Let's face it. .env
files are amazing. They have an easy-to-read syntax that stores all of our essential configurations into one file. Not only that, they keep our deepest, darkest secrets as web developers. They make sure that our precious API keys and database passwords are kept locally, away from prying eyes. Having such a critical role in our code bases, we are constantly reminded by the community to never share our .env
files; to treat them like how the government treats their confidential information.
This is especially enforced in the open-source community where everyone shares, copies, and reuses code with each other. Accidentally committing and pushing the .env
file is considered by many as a relatable moment. Personally, I have never done it myself yet, but I'm sure my fate is sealed at this point.
As fun as it is to talk about "that one time you committed the goods" in a casual conversation between developers in a party, it is pretty alarming that it has become a common conversation—perhaps even a rite of passage—in web development.
A stain in the commit history
Running a quick search on GitHub reveals that there are still a number of people who didn't get the memo. The occasional add .env
and remove .env
commit titles appear every now and then in the search results. Looking into the content of the commits indeed shows their precious API keys and database passwords. It's honestly funny to see how they revert their changes like how a child becomes guilty of doing something they shouldn't have.
What's more alarming about this is that there are still some others who have not reverted their commits. The .env
file is still alive and well in their repositories. For all we know, these might be real, actual API keys and database passwords they currently depend on in a regular basis. To make matters worse, sorting the search results by recently committed shows how common and frequent these commits are.
The problem with simply removing the .env
file in the working tree is the fact that Git keeps a record of all the commits made in a repository, even the earliest ones. Unless clever tricks have been made (more on that later), committing that .env
file will forever be a stain in the commit history. This is just what a version control system is supposed to do after all: keep a history of changes, even the bad ones.
Having said that, how does one handle sensitive data in a repository?
.gitignore
is your best friend
Adding a .gitignore
file to your repository is your first line of defense against these hiccups. Properly and explicitly specifying what files and directories to ignore is a surefire way of preventing sensitive data from ever reaching public repositories and prying eyes.
GitHub and gitignore.io provide general-purpose .gitignore
templates for specific languages and environments. Most of the time, these templates are more than enough to suit your needs.
In the case of already having committed sensitive data in a repository, GitHub has a handy, and rather overkill, guide on how to purge a file from the commit history. "Overkill" is not a bad thing, though. One can never be too safe when it comes to security. 😉
We all make mistakes
We're all humans here. Nobody is perfect. Mistakes are just a part of life. In fact, it is probably one of the most important realities we have to face everyday. Without facing these mistakes, we can never become better developers. A huge part of learning isn't in the affirmation of success but in the recognition of the "whats" and "whys" of failure.
Sure, committing the .env
file is a grave mistake. Sure, it will pose serious security risks in your app. Sure, the business will suffer significant losses. Sure, it will be a hassle to clean up the commit history. Sure, it will be hard to sleep at night knowing that you have been compromised. But if there's one good thing that comes out of this experience, it has to be the fact that you will come out as a better, more experienced developer in the end.
With that said, here's a toast to all the future projects that you will never commit the .env
file to. Cheers! 🥂 *clink* I shall conclude this article with a parting thought and a final reminder to you, the reader:
Please don't commit .env
. You wouldn't want to end up at the top of GitHub's search results.
Latest comments (76)
When users access the front-end project, they can naturally capture all the content, so how can the front-end project prevent API key exposure? Is there no way?
The most effective way to combat this is to just never embed API keys in the front end. Instead, we use API calls (e.g.,
fetch
) so that some back-end server somewhere performs the API request on behalf of the front end. The API keys thus remain secret in the back end. This is an unfortunately cumbersome but necessary measure.If so, access the API encryption interface through the fetch reappropriate back-end interface. So the API interface encryption on and from the front and back ends is meaningless.
What do you mean by encryption, by the way? Are you referring to API keys that are embedded inside JWTs? Or are you referring to the general pattern of encrypting API secrets before sending to the front end?
general pattern of encrypting API secrets before sending them to the front end
I suppose that is one possible way. Personally, though, I would still prefer just having a centralized server that acts as a proxy for privileged API calls.
Just to clarity all this text above, which is misleading, the issue is not to commit the
.env
file!The issue is storing sensitive data in plain text -be it
.env
or any other kind of file- inside the repository, either be public (as an obvious security measure) or private (as a double security measure).It's perfectly fine to have a
.env
that stores your DB user, password and model if that's meant to point to a local database in your computer (Docker-ized or not) for development purposes or an API key that's meant to work just in local environment within your app cluster. You can then override these values with the production -or any other public environment- ones in the pipeline with the sensitive data values stored in a vault or repository variables.Best regards
I totally agree with this. I should have been clearer in emphasizing this point. Of course, hindsight is 20-20 five years later (when this article was first published). 😅
This is actually my current stance on the subject nowadays. However, strictly out of abundance of caution, I still avoid committing an
.env
file into any of my repositories. It's more about the fact that I know I will be working with other people, so I just outright.gitignore
potentially sensitive files from the get-go. It is not to say that committing.env
is inherently evil, but proactive measures are still better than reactive measures when it comes to computer security.true! 😁
Got a question probably a dumb one :D. Let's say the configs in our .env are just for a client side application like react and it includes api keys and such (no database or backend user pass stuff). Why would we care about committing since after the build all the keys are going to be somewhere inside the built file as well? and its visible through the browser when user is working with the website?
First and foremost, that's not a dumb question! 😂
You are correct that it shouldn't really matter if the environment variables are truly meant to be deployed with the file bundles.
However, the real issue here is the fact that sensitive keys are publicized in the first place. Needless to say, this is not exactly a secure deployment strategy. Even if the bundles are minified and such, this is a potential attack surface nonetheless.
API keys must be stored and indirectly served via some in-house server-side API, never directly through the client-side code. Potential security risks include (but are not limited to) denial-of-service attacks, impersonation, and backdoor access to app internals. That's no fun!
Though, if the environment variables only include non-sensitive static build configurations and such (i.e. theming options, CSS variables, etc.), perhaps it may be alright to publicize them. Otherwise, you should be extra wary about this deployment strategy—if not reconsider it altogether.
I need the
.env
file during build on CI how do I make sure GCP receives my .env file if I have not committed it on github ?/settings/secrets
.env
file/settings/secrets
) same in local setup ?To be honest, I am not exactly versed in the Google Cloud Platform. However, I do know they provide a (paid) Secrets Manager service that does exactly what you need in a secure manner. Otherwise, there doesn't seem to be a "free" workaround.
But that doesn't mean other cloud platforms have this limitation. Heroku, for example, allows you to directly set environment variables in the app itself.
So at the moment, I can't offer you any solutions with GCP. I'm unfortunately not familiar with it.
cool. thanks
No problem, Rajesh! Thanks for reading the article! 😉
Yup, pretty much! The best part about adding a sample file is the fact that it can also serve as documentation for environment variables. One can simply annotate and comment on the sample file.
Hm, I would recommend adding an
.env.sample
file in the repository. Instead of filling in the fields with real values, you can add placeholders, redactions, and "pseudo-values".Then, of course, this would have to be documented and explained in the appropriate
README.md
.Or perhaps, you could even forgo the
.env.sample
file and declare the environment variables in theREADME.md
itself, but I personally prefer the former because it is more explicit.And you can also use
dotenv-linter
to lint your.env
files for consistency and best practices: github.com/wemake-services/dotenv-...Here's a relevant article.
This is really intricate and interesting. Thanks for sharing! It was a great read (yet very disturbing 😬 from a security standpoint).
As if it was fault of Mailgun... It's your friends fault and yours for giving him these credentials.
Take a look at a package I wrote on NPM: (envup). It allows you to version control the structure of your environment as a separate file making it easy for others to setup their own .env file without commiting any of the data to git.
Why have it in your git directory at all? Most frameworks let you override the config location so set it to
~/.app-name/.env
or similar. If you need to provide an example dummy config you can check in an.env.sample
file.Yes, that's true. At least for me, I think it's just easier to have a
.env
file because it requires minimal setup and messing around with global configurations.At least in Node.js, all you have to do is
npm install
andrequire
the dotenv package. Then in your code, just invoke thedotenv.config()
and it should all be running smoothly via the process.env object. This way just saves you from the little extra effort you have to do with the nitty-gritty configurations.But to each its own. Whatever workflow works the best for you, you should apply it, not because everyone does it but because you feel productive with it.
.env
is an anti-pattern to me, because it requires overhead to keep it secure.It seems to be quite a popular anti-pattern nowadays. 😅
Besides, adding one line to the
.gitignore
file shouldn't be that much of an overhead.Oh, and this doesn't just apply to auth and application keys from cloud providers. It can also apply to the credentials you use for demo and testing servers for your client-side app, and database credentials for server-side apps.
I noted this a few years ago, before ".env" became the norm, but the effect was the same. People were putting their OAuth keys into configuration files, committing them to git, and a bot search came across several thousand exposed keys just looking for Amazon's. Similar numbers probably would have been found with google and others.
In that case, I wanted to make sure that my users knew where to put their stuff, so I created a credentials.template file that showed the format, and that got committed, but my own credentials did not. One could do the same here by having a README.env.txt file to document what to do, and cat that file to the console in an npm post-install hook.
The negative of that, though necessary, is it means you're not distributing running code. They can't just pull your files down and npm start and everything works. They have to finish the init by creating their own files. It may also complicate automated testing systems that would have to be configured to provide that file before running.
If you have made this mistake already, one possible way to fix it is to interactive rebase back to the sha that introduced the problem, wipe the file and add the .gitignore line there, and then deal with the merge conflicts as it pushes the rest up if you ever had to touch that file again (either in format or in updating the data in it).
Of course, how much work that is depends on the age of your code (how many commits and how many branches).