I've just switched to Roots Sage starter theme for Wordpress: roots.io/sage/docs/
and I'm currently reading up on deployment processes.
My processes is usually:
- make changes
- build with grunt/gulp
- commit (including compiled scripts)
- deploy
Sage's .gitignore file removes the dist folder (compiled files) from the repo ie. no css/js in repo. Are you supposed to install node/npm and build the assets on staging/production environment after deploy? If so, how do you trigger a gulp/grunt task on a remote server after deploy?
I'm using https://www.springloops.com/ for managing git and deploy.
Are you supposed to install node/npm and build the assets on staging/production environment after deploy?
You should avoid doing this. There is mixed opinion about committing compiled assets to a VCS as you stated you were doing previously, too.
Let's look at an example.
You finished all your testing locally. You haven't run an npm update in a few days and one of your dependencies has a loose version constraint specified; something like "~1.0.0".
You deploy. On the server, npm install is run before gulp or grunt. gulp runs, the build of your assets completes successfully, and the new version of your app is now live.
Unknown to you, version 1.0.1 of that dependency was released yesterday. For whatever reason, 1.0.1 introduced a change that breaks functionality within your app. That breaking change is now live on your site in production.
Even if you could guarantee all dependencies pulled from npm install on the server will mirror what you had locally/in staging, the headache of maintaining yet another set of software on the server (node.js, ruby, etc...) just for compiling assets should be enough to keep you from doing compilation on in production.
IMO, you should keep compiled assets out of your VCS, and rsync them to your server(s) as part of your deployment.
Related
I have an ubuntu staging server where I have installed apache, php, mysql, git, composer installed. I have a private git repository setup on the bitbucket, the project is already cloned to the staging server and to my local development machine. The Laravel setup is working perfectly fine on both machine.
What I am currently doing is whenever there is an update to the git repository, I do login to the staging server, pull the latest code from the git repository and do composer install, npm install, bower install.
I want to automate this process via capistrano tool. I checked the tutorials online, but all of them do the clone of repository whenever, I issue a deploy command and creates a fresh installation every time. Can't capistrano helps me to work on the existing folder that is already setup?
The basic premise of Capistrano is the idea that a new installation is created each time, such that there is not much to be done initially in terms of setup. If you'd rather use a different mechanism, a different tool would work better for you! For such cases, you could try to write a script using SSHKit directly (fairly advanced), or write a makefile or some other tool to automate your process.
If you do want to try to make Capistrano work on its terms, look into how linked_dirs and linked_files work in it. They allow you to have some files (e.g. config files, log dirs, etc) which are outside of the deployment directory and as such are shared between deploys.
We are building packages for multiple deployment environments using TeamCity server and OctoPack. The problem is that tentacle agent chooses the latest by number version of the package, so it's the same (latest) package that is deployed on all environments. Here's the summary of our setup:
Environments DEV and STAGE;
Deployment to DEV is triggered from Git "dev" branch;
Deployment to STAGE is triggered from Git "stage" branch;
OctoPack is configured to generate packages MyProduct.1.0.0.dev-%build_counter% for DEV build configuration;
OctoPack is configured to generated packages MyProduct.1.0.0.%build_counter% for STAGE build configuration;
TeamCity is configured to expose OctoPack artefacts (NuGet packages) via its NuGet feed;
Octopus project is configured to deploy packages with NuGet Id MyProduct from TeamCity NuGet feed.
So what happens is that since DEV builds are run more frequently, they have larger %build_counter%, and STAGE doesn't get a chance to get a deployment of its own packages - Octopus tentacle preferes packages with 1.0.0.dev-* suffix.
This must be fairly common scenario, but I haven't found a simple way to solve it.
There are some parts that are not documented here: https://github.com/OctopusDeploy/Octopus-Tools. But if you look at https://github.com/OctopusDeploy/Octopus-Tools/blob/master/source/OctopusTools/Commands/CreateReleaseCommand.cs it is possible to figure out what you can do.
I think the tools is backward compatible, but not 100 % sure about that.
When you are using the octo tools, which I expect that you use, you can set the version (also called releasenumber now) option to specify the release number. If you doesn't specify anything else it will take the latest package so what you want to do is set the packageversion (also called defaultpackageversion now) that should be used for the release.
I think that should do it. If it doesn't, what are you using to create the release?
Example of what we are using from our TeamCity when using octo tools which we have added to the environment path on the build agents:
create-release --server=%conf.OctoServerApi% --project=%conf.OctoProject% --version=%env.OctopusPackageVersion% --deployto=%conf.OctoDeployEnv% --packageversion=%env.OctoPackPackageVersion% --apiKey=%conf.OctoApiKey% --waitfordeployment %conf.OctoExtraParams%
UPDATE:
The documentation for 2.0 is much better: http://docs.octopusdeploy.com/pages/viewpage.action?pageId=360596
Inspired by Tomas Jansson's answer, simply adding the following to Additional command line arguments in the OctopusDeploy: Create release build step (TeamCity v9) worked for me:
--packageversion=%build.number%
I have an MVC4 + EF4.0 .NET 4.5 project (say, MyProject) I'm able to run the project locally just fine. When I FTP deploy it to Azure Websites (not cloud service) it runs fine too. However, if I do a GIT deploy, the site 'runs' for the most part until it does some EF5.0 database operations. I get an exception Unable to load the specified metadata resource.
Upon debugging I noticed that if I:
GIT deploy the entire MVC4 project (as before)
FTP in and then replace bin\MyProject.dll with the bin\MyProject.dll file that I just built locally (Windows 8 x64, VS2012, Oct'12 Azure tools) after the GIT push (i.e. same source)
then the Azure hosted website runs just fine (even the EF5.0 database functionality portion).
The locally built .dll is about 5KB larger than the Azure GIT publish built one and both are 'Release' mode. It's obvious that the project as built after the GIT push (inside Azure) is being built differently than as on my own PC. I checked the portal and it's set to .NET 4.5. I'm also GIT pushing the entire solution folder (with just one project) and not just small bits and pieces.
When I load the locally built as well as the remotely built MyProject.dll files, I noticed the following difference(FrameworkDisplayName)
local: System.Runtime.Versioning.TargetFrameworkAttribute(".NETFramework,Version=v4.5", FrameworkDisplayName = ".NET Framework 4.5"),
remote: System.Runtime.Versioning.TargetFrameworkAttribute(".NETFramework,Version=v4.5", FrameworkDisplayName = ""),
Anyone knows why this is happening and what the fix might be?
Yes, this is a bug that will be fixed in the next release. The good news is that it's possible to work around it today:
First, you need to use a custom deployment script, per this post.
Then you need to change the MSBuild command line in the custom script per this issue.
Credit goes to David above for the pointers and hints. I voted him up but I'll also post the exact solution to the issue here. I've edited my original post because I found there was a major bug that I didn't notice until I started from scratch (moved GIT servers). So here is the entire process, worked for me.
Download Node.JS (it's needed even for .NET projects because the GIT deploy tools use it)
Install the azure-cli tool (open regular command prompt => npm install azure-cli -g)
In the command prompt, cd to the root of your repository (cd \projects\MyRepoRoot)
In there, type azure site deploymentscript --aspWAP PathToMyProject\MyProject.csproj -s PathToMySolution.sln (obviously adjust the paths as needed)
This will create the .deployment and deploy.cmd files
Now edit the deploy.cmd file, find the line starting with %MSBUILD_PATH% (will be just one)
Insert the /t:Build parameter. For example:
[Before] %MSBUILD_PATH% <blah blah> /verbosity:m /t:pipelinePreDeployCopyAllFilesToOneFolder
[After] %MSBUILD_PATH% <blah blah> /verbosity:m /t:Build /t:pipelinePreDeployCopyAllFilesToOneFolder)
Push to GIT (check the GIT output if everything went ok)
Browse to your website and confirm it works!
I'll be glad when it's fixed in the next revision so we won't maintain the build script
We have a gem that contains shared code for multiple apps. It is hosted on a private github repo.
I want each app to automatically grab the latest version of that gem every time bundle install is run, so it is easy for the other members of my team to always be up to date, as well as having a simple deployment on Heroku and our Jenkins CI server.
It is my understanding that when bundle install is run, if some version of the gem has already been successfully installed, it will be used instead of any newer version.
Is there any way to force bundler to always get the latest version of the gem?
Do we just need to make bundle update a regular part of our workflow when we deploy or push to master (triggering a Jenkins run)?
As you said, I believe the update command seems like a better fit for what you are trying to achieve, since you can force the private gem update without affecting unrelated gems.
bundle update mygem
Per the bundle-update man page:
Update the gems specified ... ignoring
the previously installed gems specified in the Gemfile.lock.
In your dev environment you could create a bash or other script for running this in tandem with a standard bundle install.
As far as Heroku deploys, once you have updated and committed your Gemfile.lock changes to your git repo, Heroku should use that version, per their docs:
Gemfile.lock ensures that your deployed versions of gems on Heroku match the version installed locally on your development machine.
I am trying to setup a internal Mercurial HgWeb server on a Windows 2003 server. The Hgweb part is working. I could just share a folder to put released binary files for each projects. But I am wandering could I still somehow link the version control system with binary build output. So when there is a commit, the build output will automated get update as well for a release?
I know I could have a build system on the server end. But for Delphi, C#, ASP.NET projects and with a few third-party libraries, it seems much more work.
Right now, I am thinking about for each project I will have two repository, one for development (not output binary), the other for release which will including everything including the build result binaries (or only build result including dependency will be a better idea?). But I don't know yet how to make those two synchronize automatically without manually commit twice.
Maybe simply a hook on Dev repository fires every time commit to Master branch which will make another commit to the Release branch?
You really need a build system like CruiseControl.NET to build your binaries after pushes happen to a remote repository that CC.NET is watching. The binaries built can then just be copied to a standard Web server to be served up for download. CC.NET is not complicated to configure and supports Mercurial out-of-the-box. Using a system like this, you can get the extras like build stats, run unit tests before pushing a build to be downloaded, and lots more.