This message was deleted.
# atlantis-community
s
This message was deleted.
w
@Matt McLane That, and parallel plan may also cause transient issues because of the shared cache directory
j
^^ That too if you are not using a different workspace for everything
w
there are some proposals in there (for example, having a small amount of splay), but to my knowledge, none have been implemented
j
I was able to "work around" most of those plugin caching issues by using 2 custom run commands that do this: It sort of negates the whole idea of having a cache because it will grab the plugins/providers for every init in every config/state but I have found that better to deal with than constant
text file busy
errors when planning in parallel.
Copy code
run: terraform get
run: terraform providers mirror .terraform.d/plugins
init
plan
w
I’ve found it’s most problematic when the version is updated, presumably because in most other cases, it’s already in the cache
we have parallelism limited to 3, but a lot of states with similar dependencies; tbh, I mostly just deal with planning a second time when it does come up
your solution is more elegant, and is similar to what I’ve seen Hashicorp recommend about this type of problem
I have a (non-atlantis related) CI pipeline where I ended up not running the inits in parallel vs. trying to do something like that, just because of the complexity of making sure I got all the right stuff mirrored
m
I did set it to 1.5.5. I will try setting this environment variable and the two recommended steps in my workflow.
So far so good, thanks for all your help
j
Now if we could only get github to stop having webhook delays we would be in business
w
😆
yea, last couple days have been interesting. Maybe this is their strategy to further adoption of GitHub Actions
Maybe someone who’s an admin here can add the angry unicorn emoji at least