dumb-init VS vault-exfiltrate

I think you are on the right track in thinking it’s a signal handling issue. You mentioned using some “bash scripts”, have you tried something like dumb-init?

At the other extreme, there's dumb-init - it implements the special pid-1 behaviors and acts as a wrapper around the one script you want to run. It's ideal for containers or virtual machines that don't need user logins or more than one service.

Oh, and another thing. Many containers launch their main process from a shell script. When this happens, the shell script receives the SIGTERM event, not the application. Your shell script MUST relay SIGTERM events back to the main process, and it doesn’t happen by default. You can use a shell script wrapper, like dumb-init (https://github.com/yelp/dumb-init), as your entry point if you need to use a shell script on container startup.

It's a plus because Gentoo fully supports the choice of Systemd or OpenRC. It also has minit, dumb-init, sysvinit, cinit in tree for the more adventurous. No one was calling the AUR bloat, the parent comment just mentions that Gentoo has an equivalent project, GURU.

, but I prefer to not have to make this assumption and use an init system instead.

[1]: https://github.com/Yelp/dumb-init

> It has no init system.

Apologies that I can't link directly to the "--init" flag but docker actually does have an init, it's just (err, was?) compiled into the binary: https://docs.docker.com/engine/reference/commandline/run/#op...

My recollection is that it either adopted, or inspired, https://github.com/Yelp/dumb-init#readme which folks used to put into their Dockerfile as the init system back in the day

Folks (ahem, I'm looking at you, eks-anywhere[0]) who bundle systemd into a docker container are gravely misguided, and the ones which do so for the ability to launch sshd alongside the actual container's main process are truly, truly lost

0: https://github.com/aws/eks-anywhere/issues/838#issuecomment-...

You can use something like dumb-init which is designed to correctly handle signals

Vault attempts to protect against host compromise scenarios, but it's a very hard problem. Ultimately, in order to do anything useful, Vault deals with plaintext values in memory, and that means that yes, there are ways for an attacker to get access.

Here's a good example: https://github.com/slingamn/vault-exfiltrate

The Vault docs include a list of 'hardening' steps for secure production usage. These are great steps to take, but each one represents a mistake that could be made. And because the Vault process is trusted with plaintext secrets, the stakes are high. Making a mistake could lead to a compromise.

With EnvKey, the host server is never sent secrets in plaintext. For defense in depth, we also follow best practices for hardening our networks. But I think we've seen with Okta and other incidents that despite best intentions, best efforts, and strong engineering, trusting the host server whatsoever just isn't good enough anymore.