What kind of tools and strategies do you use for your apps' deployment Do you do it manually via Git? Do you use a PaaS like Dokku? Or maybe just create a custom bash script that does the work?
There are so many options out there and I'm curious what do you guys use in your company or private projects.
I'd love to hear your thoughts on this topic! :)
Cover image by Taylor Vick.
Top comments (35)
I just build everything into docker images, then setup docker-compose yaml file for your stack and just upload or git clone the yaml file on the server, and run the docker compose up command. If it works on your development machine, then it will works on production too. That's how docker was born.
For project that need to be scalable, You may take a look at kubernetes.
Hey Jordan! Thanks for your reply!
I'm aware of docker and it's a really easy and cool solution for deployment. Unfortunately, I often work on shared servers that do not offer virtualization (docker is not supported) and I need to handle it somehow different.
But for docker-compatible servers - I think it's one of the best solutions out there.
Btw. do you use Kubernetes in your private projects or is it just too much for the task that's need to be done?
Shared server? You mean CPanel? As long as you have root access to the server and the server is using Linux, suppose you are able to setup docker.
I didn't use Kubernetes for my private projects. Because I am still not confident enough on my Kubernetes knowledge. But I ever use Docker Swarm which is simpler and lesser features version of Kubernetes. Honestly it is steep learning curve to learn Kubernetes. 😅 Therefore, if you are interested and feel difficult on learning Kubernetes, u may start from docker swarm first.
Yeah, something like a CPanel. Even if you have root access there, those hosts are often based on a virtualization technology which does not support Docker.
Thanks for suggestions!
Okay, y'all are saying a whole lot that's seeming alien but here goes nothing, just in case my hunch is right.
Github, Netlify.
Case study: cirphrank.com
Processes:
CREATION>
XAMP's active, VSCode open. I create my app on VSCode, check on the go till I'm satisfied via chrome, localhost powered by XAMP.
DEPLOYMENT>
To cut costs, save for clients' and main business' projects, I use Netlify now (so long site's supposed to be static), used to use blogger but that had extreme limitations(optimum for blogs though).
I use Github for Desktop to upload my app to a private repository on my Github account, then I proceed to deploy via Github on Netlify.
-----At this point, I have cirphrank.netlify.com but to go custom I sync DNS and CNAMES for free SSL coverage...
Too much talk already and I'm not even sure if it's under the right topic. Çiao
It is! Thank you for this extended description! 😃 Your flow is interesting!
🤗My pleasure. Thank you too for the platform thread to express.
My team uses Jenkins, in future we might adopt CodeStar from AWS. Unfortunately, using 3rd party tools like GitLab in the cloud is not an option for security reasons. Perhaps it's possible, but getting this is way above my pay grade. On the other hand Jenkins gives you most control you can have. All jobs are pipelines in the code, stored in git.
Thanks for your reply!
Let me add one more question: you use Jenkins as a deployment tool, but how do you actually deploy your code? How do the scripts look like? Do you just copy your code to the remote server, or git pull on the target server and restart the app? How does it look like?
With Java - the deployment flow is two-part - one is to build the app and it's assets, and execute unit and integration tests, and put artifact into artifactory. Second part is about pushing stuff to one or multiple tomcats. Each time there is a loadbalancer that has to be disconnected, artifact to be deployed, sanity checks (both automated and manual), then loadbalancer reconnects, then we proceed to next server.
For some - flow is ansible based - put service down with systemd, upload new artifact, then bring it up
We work on having aws lambda, im currently finding the best approach, there is a package to be built, deploy as staging, run checks, promote to production build, and discard older versions.
There is also a package of static files that is deployed by just "git checkout".
Thank you! :)
In my pet project I do semi-automated deploy. When I merge commits into the prod branch I push it to GitHub. TravisCI performs all kind of checks and invokes the deployment script that executes Ansible that copies WAR file to a server and restarts the service. No docker so far :)
Nice! How was your experience with Ansible? Is it easy to use and "get into", or does it have a difficult learning curve?
It was easy to me when I found it (around 5 years ago). But I had experience with puppet at that moment so maybe because of that I had the smooth experience.
At my first job, we used Jenkins to build Android apks, someone downloaded the signed build, and uploaded it to the Play Store. Once tools like Testflight and HockeyApp came out, we had a Jenkins pipeline to manage distribution.
My next job was in a web services context. We had an infrastructure on AWS with hosts managed by Chef. "Deploying" meant that an ops team would need to build out a cookbook for your service which had a lead time of 3-5 months. Then once the infrastructure was set up, the ops team would need to update a manifest.yml file which had the master list of which build/version would be released on which hosts. The file would deploy about 15 minutes after that.
This could only be done in prod after 10pm since it incurred downtime, could only be done by the ops team since it was a fragile process. no one had enough data to figure out whether we had any users around at that time. Some pieces of it were managed by Jenkins, but it was what I would call a manual automated process - a bunch of manual steps that were just as bad as having to deploy manually. ... But the repo for this process was named "devops" so it must have been a devops process, amirite?
My job after that was to build an in house tool to make push button deploys for legacy services and new apps. It is basically a REST API with a front-end around a few deploy.sh scripts - one for backend services that deployed chef cookbooks, another for deploying to S3, and most recently, one for deploying containers to AWS Fargate.
That tool took care of authentication/authorization and integrated with JIRA to manage approval processes.
It also provided a few dashboards for reviewing all deployments for a particular service, what the current version of that service is in each environment - available to read for everyone in the company. Even POs could deploy... from a bar ...with proper approvals.
Nowadays, we use Jenkins to build and test artifacts, upload them to artifactory, deploy through the deployment tool (Jenkins does it for test environments, a human does it in prod). The tool just does an S3 sync to a bucket behind Cloudfront.
Wow, this is an extensive description, thanks! 😄
git push;)Do you mean solutions like Heroku or doing stuff with git hooks? 😄
I've been using Gandi for my single-dev apps lately, but I suppose Heroku could work as well :-) So, yeah, Option #1.
For my personal projects I've just discovered Netlify! I highly recommend a look into it. I use that for continuous deployment from GitHub. But like you said, there are SO many!! ☺
Hey Lorna! Thank you for your reply 😃 I've tried Netlify and it's amazing! Unfortunately, it works only with static websites, and sometimes I need to do some more complex server-side solutions. Netlify is my go-to with static sites, but I'm still looking for something I'll fall in love with when working on server-side solutions 😄
I use Zeit Now for my personal projects and my own pipeline with github webhooks and NodeJS for work (check out my latest post)
I've just discovered Zeit Now myself. How does it work for you?
As advertised I guess?
I didn't have time to try out Zeit Now for now, but it looks promising! I'll definitely take a look at your post, thank you for your reply! 😄
We're using BitBucket pipelines internally, which deploys all our serverless endpoints to AWS and our React apps to S3.
I still didn't have time to take a look at serverless backend. Does it scale well? Do you have a lot of endpoints or rather a few of them?
It scales infinitely, that's the cool thing about it (of course there are bottlenecks like DBs, so it's rather a theoretical infinity). Actually, we currently have 3 endpoints, and each of them exposes a GraphQL API.
Started with basic Bash scripts, now I use GitLab CI/CD + DCR or GitLab CI/CD + rsync to my EC2 instances (mostly for my throwaway POCs).
PS. I have a self-hosted GitLab instance.
I'm sorry but... what stands behind
DCR? 😄That's a cool setup! Do you host GitLab on your own server? Are the costs high for something like this?
I deploy those images to a docker engine running on an EC2 or directly to ECS.
I was using DCR way before ECR hence didn't migrate.
Yep, I do host GitLab on my own server and the costs are not high, but not cheap either.
It gets the job done without me having to worry(much) about security of my codebase