So far, I've encountered different scenarios of deployment, at least three types.
Full Build, Full Deploy. Like most of Java, .Net applications, have a Jenkins job builds the whole application and deploy the whole application.
No Build, Incremental Deploy. Like Mainframe, Informatica, Vitria, even database, these types of applications, they only deploy files got changed.
Full Build, Incremental Deploy. e.g. a not well structured Java application. The build generate total 7 jars, but only one of them is actually changed, and they want to deploy only this jar. Same for .net applications.
After google, I believe the 3rd case is not following best practice and should be resolved at the application architecture level. How to structure, partition, and build large MVC application for deployment in small incremental pieces?
The 1st case, is simple. We take everything from the source control system, build it and deploy it.
The tricky one is the 2nd. I have to generate a list of changed file in this build and deploy them. Any good experience on how to handle this well?
For incremental deployment, I use rsync. It's a program for synchronizing files, directories, permissions, etc. By default, it performs the task of figuring out which files have changed. When you have gigs of data and fifty thousand of files, re-deploying is fast and can be automated.
Related
My use case is composed of several ReactJs projects, where we collaborate using Git.
we are building a workflow to be used using Git, and that's our current thinking:
Each programmer works locally feching the next branch
They build their own branches but at the end its all merged to next
When all the pending tasks are done, we move create a branch test from next
Once test is fine, it is branched to beta
When stable, branched stable
This is the development phase.
For deployment, our first tought is to "build" the bundle on test, beta and stabe and copy it to the respective servers for running, as we keep built bundles on a normal filesystem (this is how we do it today, keep several bundles for several versions, not using Git)
Our production environment has dozen of servers in different customers, and every time we need to update, we need to copy the respective bundle from the correct directory to the server and install it (all bundles are build with an installation tool).
So, I have 2 doubts here:
a) Is the development workflow a good practice? Any suggestion?
b) How we make the deployment workflow smoother? Should we keep the bundles together in the code on Git ? Should we use something different?
Ideally we would need the server to autoupdate from our command. What would be the correct way to accomplish that ?
In our current dev. workflow there is main database --> DbMain. There is the process that takes the latest version of the project and automatically deploys it there and after that it triggers unit tests. As we would like to always have working version of the project in the source control each developer should be sure that he checks in the working code and all tests would be passed.
For this purpose we decided to create individual databases for each developers that has following naming convention --> DbMain_XX (where XX are the developers initial). So every developer before the check-in is suppose to publish all the changes to that database manually and run the unit tests. It is useful to setup publish config for this purpose with that is the copy of the main publish config with the only difference in the database names.
That would introduce that we will have a lot of different publish profiles in the solution that is quite a mess.
If we will not add these profiles to the source control, then .sqlproj file would still have reference to these files so the project will have reference to the not existing files.
So the actual question. Can I have single publish profile for all developers where the database name will be changed using variables? For example DbName_$(dev_initials)? Or can we have that each developer would have their own publish configs only locally and it wouldn't break the project?
UPDATE:
According to the Peter Schott comments:
I can create local publish profile, but if I don't add it to the source control, then the still be an entry in sqlproj file, but the file itself will be unavailable.
Running tests locally have at least 2 disadvantages. The first one is that everybody is supposed to install SQL Server locally. We are mainly working via virtual machines and the disk space is quite limited there. Another thing is that developers will definitely forget or not will not run tests manually every time. Sometimes they will push changes to the repo without building it or/and running tests. We would like to avoid such situations and "catch" failed build as soon as possible.
Another approach that was mentioned is to have 1 common build database. And in my case we have one (DbMain). All of developers can use it for it's needs but we will definitely catch the situation when the 2 developers will publish at the same time and that can make a lot of confusion by figuring out what's really went wrong.
A common approach to this kind of thing - not only for SSDT publish profiles but for config files in general - is to commit a generic version of the file with a name something like DbMain.publish.xml.template, and provide instructions to the developer to rename the file to DbMain.publish.xml - or whatever - and .gitignore this local copy of the file, allowing the developers to make whatever changes they want, but inherit the common settings from the .template version of the file.
Publish profiles don't need to be added to the .sqlproj to be used at deploy time, this is merely a convenience in Visual Studio to make them easier to find and edit, so you don't need to worry about broken references.
You are right in wanting to avoid multiple developers publishing to a common "build" database, this is a recipe for frustration.
Really, you want the "build" database to be published to as part of your CI process, meaning after the developers have pushed their changes.
I'm looking at implementing Team City and Octopus Deploy for CI and Deployment on demand. However, database deployment is going to be tricky as many are old .net applications with messy databases.
Redgate seems to have a nice plug-in for Team City, but the price will probably be stumbling block
What do you use? I'm happy to execute scripts, but it's the comparison aspect (i.e. what has changed) I'm struggling with.
We utilize a free tool called RoundhousE for handling database changes with our project, and it was rather easy to use it with Octopus Deploy.
We created a new project in our solution called DatabaseMigration, included the RoundhousE exe in the project, a folder where we keep the db change scripts for RoundhousE, and then took advantage of how Octopus can call powershell scripts before, during, and after deployment (PreDeploy.ps1, Deploy.ps1, and PostDeploy.ps1 respectively) and added a Deploy.ps1 to the project as well with the following in it:
$roundhouse_exe_path = ".\rh.exe"
$scripts_dir = ".\Databases\DatabaseName"
$roundhouse_output_dir = ".\output"
if ($OctopusParameters) {
$env = $OctopusParameters["RoundhousE.ENV"]
$db_server = $OctopusParameters["SqlServerInstance"]
$db_name = $OctopusParameters["DatabaseName"]
} else {
$env="LOCAL"
$db_server = ".\SqlExpress"
$db_name = "DatabaseName"
}
&$roundhouse_exe_path -s $db_server -d $db_name -f $scripts_dir --env $env --silent -o > $roundhouse_output_dir
In there you can see where we check for any octopus variables (parameters) that are passed in when Octopus runs the deploy script, otherwise we have some default values we use, and then we simply call the RoundhousE executable.
Then you just need to have that project as part of what gets packaged for Octopus, and then add a step in Octopus to deploy that package and it will execute that as part of each deployment.
We've looked at the RedGate solution and pretty much reached the same conclusion you have, unfortunately it's the cost that is putting us off that route.
The only things I can think of are to generate version controlled DB migration scripts based upon your existing database, and then execute these as part of your build process. If you're looking at .NET projects in future (that don't use a CMS), could potentially consider using entity framework code first migrations.
I remember looking into this a while back, and for me it seems that there's a whole lot of trust you'd have to get put into this sort of process, as auto-deploying to a Development or Testing server isn't so bad, as the data is probably replaceable... But the idea of auto-updating a UAT or Production server might send the willies up the backs of an Operations team, who might be responsible for the database, or at least restoring it if it wasn't quite right.
Having said that, I do think its the way to go, though, as its far too easy to be scared of database deployment scripts, and that's when things get forgotten or missed.
I seem to remember looking at using Red Gate's SQL Compare and SQL Data Compare tools, as (I think) there was a command-line way into it, which would work well with scripted deployment processes, like Team City, CruiseControl.Net, etc.
The risk and complexity comes in more when using relational databases. In a NoSQL database where everything is "document" I guess continuous deployment is not such a concern. Some objects will have the "old" data structure till they are updated via the newly released code. In this situation your code would need to be able to support different data structures potentially. Missing properties or those with a different type should probably be covered in a well written, defensively coded application anyway.
I can see the risk in running scripts against the production database, however the point of CI and Continuous Delivery is that these scripts will be run and tested in other environments first to iron out any "gotchas" :-)
This doesn't reduce the amount of finger crossing and wincing when you actually push the button to deploy though!
Having database deploy automation is a real challenge especially when trying to perform the build once deploy many approach as being done to native application code.
In the build once deploy many, you compile the code and creates binaries and then copy them within the environments. From the database point of view, is the equivalent to generate the scripts once and execute them in all environments. This approach doesn't handle merges from different branches, out-of-process changes (critical fix in production) etc…
What I know works for database deployment automation (disclaimer - I'm working at DBmaestro) as I hear this from my customers is using the build and deploy on demand approach. With this method you build the database delta script as part of the deploy (execute) process. Using base-line aware analysis the solution knows if to generate the deploy script for the change or protect the target and not revert it or pause and allow you to merge changes and resolve the conflict.
Consider a simple solution we have tried successfully at this thread - How to continuously delivery SQL-based app?
Disclaimer - I work at CloudMunch
We using Octopus Deploy and database projects in visual studio solution.
Build agent creates a nuget packages using octopack with a dacpac file and publish profiles inside and pushes it onto NuGet server.
Then release process utilizes the SqlPackage.exe utility to generate the update script for the release environment and adds it as an artifact to the release.
Previously created script executed in the next step with SQLCMD.exe utility.
This separation of create and execute steps gives us a possibility to have a manual step in between, so that someone verifies before the script is executed on Live environment, not to mention, that script saved as an artifact in the release can always be referred to, at any later point.
Would there be a demand I would provide more details and step scripts.
Twist to the standard “SQL database change workflow best practices”
Background
ASP.NET/C# Web App
MS SQL
Environments
Production
UAT
Test
Dev
We create patch scripts (XML and sql) that are source controlled in Mercurial. We have cmd line utility that installs patches to DB (utitlity.exe install –patch) from a Release folder the build packages. Patches have meta data that helps with when patch should run and we log patches installed in a table in the target DB. All these were covered in the 3 year old question:
SQL Server database change workflow best practices
Our Problem/Twist
I think this works well for tables, views, functions and stored procedures. We struggle with application configuration data. Here are some touch points on application configurations.
New client. BA performs system study and fit analysis. Out of this comes a configuration word document of what application configurations need to be setup. Note some of these may also come in phases over time. We need to get these new configurations into the system for the developer and client UAT.
Developer works on feature request or bug fix. A new configuration change comes out of that change. The configuration needs to make it into the system for testing and promotion to UAT and up.
QA finds that the developer missed an associated configuration change. That configuration needs to make it into the system for promotion to UAT and up.
Build goes to UAT. Client performs acceptance testing but find they really want to change another unassociated configuration and have it promoted with the changes. In other words they found they want to change a business process by a configuration. The configuration needs to make it into the system for promotion to PRD.
As the client operates in PRD they may tweak application settings. These configurations need to make it into the system for future development and testing.
The general issue is making sure we are accounting for all the configurations and accidently not miss any during promotions which causes grief.
Our Attempts At A Process
a. We have had member of the QA team to write patches (xml and sql) and check those in. This requires a build to make sure those get into the package. With this approach it really just took care of item 1 above and we fell apart on the other items. The nice thing is for the items that made it into the patches it was just an install with the utility.
b. A developer threw together a Config page on the application. All the configurations could be uploaded and downloaded via XML document but it requires the app to be running. For item 1, member of QA team would manually setup configurations in the application and then would download the Config.xml file. This XML file would be used to upload configurations in other environments. We would use text diff tool to look at differences between config.xml files from different environments. This addressed item 1 and the others items but had problems. Problems were not all configurations made it into the XML document (just needs to be fixed by developer), some of the configurations didn’t have a UI in the application so you still had to manually go to the database on some, comparing the XML document with text diff was difficult at time (looked mostly due to sorting but I’m sure there are other issues), XML was not very human readable and finally the XML document did not allow for deleting existing incorrect or outdated configs.
c. Recently we went with option B, but over time for a new client we just started manually tracking configs and promoting them manually by hand (UI and DB) through the promotions. Needless to say lots of human errors.
So we have been looking at solutions. Eventually it would be great to get as much automation in as possible. I’m looking at going with the scripting approach and just focusing on process, documentation and looking at using Redgate data compare in addition to what we had been doing with compare on config.xml. With Redgate we have to create views though and there is no way to create update scripts from that approach except to manually update the scripts. It does at least allow a comparison without the app running. I’m also looking at pulling out the configs from our normal patches and making it a system independent of the build (utility.exe –patch –config). When I say focus on process it will be things like if we compare and find a config change either reported by client or not, we still script it, just means we have to have a process in place to quickly revalidate config install before promoting to the next level. As for documentation looking at making the original QA document a living document instead of just an upfront document. The goal is to try and enhance clarity and reduce missing configurations during promotion. Unfortunately it doesn’t improve speed of delivery.
Does anyone have any recommendations or best practices to pass along. Thanks.
Can I ask exactly what you mean by application configuration. I'm interpreting that as both:
Config files in the web application
Static reference data inside the database
Full disclosure I work for Red Gate. You might be interested in taking a look at Deployment Manager, it's a deployment tool that deploys applications, databases and configuration. It's free for up to 5 projects and target servers.
The approach it uses is to package application code and the database state into packages. These packages can be deployed into dev, test, staging and production environments. The same package is deployed to each environment.
Any application configuration that needs to change between environments is handled in one of the ways below:
Variable substitution in web.config. The tool allows you to specify override values for variables in these files, and set these per environment/server
Substituting the web.config file per environment.
Custom powershell scripts that are run pre/post deploy. You could use these to execute custom SQL based on the environment or server.
Static data within the database, using SQL Source Control's static
data feature. I've written a blog post about how to supply
different sets of static data to different environments/customers.
This allows you to source control the application configurations and deploy them to different environments.
Our team develops distributed winform apps. We use ClickOnce for deployment and are very pleased with it.
However, we've found the pain point with ClickOnce is in creating the deployments. We have the standard dev/test/production environments and need to be able to create deployments for each of these that install and update separate from one another. Also, we want control over what assemblies get deployed. Just because an assembly was compiled doesn't mean we want it deployed.
The obvious first choice for creating deployments is Visual Studio. However, VS really doesn't address the issues stated. The next in line is the SDK tool, Mage. Mage works OK but creating deployments is rather tedious and we don't want every developer having our code signing certificate and password.
What we ended up doing was rolling our own deployment app that uses the command line version of Mage to create the ClickOnce manifest files.
I'm satisfied with our current solution but is seems like there would be an industry-wide, accepted approach to this problem. Is there?
I would look at using msbuild. It has built in tasks for handling clickonce deployments. I included some references which will help you get started, if you want to go down this path. It is what I use and I have found it to fit my needs. With a good build process using msbuild, you should be able to accomplish squashing the pains you have felt.
Here is detailed post on how ClickOnce manifest generation works with MsBuild.
I've used nAnt to run the overall build strategy, but pass parameters into MSBuild to compile and create the deployment package.
Basically, nAnt calls into MSBuild for each environment you need to deploy to, and generates a separate deployment output for each. You end up with a folder and all ClickOnce files you need for every environment, which you can just copy out to the server.
This is how we handled multiple production environments as well -- we had separate instances of our application for the US, Canada, and Europe, so each build would end up creating nine deployments, three each for dev, qa, and prod.