This message was deleted.

Plan Error

running "/atlantis/bin/terraform1.5.4 plan -input=false -refresh -out \"/atlantis/repos/ORG/terraform/100/default/TERRAFORM/PROJECT/PATH/PROJECT-DEFAULT.tfplan\"" in "/atlantis/repos/ORG/terraform/100/default/TERRAFORM/PROJECT/PATH": signal: killed

No problems detected!

I run another

atlantis plan

Plan Error

running "/atlantis/bin/terraform1.5.4 init -input=false -upgrade" in "/atlantis/repos/ORG/terraform/100/default/TERRAFORM/PROJECT/PATH": exit status 1

Initializing the backend...

Initializing provider plugins...
- Finding hashicorp/aws versions matching "5.10.0"...
- Installing hashicorp/aws v5.10.0...
╷
│ Error: Failed to install provider
│ 
│ Error while installing hashicorp/aws v5.10.0: open
│ /atlantis/plugin-cache/registry.terraform.io/hashicorp/aws/5.10.0/linux_amd64/terraform-provider-aws_v5.10.0_x5:
│ text file busy
╵

I've run into this using

plugin_cache_dir

in another tool and had to disable it: https://github.com/hashicorp/terraform/issues/31964#issuecomment-1271597992 Curious if other folks are able to successfully use

plugin_cache_dir

Hashi says this isn't a bug and that the plugin cache was never intended to be concurrency safe. I highly doubt they will fix it. I've worked around this by doing a few things here at OM1. • Reduce the parallel pool size to 5 so only 5 inits can run at any given moment. • Add a custom command to my atlantis.yaml that runs tf providers mirror to a different directory per-init (this sort of kills the whole concept of using a cache but I want stability over speed) Here is a link to another thread where I posted a workaround that is currently working for me: https://atlantis-community.slack.com/archives/C5MGGAV0C/p1693429478890169?thread_ts=1693428770.117169&cid=C5MGGAV0C

I submitted a patch to hashicorp for it, they were not interested. It can also be fixed in the backend library they use to download and unzip files, but getting that changed and then getting hashicorp to use a new version of their own backend library sounds even worse. However, you can work around the annoying terraform behavior perfectly by setting the environment variable

TF_PLUGIN_CACHE_MAY_BREAK_DEPENDENCY_LOCK_FILE

.. In fact atlantis should probably recommend or at least mention setting that.

For us, before TFv1.4, since we are using custom python wrapper for Atlantis plan, we parse the error for plugin related message and retry a few time for failed plans due to plugin cache issue. This works well for us. After TFv1.4, we start to use

TF_PLUGIN_CACHE_MAY_BREAK_DEPENDENCY_LOCK_FILE

since the retries are no longer helping for the

Text file busy

issue in TFv1.4 and above...