Table Of Contents
- Hello Reader
- The problem with .env files
- Config servers to the rescue
- More Pros of a Config Server
- About the Author
Hello Reader 👋🏽
I've worked on huge enterprise systems and I'm disappointed to see 99% of all JavaScript tutorials tell you to use a .env for production config and secrets which is dumb.
It's time to correct this behavior plaguing the community. Let's explore the problem with .env files and the ultimate solution.
The problem with .env files 🚫
1. Storing the .env file
The problem is well... It's a file. You are not supposed to store your "production" file in your repo; where do you put it? Directly in the VM root? What about Docker containers, do you bake your secrets directly in the image? If the image leaks, everyone has access to your secrets.
2. Updating a secret config
What happens if you need to update a database password? Whoever updates the .env will have access to every secret in the file. I don't even have to explain why this is bad. Every time a config has to be updated whoever is updating it can see EVERY secret.
3. Versioning
Let's say you are deploying a new feature that requires updating your config / secret variables, something goes wrong and the application gets rolled back.
Uh-oh does anyone remember the last values we had in the .env? We have no history and the application requires the values we just deleted/overwrote.
Config servers to the rescue 🦸♂️
The perfect solution to all these issues is using a config server. A config server is an externalized application for storing configs and secrets. It's considered the central hub for managing secrets across environments.
Multiple cloud services can function as a config server like AWS Parameter Store, Google Secrets Manager, or HashiCorp Vault for my open-source enthusiasts.
Once you create a secret or config, most of the time you are given a URL like https://app.com/config/DB_PASSWORD/v1.
Let's see how a config server solves all the issues with .env files.
Solving: 1. Storing the .env file
With a config server, all the secrets are centralized in one place, encrypted, and can only be accessed by applications and users you approve.
Solving: 2. Updating a secret
With a config server, instead of updating a secret, you create a new version. After creating a new version, you update your secret URL like .../config/DB_PASSWORD/v2 or .../config/DB_PASSWORD/v3.
Solving: 3. Versioning
Like with solution 2, most config servers have automatic versioning. If you need to update a secret, create a new version and update the config URL in the application. If the application gets rolled back, it will automatically use the last config URL you had in place.
This way secrets are protected and your applications can get rolled back with no issues with the previous secrets.
More Pros of a Config Server ✅
Automatic Secret Rotation
Secrets can become stale which can lead to your systems becoming compromised! A centralized config like AWS provides automatic Secret Rotation.
If you team or org all pull the same config, they will automatically get updated with the latest values. No need to send it over slack, no need to get a "trusted" dev, or having the possibility of forgetting to update one of your .env files. Easy and painless!
VPC Endpoints
A config server is usually hosted on a VPC which provides a localhost like connection for your services. This allows the ability to pull secrets without them ever having to touch the internet (virtually no latency), nor allowing outsiders to query the server.
Audit Logs
Did a secret get leaked? Config servers usually carry audit logs so you can check when or where a secret was accessed.
Does an updated config cause an error? Check when it was last updated and if needed, preform a global update for all your services. No plain text file editing needed.
Final Thoughts 💭
I hope now developers will start using centralized configs for their production services. .env files are unreliable and have no access management, versioning, or safe updates.
I'm not saying .env file don't have a purpose. They can be used for local or development-oriented environments.
What I'm trying to say is don't store valuable secrets in simple files, especially if my data is in your service.
About the Author 👨🏾💻
I'm Gregory Gaines, a simple software engineer @Google who's trying to write good articles. If you want more content, follow me on Twitter at @GregoryAGaines.
Now go create something great! If you have any questions, hit me up on Twitter (@GregoryAGaines); we can talk about it.
Thanks for reading!
Top comments (290)
I get it...you're like overly cautious.
After years and years of dealing with hundres of sites/apps/applications I can tell you one thing.
NO ONE gets hacked via their .env file. Ever.
First of all, it's really easy to lock it down so no one can access it except for the devs who need access. At some point there's at least one human being [preferably 2 in case one gets hit by a bus] who should know where and how to access and change all your passwords. Therefore having them in one place is just as safe as having them in 20 different places where you will absolutely 100% guaranteed forget some the next time you need to update things or something is on fire.
Overly cautious security leads to a mountain of steps to get to sensitive info that just hurts development and recovery times more than it helps to actually protect anything.
In other words...it's a pain in the buttocks. Don't do it.
Use .env files and learn the few steps it takes to protect them. There is NOTHING wrong with that.
Sounds like more work in my opinion. With a config server, there is no need to trust the "2 devs" since the server has permissions and access restrictions so only permitted devs can add secrets without viewing existing ones.
I don’t think it’s overly cautious, it’s being modern. There is no mountain of steps. Whitelist your application and use the config url like the examples in the article, easy and no pain in the buttocks.
Please re-read the article, I explained how easy recovery (roll backs) and development time is simplified using a config server. I COULD use a .env file and take steps to secure it, or I could live in 2022 and use a config server with permissions, auth, versioning, real-time updates, and an audit trail for extra security.
Personally I don’t want my secrets being stored in a glorified text file.
Thanks for the comment!
So you store them at an external company? What if they are down, or have a data breach? (They are a bigger target than most of the websites). I'd like control over my secrets and not be dependent on third parties for such important stuff, and storing them in a
.envis perfectly fine with the precautions and correct configuration mentioned by @ravavyrAlso, those services are not free. True, you can self-host HashiCorp Vault, but that costs also money.
True, nothing is full proof. But that's why we choose techniques or companies that gives us the best guarantees.
Also no matter how secure you keep that
.envfile, you are giving someone FULL view of all secrets whenever its updated.The services I listed have free tiers. HashiCorp can be self hosted for free forever, Google Cloud has a free tier for 6 secrets, AWS has a generous free tier for storing configs in parameter store.
If you don't want to go with an external company and want full control over your secrets, thats fine. There are free, open-source, or self-hosted alternatives. There's Spring Cloud Config, Github Secrets (if you trust the company enough), or roll your own.
I think the point that @tristan is making rather well is that your recommendation is just increase the surface area of your vulnerable systems.
The mindset that doesn't even trust employees with private information would deem providing that information to a third party (especially a faceless tech company) to be less secure than, keeping it contained within a closed system that is completely under their control.
The arguments you're offering for using a third party to host private information are actually the arguments against using such services. They are the weak points: If a system is compromised, it's almost certainly due to leaky abstractions like this...
It's an increase in surface area, but with the benefits I listed above.
Its not distrusting employees, it's being safe and following a secure practice. I don't know about faceless companies, I mentioned Google, AWS (which is the leading in the market for cloud) and other credible companies which provide a config service. I may be a little bias since I am a Google employee.
And if thats the case, why trust anything? Why host private code on Github or deploy on Digital Ocean? Its for the ease and guarantee.
I also mentioned open source config servers that you can host or deploy yourself.
If a system is compromised, most likely its from using bad practices...
DigitalOcean with firewall configurations is the problem solver. Only SSH works on my IP and only Cloudflare can access the HTTP directly. My name literally says "trust me".
What happens if your service expands and you have to share common secrets across different teams?
Do you just copy and past the file or use a centralized config?
I could very well say let's use Github/DigitalOcean for secrets and containers, but I work by myself and I have one project that runs about 20 servers (API's, webhooks, crons, etc.) with each having different slightly different .env's. I code in Rust instead of PHP or JS in the backend, so I'm more concerned with supervisor's configuration more than the actual env.
@tristan If you're having serious infrastructure you're already relying on external providers (whether it's AWS, GCP or Azure) so using different service for secrets doesn't make much difference.
But, how do you access your config server? At some point, there are credentials being held somewhere? If someone has access to your server where the code is deployed (which they must do if they can access your
.envfiles) then they can also do what they will with your code to access whatever you have held in a config server.Interesting, your screenshots are specifically from the Google offering, and you didn't disclose in this blog post that that is where you work.
People absolutely get their secrets stolen through .env files if they commit them to their repository.
Exactly, like, learn the BASICS of how to use them and you won't do that.
You should have a default gitignore that omits any "dot" files [chances are you don't want any of those to ever commit to your repo]
So yea, newbies make that mistake. Besides, so what?
You add it to gitignore , change all your passwords, and you're fine again.
Or you can use a config server and never have to worry about that "newbie" making that mistake with a modern solution.
@ravavyr, you're fine until another newbie comes into the team and needs to edit .env but makes a backup of it to avoid making a mistake, since it's not in GitHub. Later, when doing a
git add .the person forgets that the backup isn't in .gitignore -- he/she may not even know what a .gitignore is -- and accidentally commits the secrets.This scenario might sound a bit far fetched, but I have seen things like this happen many times. Also, some people will hide the mistake instead of going and changing all the passwords... it's easier to just pretend it didn't happen, that reverting the commit will fix the problem, and that no hackers will ever look at git history. :)
I will say that config servers were initially very intimidating, but once I got used to it, I personally can't imagine going back.
From a security standpoint, it's not perfect though. A tool is still only as good as the people using the tool. We still struggle with how to avoid getting the secrets into the config server. IT has the access, but the devs all have the know-how regarding what the secret is for. Sometimes this means devs create the secret and then share it in the chat system (a bad practice) so that the IT person can then add it in the config server... I guess things can still not make sense even with a config server.
I was strictly referring to your somewhat bold statement:
If your secrets became public then the env is already irrelevant.
No one gets your secrets from your env unless they've already gotten into your server.
Even the most basic servers have [dot] files blocked from access (or should, i'll admit some don't do it by default)
And yes, if your env got into your repo, changes the passwords and keys and make sure you add it to gitignore so it doesn't happen again.
Either way, the .env file has never been the problem.
If .env is in your repo, it public knowledge. Even if its a private gitlab, nobody keeps track of what secret is where and secrets in repos should therefor always be considered public knowledge.
There's plenty wrong with using
.envfiles, especially in production. Permissions could be wrong, it could expose passwords to folks who are authorized to support the application but are not authorized to access the database, it could be slurped, version-controlled, etc.Just because it is easy to maintain security of a single file in a single filesystem doesn't mean that the
.envfiles sitting on your laptops and production servers (if you're using containerization and clusters than it's one copy per machine).Add on to that the fact that if you use
.envfiles then you have to go manually change the files whenever you rotate or change tokens/keys/passwords, which frequently leads to people not changing those aforementioned credentials.Use a security service which encrypts, like a Secret Manager (GCP, Azure, and AWS have these), you could use Hashicorp Vault, there are a bunch of choices.
Don't use
.envfiles, the odds that your program will suffer a security event or will perpetuate bad security hygiene is substantially higher than if you use a vault or managed service for your secrets.Do use .env files to define environment variables in local dev (which is what it is for). Don't commit .env files to your repo. github.com/motdotla/dotenv#should-... .env was never ever meant to define environment variables in production/hosted environments, its for local env ONLY.
You make a great point because in AWS permissions can't be wrong. 0600 is really hard to get right, but IAM policy docs are super straightforward.
The passive aggression isn’t helpful. It is possible for us to have a respectful conversation while disagreeing.
IAM is an auditable, traceable mechanism which can be monitored and alerted on. I can see who makes an api call to fetch credentials in CloudTrail. I can see if my policy statements are too permissive with things like AWS Config, Trusted Advisor, or Access Analyzer. I can set and forget those roles, and make access to credentials something that adheres to least privilege.
I’d much rather risk a detectible fault IAM policy documents and roles than try to manage
.envfiles across a bunch of infrastructure, and constantly monitor engineers who will accidentally commit and push credentials to source control.If we were still in the days of monoliths on bare metal servers running in data centers, I wouldn’t be disagreeing as I am. But we’re not, at least not all of us are. The threat is much bigger than you seem to be giving it credit for. Using
.envfiles for production workloads in contemporary containerized cloud deployments is clearly a security anti-pattern.Lol I was just messing around. You're right though. There's definitely not a way to audit syscalls about file and attribute operations and send those to a SIEM and alert on them. And even if that existed it's not like it's best practice to monitor those things anyway. So yeah I'm in your camp on this one.
Lol mate...
You're so dead wrong on every point.
This is why nconf with it's ssm layer exists.
care to elaborate or you just know one solution and that must be the best safest way to do things all the time?
use what ever solution you like as long as :
So yeah, that pretty much eliminates your one trick pony.
you don't even need nconf, i could come up with something in several languages that do this:
Didn’t Uber just got hack recently because they put their secret in their script?
neat story:
portswigger.net/daily-swig/uber-ha...
and yea an admin screwed up, that's the point, people screw up, but that doesn't mean using environment files with secrets in them is the actual problem. Using them improperly is.
Please can you teach me the the proper way to use . env file?
no one you know or no one ever ?
if you're gonna say someone did, point them out pls.
I won't point them out cause I'm not here to expose anybody, but I've seen it personnally, on multiple occasions.
Good for you if you got some likes but your comment makes no sense to me. Besides, why fuzzing/hacking tools would include
.envin their list if it's that irrelevant.Look, i can't help you see the reason this entire argument is pointless.
Suffice it to say that all security measures are flawed because they are implemented by human beings and have to be maintained by human beings.
What does hurt projects often [that i've experienced with at least half a dozen clients] is being overly paranoid and trying to secure everything to the point where basic assets are not accessible and sites go down when they shouldn't. At that point it's hurting more than it's helping. And having a .ENV file in 16 years has not once been the problem. So per my experience, it's not an issue. You claim otherwise, and as everything in this industry, we can leave it to personal preference.
I see you're really concerned, but you don't demonstrate anything. Why do you consider not using .env is being "overly paranoid"?
If you care about error 500 and other inconveniences, it happens a lot with .env, and many beginners have difficulties using them properly. Most of the time, teams use it because the framework forces them to use it, not as an internal methodology.
I don't know you, but it sounds like "I don't want to change my habits, I've been doing that for 16 years." If you're careful, which I assume you do, you might be ok with that approach, but that does not mean it's the best one.
Config files like that are primary targets that will be automatically scanned, and sadly, people tend to use the same credentials and API keys in all environments, sometimes making the attack even easier.
lol
500 errors only happen if you wrote bad code or didn't debug it enough. This is fact. I fully expect 500 errors if i forget to setup the correct configs in the .env file. You're supposed to fix those across your application and account for any combination of them and make sure to log them and keep an eye on those logs for new ones and then fixing them.
note, my habits extend 16 years, which means i've been coding since before env files existed and i still run some older monolith systems while also keeping up with various frameworks, platforms, services, tools and whatever else people keep coming up with. Config files only get scanned automatically if you don't secure the damn things which again shows that maybe you just lack experience in the field. Credential sharing happens yes, and it's just as bad as not securing your env file from external access, but so is clicking on a bad link in an email, or not setting folder permissions correctly, or a mountain of other issues. ENV files are not the problem, nor were they ever.
Discuss starts with "lol." If you think this is a battle, then win. Sorry to say that again, but you don't demonstrate anything. I've seen many uses of
.envfor convenience I did not find convenient or particularly efficient, but if I have to use them, then I use them.Does not mean it's the best strategy out there. Does not mean you have to migrate all legacy projects right now because someone said you have to. Maybe think about other approaches for your next project.
I think @ravavyr is largely spot on here.
That said, .env files do have their weaknesses. We are addressing those with dotenv-vault - from the same people that pioneered dotenv.
At the end of the day, your dotenv-vault is functioning like a config server.
At that point it doesn't matter who's spot, this is a win for me in my books and my principles are still in play!!
Good job, can't wait to try it out and see people using my principles whether they know it or not 😈.
Yes, you are correct.
I think we are toward the same end here. You recognize the problem we see as well.
But throwing out the .env file will be a mistake. They need to work together. The config server should layer on top of .env files.
Currently, all implementations of config servers require you to learn a new proprietary system, rewrite code, and get locked into it. Plus, there is training on the new system for your dev team.
That's why we think all config servers should be built on top of the defacto .env file standard. This way, you get all the security benefits of .env files PLUS solve the insecure sharing and config issues.
That is what we are doing with dotenv-vault.
This has the added benefit that you could choose to leave dotenv-vault, and everything would still work. Or you could switch to a different provider that syncs your .env files for your team.
Sounds good to me no matter the underlying system as long as the benefits listed in my article gets implemented.
But seriously guys good job, it feels good to have the brains behind the bases of my very argument make amends with me. I’m so happy right now!!
every single security measure has its weaknesses, going by the original post here, using .ENV files is a thousand times simpler than what he proposes and maybe slightly less secure, but a thousand times easier to repair in case of an attack than the original post's process. That's all i was getting at and this thing blew up lol
True, and he also forgot to mention the dependency of SDK, .env isn't bad, if steps are taken to secure it, and infact there's no way people are going to directly update the code or version the .env, they are going use it for local development, and rather build validation in code to check for environment variables and add up new vars progressively, Secret Managers are special purpose services, they add up complexity for small systems, but solve the problem of scattered and untracked env for large systems, obviously with overhead of using client sdks. Still if you still wish use some secret store for even local dev, use Doppler.
Then why do the people who made dotenv explicitly tell you to NOT commit .env to version control? github.com/motdotla/dotenv#should-...
People never stop to disappoint me with their clinging to bad practices.
not commiting it to version control is the right thing to do, but using it as source of truth for your variables is simplest thing to do, if one needs unified way of injecting secrets, from various sources there are tool out there like tlr.dev/ which can source secrets from AWS, Vault, etc. all in one place without even using any SDK. That totally depends on choice of the devs.
Depending on which service you use, you could slide by without an SDK.
I've never heard of a
.envused for build validation, can you give an example.That's kind of an overstatement. A quick regex search on Github reveals tons of projects still include their .env in the repo, and of course with their secrets in it
Exactly
Honestly... This sounds great compared!!
Read this and join me in knowing that people definitely do get their secrets exposed through .env files...
dev.to/mittal69353530/adding-an-en...
Oh my, oh my. I'm feeling a sense of irony from some of my criticizers.
It is myth busting time.
1.
.envfiles are unreliableHaving the environment settings come via network adds a whole lot of overhead.
A lot of bad things can happen.
You can by all means make a web service reliable. But how is an external web service more reliable than a file I have access to locally?
Sure, you can write bad code that allows attackers toaccess arbitrary files. But you might as well leak your environment variables even when you would retrieve them through a central environment provider. Object property injection is a thing. Essentially allowing you to instruct your server to serialize and send you the
envvia any API.There is absolutely nothing from preventing you to write bad code. No matter if env is directly injected to the app or read from an
.envfile.2.
.envfiles have no access managementYou can. Both locally and remotely.
How do you run your apps? If I run my docker containers, I can have my compose file read from an
.envfile and provide that to the containers. The app doesn't even need to know an.envfile exists. But the only user starting apps would have access to the.envfile. You can control very well on any system who is going to be the guy that gets to run apps.Secondly, you can also manage access who is allowed to read and write the
.envfile.3.
.envfiles have no versioningCheck your
.envfile into git - no, not the public project. Create a separate repository that you can make private and only give access to limited people.4.
.envfiles have no safe updatesThere is a little bit of lack of tooling here maybe, but you can create a contract between your app and what environment it expects.
In TypeScript and NodeJS you can simply overwrite the nodejs
process.envinterface, providing a clean contract between your app and the environment vars it needs. It would be even better, if you could generate that from a.envfile. Problem solved.When I roll out a new container, I update both my repo containing the new .env file for compose + the app.
Summary
.envfiles are a battle-tested and widely adopted and supported feature. They make deployments predictable and secure. If you have trust issues, then you can sufficiently lock it down and only inject it via docker containers.I absolutely don't see a need for involving yet another server.
I'm not sure you read the article completely.
Yea, nothing is full proof; anything can go down, doesn't mean its not a viable option. When config variables are updated, they keep the same entry point ex.
config/DB_PASSWORD/v1. Only the value is changed, this is how an paper trail is kept.When that .env is updated the updated has full access to all secrets, no matter how secure that file is nothing will change that. My point wasn't that the file couldn't be stored properly.
The creaters of dotenv tell you not to commit it in your repo. Look below, tons of comments say the sme thing.
Summary
Please re-read the article, your points are not aligning with what I was saying. Yea its widely adopted, but change takes time.
From what I understand , he's more like talking about a file-based config management solution, with git-based version control, meaning a dedicated private git service to manage the config file versions not exposed to outside access, that is quite different from "commit it in your repo" which means more about committing it in your application code repo.
Also you can easily have a system that automatically merge multiple config files into one .env file, which should not be directly accessible to anyone, so people updating the individual config files won't have access to configs in other files.
Like I said, you can have a file-based solution that achieves more or less the same goals as a database-based solution, which may be easier or harder to implement depending on your specific situation and needs. After all, databases are file-based solutions themselves, they too store data in files (well, some store non-persistent data in memory, but you get the idea :P )
Sounds interesting but still opening one .env exposes all secrets in it. A config server mitigates that risk while giving feature like audit logs and automatic secret rotation.
You also get the benefit of a ui if you care about that.
The point is in such a file-based config management solution no one should be able to directly open the .env file, it should be dynamically generated and managed by the system, more akin to a "flat database file" if you like to think that way. And features like audit logs and automatic secret rotation can be implemented with no more than a couple dozen lines of code, without the need of a config server. Like I said, the main point of your post is actually not about "config servers" (albeit you are advocating their usage), but rather about storing the config data in remote databases vs. local files/environments.
And the existence of a UI can be a pro or con in the case of config administration depending on specific needs. It's also quite trivial to setup a simple database admin UI if you decide to store config data in remote databases anyway.
Honestly at this point you are just implementing a config server yourself. Every feature I mention you are saying just write it. If that's what you want to do by all means. We can agree to meet in the middle at this point.
That means you can just implement a file-based "config server" with .env files, and your "config server" itself would also use .env files for config, and it could have better security, cleaner code, and maybe even easier and faster to implement than trying to deploy and adapt to use one of those enormous "config servers" you mentioned in the post depending on your particular needs.
Which is exactly the point:
1) .env files (or config files in general) based config management solutions are not inherently worse than those based on remote databases. Don't just blindly diss config file based systems in favor of those based on remote databases, they both have their pros and cons, and one may be improved very easily over the other depending on your particular needs.
2) for experienced devops, it's usually quite trivial to implement a bespoke config management system for things like config versioning, sharing and access control, and it could likely be the preferred way over using one of those enormous "config server" with tons of code and myriads of dependencies.
Yea, just like you can implement a no sql database with json files, we have tools built for various use-cases. How do you know config servers are enormous, have you used or checked one out? How is tons of code a useful measurement in this conversation again?
All these features you are saying that are trivial to mention can also grow into these "enormous" config servers you keep mentioning. OOOR you can use an already established and battle-tested implementation to avoid introducing the very bugs you are talking about.
And not to be rude, but how much experience do you have?
I have checked the "config server" you mentioned in the post,, it has hundreds of dependencies and more than 600k lines of code, which is really ENORMOUS considering the fact that only several hundreds of lines of code are needed to realize the features you mentioned most in the post.
log4j was established and battle-tested, struts2 was established and battle-tested, and when there are orders of magnitude of difference in the amount of code involved, then it's a useful measurement. When it comes to managing important sensitive data like passwords in devops, I'd always trust 600 lines of code written by myself with a handful dependencies than 600k lines of code written by other people with 500 dependencies, when they basically serve the same goals for my particular needs.
And nope a simple bespoke config management solution is next to impossible to grow into those enormous config servers, for example if you code a simple server with koa and several self-made middlewares for your particular needs, it will never grow anywhere near the size of Spring or Rails. From what your post suggests, using a "config server" with 600k lines of code and 500 dependencies just to address pain points like version control, config sharing, access control, data encryption and secret rotation kind of stuff, that sounds like you wanting to deploy an entire Spring Cloud infrastructure just to serve a simple website. Or like wanting to use a NoSQL database when a plain JSON file is more than sufficient to store your data.
And like I said, the solutions to those pain points mentioned in your post is really not about config servers vs. dotenv files, but more about centralized remote databases vs. local config files.
I'll just say I have lead teams of devops with 5 to 10 years of experience, using many kinds of configuration management solutions across various projects of various sizes, from simple bespoke solutions based on config files and/or centralized databases, to distributed systems based on things like zookeeper, and tools for k8s, etc. From what I can see, those "config servers" mentioned in your post are simply WAY overkill as replacements for some config files to just address those handful pain points you mentioned in the post.
Again. WHY DO YOU KEEP BRINGING UP LINES OF CODE AS A METRIC OF USEFULNESS??
About half of HashiCorp Vault's code is unit, system, and integration tests.
They thoroughly test the front-end and the back-end. And to take it a step forward they have end-to-end tests and acceptance tests.
Tests written to avoid these "vulnerbilities" that you keep preaching have multiple lines of code create.
Ok then you can write a config server yourself if you are scared of vulernerbilities. While you are at it MySQL has a vulnerbility back in May so you might want to write your own database engine while your at it.
Also, Chrome just recently had a vulnerbility too. Don't forget to build a while web browser as well. Actually, Dev.to has a bug page, yikes you might wanna re-write this website too.
I mentioned in the article that config servers are usually accessed form VPCs so even if one of those "hundreds of dependencies" has a bug, the config server is protected from the internet.
Also I find what your saying kinda funny.
If a project uses tons of dependencies, then the project is bad. Instead, if they write the features by hand which increases the lines of code, that also makes the project bad.
You really need to read what others' saying before replying completely irrelevant things. If I just need to store 1k of data, I'll not use MySQL, I'll just use a JSON file. If I just want to download a file, I'll not use Chrome, I'll just use curl or wget. You must have some serious problems with your eyes IF you can't properly read what others write.
I clearly said those handful of pain points you mentioned in your post does NOT require those enormous "config servers" to address. Those are some simple goals achievable with some very simple config management solutions based on config files, or if you want, with centralized databases, which may take even less time and easier to implement than deploying and adapting to one of those "config servers" mentioned in your post. If you want to call such simple config management solutions "config servers" too, fine. I guess you can call both koa and Spring as "web frameworks", but their differences are huge.
Like I said, using one of those enormous "config servers" to address those pain points mentioned in your post is like deploying an entire Spring Cloud infrastructure to just serve a simple website with features that can be done by using just koa and a handful simple self-made middlewares. Yes you can try deploy a complete Spring Cloud stack to just serve a simple website, but it will take more time, use more resources, and exposed to more potential vulnerabilities UNNECESSARILY, so noone would recommend that.
When you can address all your pain points regarding dotenv files with some simple bespoke solution of several hundreds lines of code, deploying and adapting to an enormous "config server" mentioned in your post to achieve the same goals is really not something to recommend.
Next time try really read and understand what others have said first.
===
You seem to really have some serious problems with your eyes, or your English comprehension skills. What I said is that achieving the same goals with less code is good, and if by orders of magnitude less amount of code, then that's really really good. More dependencies usually means more code in total to achieve your goals.
And about the lines of code as a useful measurement, yes they are when the differences are ORDERS OF MAGNITUDE kind of huge. A simple "Hello World" program practically has zero bug (of course it has next to nil functionality too), 600k lines of code may not differ from 500k lines of code much in terms of complexity and potential bugginess, but if THE SAME GOALS can be achieved by 600 lines of code instead of 600k lines of code, that's some big difference.
I don't think we are going to get anywhere, we are clearly at a disconnect and throwing insults isn't going to solve much. Let's just agree to meet in the middle on this.
no insults, just curious, is English your first language? I kinda understand what you are trying to say with "agree to meet in the middle" here, but that sounds really a little odd here, just saying.
Yes english is my first langauge. What i mean by meet in the middle, I'm saying we both respectfully acknowledge the other person side and agree to disagree.
Yeah I know what it means, just don't see it usually used like this.
I don't want to keep pinging you so I'll make this the last message, but I'm curious, how do you normally see the phrase used?
usually when there's no compromise it's just agree to disagree.
I have read the article. I am very certain I understand what you propose. It does not work for the environments I work with and I know I am not alone. For the regular self hosted app running on a virtual private server, I would argue you also won't have a need for config servers.
I build industrial applications in demilitarized zones - low trust environments. The customer decides when to go online. We can't expect that an internet connection is there all the time unlike in a cloud application. Not when setting up the software, not when starting the software and also not when updating the software.
This is not a hobby project and it really misses a lot of tooling that yet has to be invented. Because neither docker, k8s or any tooling helps you in continuously rolling out your applications and updates to non-cloud environments. You need to have a very narrow Google-esque vision to assume every app is deployed, managed and run by some cloud.
The problem is a whole lot bigger than ".env files are not secure and don't version". There is a huge gap between the tooling and infrastructure a cloud provides, and the mostly self-baked solutions you have to apply for every tier below that. Docker and docker-compose does work for deploying, but there is an entire slew of tooling that we lack to properly update and version applications that don't run in the cloud.
You mostly have 3 ways of running your apps:
In the Cloud you probably have your complete kubernetes cluster and the entire infrastructure needed to run, monitor, update, version control and deploy your app and its dependencies. There you probably don't have a need for .env files. Even static site cloud hosts like netlify allow you to tune the env from a web ui.
On a virtual private server you have to come up with your own infrastructure. You can't expect everyone to run k8s for a business application. Even then, having too many moving parts can kill your software, as I outlined in my previous post. If I can avoid having a separate runtime dependency, I will avoid it.
All other external systems - A git repository, a CI server, a public key infrastructure, a letsencrypt ssl service, a monitoring service are all optional by design. You can still deploy an app if your CI server gets down. You can still build and package the software locally if your git repo is down. You just don't get monitoring data if sentry is down - the app will continue to run. A certificate will be stored locally and make your app at least secure for ~3 months. But your app won't even start, if your config server is not reachable or has other problems. Even for VPS I would use an
.envfile.On bare metal - hardware demilitarized zones and low trust environments are very valid use cases for IoT or public service applications - Both industrial and homebrewn (e.g. smart home). You can't expect that the customer connects to the internet only to get some env vars when starting, installing or updating their app. And I also wouldn't recommend a nuclear power plant to connect to the internet - no matter how secure your communication is. Even for modern industries, sensor data is sacred and they want to very tightly control when software is allowed to synchronize its data with some cloud application. Your across-the-board suggestions "stop using .env files" stops to work here completely.
For Development For development and for documentation, .env files are still great. Setting environment variables for development on windows is a nightmare. In Linux you can at least do
DB_PORT=3306 DB_HOST=whatever.com npm run start. But it still convoluted for development. Try to convey what environment variables are needed for your app in order to run to your coworkers. An .env file will document and communicate just that - a set of env vars required for your app to run. Configurable in one place.We are very far away from having best practices or tools for running software on non-cloud environments.
Docker and compose make it easier, by having predictable deployments. But everything between is left as an exercise to the developer/devops guy deploying the application.
Before I start to figure out how to get rid of a well working tool like
.env, I'd rather find out how to fill in the gap of all the missing tooling. Nx is a good job in providing better building tools, combining a slew of independent tools to make build less painful. GitLab is collapsing more and more tools into one tool - since we have experienced in the past years that running and combining a separate repository, a separate CI and a separate container registry is painful. I'd bet my money on these guys.Would love to have @jarrodhroberson opinion on this :)
They read it, you're just ignoring anyone who disagrees with your opinion and assuming they're wrong, rather than listening
Hard to believe that when most of my responses are well typed and counter-points, and most of who disagrees are the ones who assume I'm wrong, haven't read it, and don't listen.
Recently, I've decided to leave this article in the past for now. Take care!
I'm pretty sure you are the one who doesn't read counter-points of others, and doesn't listen to those who make them
.env files should only be used locally in a dev environment. Not sure using them in production is a thing anyone does?
Using something like KeyVault for production makes sense but I prefer to use CI/CD secret services that github / gitlab provide. Makes sense to me that secrets belong to the build / deploy process.
A word of caution with this approach - make 100% sure you're not exposing your secrets in your artifacts!
Can you explain why? If you're configuring secrets in your environment at the build step, they would still be accessible in the deployed system, thus no less vulnerable than if they're in a
.envfile aren't they?I mean, if I have enough privilege to access a private file on a server, then theoretically, I can inspect the server's environment anyway.
I just feel like I'm missing something here...
Using secrets in CI/CD adds an additional level of protection for secret values through additional access controls but it isn't the only security needed. Assume you have a public repo but want to keep the secrets safe from people who have access to the source code, or you are a company with contractors who need to access the source code but should not have access to secrets.
If someone is able to access the server then other security should be considered such as using a VPC for private API's. Security should have lots of layers, protecting secrets by moving them outside the source code is just one layer.
I think this depends on what kind of secrets. CI/CD secret can be used for CI/CD, eg: secrets to access Docker, K8s cluster, artifact stores ... Application secrets are different layer, so we need extra tool to provide those. All artifacts should be stateless, loading application secrets is likely effect that happens on runtime instead build/deploy time.
I don't think there's any harm in storing secrets in a
.envfile, as long as the file is kept safe and secure. In fact, I believe it's a good practice to follow the Twelve-Factor App's recommendations.In their chapter on Config, they state that config should be in environment variables. This is good advice to follow, as it can help keep your application secure.
Moreover, by taking the necessary precautions, the risks of storing secrets in a
.envfile can be minimized. For instance, be sure to keep your.envfile in a secure location and make sure that only authorized personnel have access to it. Additionally, you should encrypt any sensitive data that you store in your.envfile.By following these guidelines, you can ensure that your secrets are safe and secure while still being able to take advantage of the benefits of using a
.envfile.This doesn't solve that anyone who updates it will have access to all your secrets in plain text. No matter how secure the
.envit doesn't change that.Also, environment variables are global to that machine, if someone gains access then they can read out your variables. This was one of the main techniques when the Log4J was exploited.
As I mentioned, it's essential to encrypt your secrets rather than storing them as plain text. This way, even if someone can read your env variables, they won't be able to use it because it is a useless scrambled string of characters. Additionally, you can set the file permissions so that only the owner can read it, further protecting your secrets. Take the following Unix command as an example:
This will ensure that only the file owner can read it and that no one else can access it. This is a great way to keep your sensitive information safe and secure and ensure that only those who need to see it can do so.
What about sharing a secret to other? With a config server it’s as simple as ‘corp.secrets.com/config/DB_PASS/v2’.
In this case, you can create an Inbound firewall rule or add an entry in the ACL to allow only specific users to access this file.
Again giving full access to every secret in the file to that user.
Now, the problem becomes tied to the party that you are sharing it with rather than the
.envfile itself. There is a risk in giving away your secrets to a centralized third party(Like Config servers), as it can become the main target for hacky wacky organizations. It is important to consider who you are sharing this information with and whether or not they can be trusted.The best solution sounds like hosting your own config server so you don't have to consider if the person you're sharing information with can be trusted while utilizing permissions and access restrictions. Win:Win.
Exactly. However, many startups nowadays are content to take the easy way out, opting for quick and easy solutions that don't really offer much in the long-term. They don't understand the risks involved in this approach, and only realize it when it costs them time and money.
Exactly! This is how I recovered some secrets back in the day. You SSH to the working machine,
env MY_SUPER_SECRET_LOLand you are done and done.I think that's probably because your ssh'ed into the server as the owner of the file or have root access. However, one thing you can't do is open a session to the server and get access to its environment variables. Nor can you log into a shell on that server and get them unless you have root access or own the web server process.
Exactly, it's more or less a moot point to talk about security if the root access to the system is already compromised. If someone has root access to the application system, then it's trivial to extract/intercept the passwords, whether you store them in files, environment variables, or get them from "secret URLs"
Then where do you store the configs and passwords/encryption keys of your config server? Using a config server just means you store your application database passwords (and other things) in another database, then you still need to store THAT database's password somewhere, and eventually you will need to store SOME passwords and/or encryption keys in either plain text or environment variables.
I think your entire post is basically the same as saying "when you have to manage many complex systems, don't store their production configs and passwords in files, store and manage them in another database system", which is quite valid for the ease of config versioning, sharing and access control (albeit the same goal can be achieved with file-based solutions too), however in the end you have to store some configs and passwords outside of a database system anyway.
Also if data security is really important and you want to serve sensitive data like passwords via remote APIs, maybe you should just code some simple bespoke config management solution instead of using those enormous "config server" things. I'd say with several hundred lines of code and a handful of dependencies, you should be able to have most of those particular config versioning, sharing and access control capabilities you need, and you have full control of the code that manages your extremely sensitive data, instead of trusting those config server's thousands of lines of code and hundreds of dependencies that you have no grasp of, where a bug in the code or a vulnerability in some dependency may just compromise your "secret URL" completely.
Sure, build your own solution if you don't want to trust a third-party. If your config url was exposed it wouldn't matter. Most config servers are connected through VPCs which prevent internet access and can only be accessed by your servers.
Well, remote APIs are inherently less secure than local access, and using config server just means you move your sensitive data from local files to a remote database. Of course it can be less prone to human errors and much easier to use, but in the end I doubt either approach (remote database vs. local file) can be said decisively safer, it really depends on the specific situation and implementation.
I do agree if you are managing many complex production systems you should not manage configs with individual files, unless you have some good file-based config management solution in place.
But then in the end the configs and passwords for this remote database/config server thing still needs to be stored in some environment variables and/or .env files (or rc files, or whatever local config files you want to name them) anyway.
When using a VPC you are essentially using local access. Yes you are moving sentive data behind permissions and access retrictions. I do however believe that .env should be used for local / development oriented enviornments like I mentioned in the article.
I guess we can agree to meet in the middle.
You should never commit your .env to version control: github.com/motdotla/dotenv#should-...
Yeah, but you can commit something like this and generate the
.envfile upon build/deploy and set the values.But why? All hosting providers I can think of, provide a way to specify environment variables. Why make it more complicated than it needs to be? Environment variables: it's in the name, they're specific to an environment. It's illogical and a potential security risk to put something thats environment specific into something that is not (your source code/version control).
Despite the grabby article headline that I usually avoid on principle. I vehemently agree that .env was and is a terrible idea from multiple viewpoints, including software delivery and security. The best solution though, is to use passwordless identities to establish trust then nobody, including developers, need to know any secrets and when they leave the organization their access leaves with them. PKI should be considered first, not last.
I don't get it, like I mean
.sshpasswords and keys are also stored as files somewhere on your server. Having a.envfile in your www folder (or whatever you are using) is not inherently insecure. Obviously if you made it publicly available, it would be. I get that GCP or whatever builds this into vaults and that's great if you're using GCP, and perhaps Google builds in extra security, but there is really nothing inherently wrong with using a.envfile.If you're an experienced security consultant working with other experienced security consultants, I'm confident no one will commit that file, but most developers who are not in the security field just don't think that far ahead.
In my experience, someone will eventually commit that file. It won't be you or me, but someone at some point will do it, and it also might not be discovered for months until someone like you or I does an audit and discovers it in the commit history.
With that said, people can still misuse the config servers and secret managers too. At the end of the day, the tools are only as good as the people using them.
Great article.
I worked at a company using docker, we build docker without any user inside, eg you could use the service it provids but it hade no user inside to access.
If we hade to fix any thing, just build a new one.
Anders way of thinking about security in a modern cloud environment works for me, it's the 'cattle-not-pets' approach: deploying immutable closed systems (no user access) that have no need of configuration / comissioning, a nice blog post here: tempatrumah.blogspot.com/2021/07/e...
This strategy allows security controls such as full system integrity monitoring and automated intrusion / modification handing (eg: isolate and freeze the affected system for forensics, deploy another one via the automated scaler / pipeline).
Add a well-thought out separation of concerns strategy (be that managed access to deployment pipeline data that gets baked into the system, in-cloud secrets managers that lean on machine identity for access control, separate production data repo with managed access, whatever works for your scale) and you have addressed most of the risk being discussed here.
The problems arise with production systems that are pets, lots of humans poking about and little to no chance of spotting a problem fast enough to act effectively.
Interesting. I will make some time to investigate these and see if I can integrate support for them into wj-config. Basically create a data source that understands how the server works.
And yes, I agree that
dotenvis not the greatest. If it were, I would not have had to make wj-config in the first place.What is the main use-case for wj-config?
wj-config is a versatile configuration package that does for JavaScript what Microsoft did for .Net: A hierarchical configuration system composed by multiple configuration data sources. Main use case? Any JavaScript project, basically, that requires configuration that has the concept of environments or similar scenarios, like per-tenant configuration, for instance.
The author seamingly has no idea of git ignore/exclude (decades old), the concept of pipeline variables (which obviously can be files or generate files as well) and the multiple risks which come when you hand over your business secrets to third party companies or tools (regarding technical security as well as business stability).
So I don't say that too often but don't listen to that load of bull crap and DO use .env files and DO use them correctly (which can depend on company size but yes, even as multi million dollar comapny and above this is still the right approach!).
Please re-read if this is your conclusion.
In your first point you rumble on that there is no good location for the .env file since you cannot put it in your repo (structure). If you don't see how this corelates to my statement and thus can be resolved with git ignore/exclude idk how to bring it to you...
And I said "and tools". With those tools you increase your attack surface by miles. And this is not a hypothetical thread. Just look at all the supply chain attacks and other (zero day) vulnerabilities in dev op tools of the last year alone. And for a tool which stores your very soul (secrets) of the whole operation there is simply no good reason to take that risk. You would need to do a 100% code review on every update of this tools and since it is open source you also need to update and that fast since vulnerabilities are also "openly" known. And again for a task as simple as this (just using .env files properly) you go for this risk for no good reason (besides finding the tool flashy or something).
Still you don't seem to even know the concept of pipeline variables (or understand it). In a very basic approach you could have one repo server and multiple build and deploy server which poll the repo. They also hold the secrets. Your devs never get in touch with them and never even know the secrets. The 'updater' (as you call it) of the secrets is the person in charge of that project and he also only has access to the build and deploy server for his products. This is just a very basic example and obviously pipelines in the real world are more complex than that but still very well and very secure doable with just bash and your typical and rock stable unix tools.
So you say a team of somewhat 5 to 20 devs (I rarely have seen more on just one product/scope which shares the same dev secrets - note that they only are dev secrets anyways and it wouldn't be mission critical if they get compromised) cannot share one small text file easily and secure enough? We have come far if the current generation needs a 10k+ loc tool with a flashy name for a task like that...
Than you seem to have more information on that topic than me. Strange though that all their infrastructure and most tools explicitly do support .env files 🤔 And do they just don't use .env files or config files in general? And on all projects or just some? I guess I would need more sources on that one 😆
I don’t know know where to go either with this point.
Nothing is full proof not this site, not hosting your code on GitHub, or deploying on Digital Ocean, GCP, AWS, etc…
So a person seeing all secrets is ok because he is in charge? That does not sit well with me. At Google you can be the VP of a project and you still don’t have clearance to view a secret.
Change takes time, why use anything? Tools are built to make life easier. We don’t need to copy and paste or trust anyone because we have a tool with permissions to do that for us. And it doesn’t matter how many lines of code it has, why is that a metric that need to be brought up?