store absolute addresses for resource dependencies in the state #23252
Conversation
We need to be able to reference all possible dependencies for ordering when the configuration is no longer present, which means that absolute addresses must be used. Since this is only to recreate the proper ordering for instance destruction, only resources addresses need to be listed rather than individual instance addresses.
ghost
added
the
sdkv1
jbardin
changed the title
| @@ -202,6 +205,20 @@ func prepareStateV4(sV4 *stateV4) (*File, tfdiags.Diagnostics) { | |||
| } | |||
| deps = append(deps, ref.Subject) | |||
| } | |||
| obj.DependsOn = deps | |||
There was a problem hiding this comment.
It looks like in principle we have enough information here to convert from an addrs.Resource or addrs.ResourceInstance to the absolute equivalent by calling addr.Absolute(moduleAddr). Would it be reasonable to normalize this difference away in the loader so that the rest of Terraform can deal only in Dependencies, or is there information captured in depends_on that can't be mapped directly to an addrs.AbsResource? (Perhaps depends_on sometimes contains non-resource references that we need to preserve for correct behavior until it gets upgraded?)
There was a problem hiding this comment.
Yeah, depends_on contains references to modules and values as well. We need to load those into the graph to find the absolute resources they will eventually connect.
There was a problem hiding this comment.
That makes sense... thanks.
As a compromise, can we document the old field as being deprecated and read-only and make sure that any time we update a state object we'll force the old one to nil and migrate everything into the new one? At least then that puts us on a path to removing the deprecated field, because (if I'm understanding correctly) everything would be migrated over into the new field by the next refresh walk.
(I suspect this might already be true, because IIRC the apply step constructs a new state object from scratch each time rather than modifying the old one in-place, but I wanted to mention it explicitly just because I couldn't see any tests here that prove it to be true by.)
| // found dependencies to those saved in the state. The existing dependencies | ||
| // are retained, as they may be missing from the config, and will be required | ||
| // for the updates and destroys during the next apply. | ||
| type EvalRefreshDependencies struct { |
There was a problem hiding this comment.
I feel a little unsure about this one, though perhaps I'm misunderstanding it. Is this updating the state during refresh to include new dependencies in the configuration? If so, it seems like this would be the first time we've made refresh take something from the configuration and represent it in the state, as opposed to taking information from upstream systems (via providers) and representing it in the state. Perhaps that's okay, but I'm a little nervous about creating that precedent if we can avoid it, to avoid confusing the user model of what terraform refresh does. 🤔
There was a problem hiding this comment.
After trying out this idea without refresh writing out dependencies, I think it's probably worth the slightly odd idea of the refreshed state being altered by the config, even if it's not really the instance state itself. Since refresh is the only moment we have to re-write resource state (apply does not write state if there's no changes), if we don't do this existing resources would never have the depends_on field fixed up.
jbardin
force-pushed
the
jbardin/abs-state-dependencies
branch
2 times, most recently
from
292ac66
to
af12cfe
Compare
The test fixture did not like having modules when using the generic json map, so read and compare the states in the final *File datastructure.
Make use of the new Dependencies field in the instance state. The inter-instance dependencies will be determined from the complete reference graph, so that absolute addresses can be stored, rather than just references within a module. The Dependencies are added to the node in the same manner as state, i.e. via an "attacher" interface and transformer. This is because dependencies are calculated from the graph itself, and not from the config.
Refresh should load any new dependencies found because of configuration or state changes, but retain any dependencies already in the state. Orphaned resources would not be in config, but we do not want to lose the destroy ordering for the later apply.
Updates resulting from orphaned instances should happen after the deletion of the instances.
The DestroyEdgeTransformer cannot determine ordering from the graph when the destroyers are from orphaned resources, because there are no references to resolve. The new stored Dependencies provides what we need to connect the instances in this case. We also add the StateDependencies method directly in the GraphNodeResourceInstance interface, since all instances already implement this, and we don't need another optional interface to check. The old code in DestroyEdgeTransformer may no longer be needed in the long run, but that can be determined separately, since too many of the tests start with an incomplete state and rely on the Dependencies being determined from the configuration alone.
jbardin
force-pushed
the
jbardin/abs-state-dependencies
branch
from
af12cfe
to
46dbb3d
Compare
|
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 have found a problem that seems similar to this, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further. |
ghost
locked and limited conversation to collaborators
We need to be able to reference all possible dependencies for ordering
when the configuration is no longer present, which means that absolute
addresses must be used. Since this is only to recreate the proper
ordering for instance destruction, only resources addresses need to be
listed rather than individual instance addresses.
The state version remains the same with this change, since it's only
adding a field, provides an easier upgrade path for users, and
provides the ability for refresh/apply to handle the "upgrade" by
reading the old depends_on field and writing out the new
dependencies .
The inter-instance dependencies will be determined from the complete
reference graph, so that absolute addresses can be stored, rather than
just references within a module. The Dependencies are added to the node
in the same manner as state, i.e. via an "attacher" interface and
transformer. This is because dependencies must be calculated from the
graph itself, and not from the config.
The new StateDependencies are used in the DestroyEdgeTransformer
to connect both dependent destroyers, as well as creators depending on
another instance's destroyer.
The old code in DestroyEdgeTransformer may no longer be needed in the
long run, but that can be determined separately, since too many of the
tests start with an incomplete state and rely on the Dependencies being
determined from the configuration alone.
Fixes #18408
Fixes #20196
Fixes #21497
Fixes #21630
Fixes #23169
Fixes #22928
Closes #8617 (fixes the remaining issue described in the comments)
Might take care of #20823 (The linked issue sounds like a 0.11 bug. If it's still broken in 0.12 it would probably fall into this same category)