This message was deleted.

We self host gitlab, and have tons of problems with atlantis. The largest being, anything after the FIRST

external pipeline

no longer shows up under the MR itself, so you have to go into CICD/Pipelines and find the job to see what its doing

So possibly the changes I suggest on the Issue, it should start working as expected.

we're on self hosted gitlab v15.11 we have zero issues.

So we were on 15.x, now we are on 16.x same issue though, and thats in 2 different Gitlab Environments

Thank you all for the confirmation that this affects more people. Can I kindly ask you to vote on that issue so that it can be assessed as soon as possible?

yes, we have a AtlantisGatekeeper gitlabci that makes sure that the MR cannot be merged until a successful

atlantis apply

has not been run

@Mustafa Mujahid 1. How the hell did you name your atlantis pipeline? I would love to add this same thing to ours 2. The

split

pipeline for us behaves like this. First push, the

external

pipeline shows under the MR's pipelines. Every push after that creates a new pipeline with

external

that is NOT showing up under the MR, you have to go into CICD/Pipelines and find it

name your atlantis pipeline?

what are you referring to ? what do yours look like? regarding the split, I guess i haven't had an MR recently with multiple pushes, but honestly i never find myself looking at the actual CIs running, just wait for the plan to pop-up. so perhaps i need to to have a look specifically to push multiple times and explicitly look at the CI.

your

Atlantis Gatekeep

How are you doing, so it understands to wait for

external

yes, we make sure that the Gatekeeper always fails unless it detects the

if name == "atlantis/apply" and status == "success":

. we do this by using a post hook to scrape the external jobs and look for this in

.gitlab-ci.yaml

. As long as the gatekeeper fails gitlab does not allow merging based on the setting in gitlab.

Ahh got it, i was really hoping there was a way to just refer to an external pipeline without doing scraping

unfortunately not

Yes, that does work for us. But once a plan is taking 20+ minutes, they assume its broken, and start trying to push more and more

hmmm, yeah, if it stuck then yes, i see the predicament 🤔

We run in EKS, so if a node scales or w/e, and atlantis is moved

I will try it out, and let you know here if i have the same behaviour

I would say thats more of an atlantis problem,

it's a stateful workload, you can't run stateful workloads on ephemeral infra.

thats not a fair statement

how does it look like, this split pipeline behavior ?

Similar to what @Justin S mentioned, when we upgraded Atlantis we started to have the pipeline of Atlantis decoupled from the main pipelines, meaning it would not appear as your screenshots. Examples below • Expected behavior (first pic) with single pipeline • Current behavior (second pic) with dual pipelines (probably duplicate is not the best wording) Additionally in some cases it would only start if we activated temporarily the

Enable merged results pipelines

option on Merge Requests. Reverting the option would not revert to single pipeline and would forever be with dual pipelines

ahh @ @Ricardo Silveira so ours looks like yours. except the

right side

may end up being duplicated n+1 times

I really don't want to argue about this, as i don;t really care how you're running your stuff 😄 , it works completely fine for us, running even >200 projects per MR

I see, even more weird than ours then. The issue I created also states that removing the added code previously reverts the behavior to be ok, but I really don't understand what the original author was solving with it 😕 That is why I wanted to first understand his problem (or any other person that had similar issue to him) and only then propose changes

@Ricardo Silveira I will test this out, i haven't had a loko at the CIs, i'll have a look

So probably I should've started by pinging @Michel Z. Santello (I think I got the right person) to understand what the original issue was 🙂

hey, so I tried it out. I did multiple pushes to this MR, and the jobs showed up in the CI

Thanks for giving it a try @Mustafa Mujahid 🙏