looking at the logs. You need: visibility (activity / metrics), insight (alerting / mapping) , optimisation (App Insights).
John demonstrated: service health, activity log (across subscription, summary, save queries), metrics (network in / out for scale sets), metrics preview (storage account API calls which had failed), edit alerts, alerts with actions (eg. deployment failed), log alert support request (a new feature), create real time metric + compound metric (CPU+network) into activity group (eg. SMS/email/service now/ITSM/custom). He also went into Log Analytics to demo custom searches, show multiple result sets with a line graph (comparing individual computers against logical group of computers, on-prem vs Azure).
Rebecca took the App Insights portion of the session, highlighting that bugs are hard to find and that's where Azure can help. She took us through an example site (Fabrikam Fibre), and demonstrated the Application Map in App Insights: this shows what’s talking to what, dependencies, overall health, availability, and recommendations, out of the box. There's also a new and improved performance dashboard, and Azure shows you stats 5 mins before and after an error to help you can reproduce any issue; you can then create a new work item with pre-populated details directly from the error in the Portal. She demo'd the Users Usage feature, which lets you group your users by page, geography, etc. and provides a very nice chart. The User Flows is another fantastic feature which shows the path of your users through the site through site, so you can see what's being used the most and focus your attention where it's needed.
New Azure platform capabilities
Katy Olmstead ran us through some of the features of the Azure portal, but I mainly focused on the newer functionality. Don't forget to go to https://preview.portal.azure.com/ to have all the preview functionality enabled!
One of the first things to note is that you can use Azure CLI or Powershell directly in the browser by clicking the icon in the top right (between alerts and settings). Unfortunately for us here in Australia it looks like this requires setting up a storage account in Southeast Asia, which is more than a little annoying.
There are also a bunch of keyboard shortcuts for power-users, which you can view by clicking the help icon (question mark in the top right), then 'keyboard shortcuts' link. In the same section you can show your new users a guided tour.
Don't forget search, all services
Up next was a demo of creating an app, and ARM template deployment. Azure Monitor lets you view all monitoring, real time alerting, and diagnose issues.
Use resource groups for your primary method of grouping resources (based on the lifecycle of the resources) but don't forget tags (key-value pairs) which can also be used for searching and billing. The columns in every view are also all customisable
Again, as with most other sessions, it was highlighted that Azure has the most comprehensive resiliency and best SLA including 99.9% single-instance.
Advisor is a great tool (which apparently just went public) for optimisation, offering personalised recommendations using machine learning, for cost or performance. It learns as you use Azure. You can see recommendations and act on things right now, or snooze (eg. for a dev environment). You can also download it all as CSV or PDF to go over as a team and prioritise for a later date.
Service Health is a more reactive tool for diagnosing symptoms ("is it a Microsoft issue, or is it me?"), understanding the impact, and getting notified. It can be integrated with web hooks, and provides a tracking id (to let you know that MS is aware of the issue). You can also always tweet @azuresupport who are apparently a very large and active team.
Planned maintenance lets you know about anything within 30 days that might impact you, and will shortly give you the ability to control the time window and pick the time of the maintenance of your resources (so you can schedule any down-time to be out of hours).
The Resource Health blade on most resources shows shows status, last changes, solutions to common problems (recommended steps), and allows you to troubleshoot issues; you can also report if you think the status listed is incorrect.
There are 4 levels of support: developer (low severity only), Azure Standard (for production, 24/7), Azure ProDirect (shorter responses), Microsoft Premier (enterprise-wide proactive support).
No comments:
Post a Comment