Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix perma-diff on instance templates. #1995

Merged
merged 4 commits into from
Oct 2, 2018
Merged

Conversation

paddycarver
Copy link
Contributor

When using instance templates, if you use one of our image shorthands at
the moment, you'll get a perma-diff. This is because the config gets
resolved to another format before it is set in state, and so we need to
set that other format in state.

Unfortunately, resolving images requires network access, so we have to
do this with CustomizeDiff. CustomizeDiff was having trouble (I think?
More on this below) on setting a field as not ForceNew once the field
was already set, so I moved the decision for whether a field was
ForceNew or not into CustomizeDiff. I also resolved the old and new
images, and if they were the same, cleared the diff.

Unfortunately, you can't actually clear a field on a sub-block right
now. You have to clear top-level fields only. So this will currently
throw an error. I opened hashicorp/terraform#18795 to fix that. Once
that's merged, and we vendor it here, this patch fixes the problem.

If hashicorp/terraform#18795 doesn't get merged, the next best
workaround is to keep track of all the fields under disk with a diff
in our CustomizeDiff, check whether they've all changed or not, and if
they've all changed, clear the changes on disk, which I think will
resolve the issue. That's just a massive pain, unfortunately.

When using instance templates, if you use one of our image shorthands at
the moment, you'll get a perma-diff. This is because the config gets
resolved to another format before it is set in state, and so we need to
set that other format in state.

Unfortunately, resolving images requires network access, so we have to
do this with CustomizeDiff. CustomizeDiff was having trouble (I think?
More on this below) on setting a field as not ForceNew once the field
was already set, so I moved the decision for whether a field was
ForceNew or not into CustomizeDiff. I also resolved the old and new
images, and if they were the same, cleared the diff.

Unfortunately, you can't actually clear a field on a sub-block right
now. You have to clear top-level fields only. So this will currently
throw an error. I opened hashicorp/terraform#18795 to fix that. Once
that's merged, and we vendor it here, this patch fixes the problem.

If hashicorp/terraform#18795 doesn't get merged, the next best
workaround is to keep track of _all_ the fields under `disk` with a diff
in our CustomizeDiff, check whether they've all changed or not, and if
they've all changed, clear the changes on `disk`, which I _think_ will
resolve the issue. That's just a massive pain, unfortunately.
@danawillow
Copy link
Contributor

Does this fix #2067?

@paddycarver
Copy link
Contributor Author

I don't believe so. This will fix a permadiff, #2067 requires a bit of logic/parsing in the flattenDisks helper for instance_template. I'll update that tomorrow.

Update the version of helper/schema we have vendored to take advantage
of fixes to the ResourceDiff.Clear function.
@paddycarver paddycarver added this to the 1.19.0 milestone Sep 30, 2018
"revision": "41e50bd32a8825a84535e353c3674af8ce799161",
"revisionTime": "2018-04-10T16:50:42Z",
"revision": "35d82b055591e9d47a254e68754216d8849ba67a",
"revisionTime": "2018-09-26T21:21:28Z",
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's probably not actually a big deal at all, but this has a versionExact and a revision time that is definitely a few versions ahead of the versionExact. Any idea how that happened?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If I had to guess, it's a result of govendor updating the revision and revisionTime, but not the version or versionExact. I've gone ahead and removed the version and versionExact.

@@ -420,6 +420,48 @@ func resourceComputeInstanceTemplate() *schema.Resource {
}
}

func resourceComputeInstanceTemplateCustomizeDiff(diff *schema.ResourceDiff, meta interface{}) error {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: I'd call this resourceComputeInstanceTemplateSourceImageCustomizeDiff, since presumably if we added more fields to customize we would want to do a customdiff.All to group them.

(also lol, every so often i think that maybe our resources should be like, objects, so we don't have to have these ridiculously long globally-unique function names)

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fair! I've been doing massive monolithic CustomizeDiff functions, but I think customdiff.All--which I just learned about!--is a better approach.

And yeah, long names are amazing. -_-

err = diff.ForceNew(key)
if err != nil {
return err
}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

shouldn't there be a continue here? otherwise the diff will get cleared on line 456

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good catch.

Fix Dana's comments, one of which exposed a bug: we weren't actually
fixing the diff. Because resolveImage can return multiple formats, and
only returns a self_link if a self_link is passed in, it was diffing
more often than not against a self_link supplied in the config. We just
had a bug in our CustomizeDiff function that concealed this.

I added the resolvedImageSelfLink helper function to turn the output
from resolveImage into a self_link, which fixes the problem. It also has
the knock-on effect of fixing #2067 at the same time, which is nice. I
decided to add another function instead of just modifying resolveImage
to always return a self_link because I don't want to deal with the
backwards compatibility problems of changing how we're storing a bunch
of things in state this close to 1.19.0. And honestly, we should
probably just be storing the self_link always, _anyways_.
@ghost ghost added size/l and removed size/m labels Oct 2, 2018
@paddycarver
Copy link
Contributor Author

I mentioned this in the commit, but the missing continue that @danawillow pointed out was actually masking that this didn't really fix the problem. Because resolveImage can return multiple formats, and only returns a self_link if a self_link is passed in, it was diffing more often than not against a self_link supplied in the config. The missing continue just concealed this.

I added the resolvedImageSelfLink helper function to turn the output from resolveImage into a self_link, which fixes the problem. It also has the knock-on effect of fixing #2067 at the same time, which is nice. I decided to add another function instead of just modifying resolveImage
to always return a self_link because I don't want to deal with the backwards compatibility problems of changing how we're storing a bunch of things in state this close to 1.19.0. And honestly, we should
probably just be storing the self_link always, anyways.

JSON is the worst.
Copy link
Contributor

@danawillow danawillow left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How does this fix #2067? That calls getRelativePath on a field that got returned from the API, and this doesn't change that.

Anyway, I'm not thrilled with this solution, but assuming that it does indeed work, I'll say it's fine. Can you file an issue about this to make sure it's something that we can make better in 2.0.0?

@paddycarver
Copy link
Contributor Author

Oh, you're right. We'd still need to use resolvedImageSelfLink here: https://github.com/terraform-providers/terraform-provider-google/blob/27bd76d9d40d0149d75a1489f83eba119a3b01d5/google/resource_compute_instance_template.go#L640

But this at least sets up a small PR to fix that?

@paddycarver
Copy link
Contributor Author

Can you file an issue about this to make sure it's something that we can make better in 2.0.0?

#2151

Going to go ahead and merge this now.

@paddycarver paddycarver merged commit c1d404d into master Oct 2, 2018
@paultyng paultyng deleted the paddy_1916_diff_fix branch October 29, 2018 19:27
@ghost
Copy link

ghost commented Nov 16, 2018

I'm going to lock this issue because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active issues.

If you feel this issue should be reopened, we encourage creating a new issue linking back to this one for added context. If you feel I made an error 🤖 🙉 , please reach out to my human friends 👉 hashibot-feedback@hashicorp.com. Thanks!

@ghost ghost locked and limited conversation to collaborators Nov 16, 2018
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

2 participants