opentofu/terraform/transform_destroy_cbd.go
Martin Atkins a3403f2766 terraform: Ugly huge change to weave in new State and Plan types
Due to how often the state and plan types are referenced throughout
Terraform, there isn't a great way to switch them out gradually. As a
consequence, this huge commit gets us from the old world to a _compilable_
new world, but still has a large number of known test failures due to
key functionality being stubbed out.

The stubs here are for anything that interacts with providers, since we
now need to do the follow-up work to similarly replace the old
terraform.ResourceProvider interface with its replacement in the new
"providers" package. That work, along with work to fix the remaining
failing tests, will follow in subsequent commits.

The aim here was to replace all references to terraform.State and its
downstream types with states.State, terraform.Plan with plans.Plan,
state.State with statemgr.State, and switch to the new implementations of
the state and plan file formats. However, due to the number of times those
types are used, this also ended up affecting numerous other parts of core
such as terraform.Hook, the backend.Backend interface, and most of the CLI
commands.

Just as with 5861dbf3fc49b19587a31816eb06f511ab861bb4 before, I apologize
in advance to the person who inevitably just found this huge commit while
spelunking through the commit history.
2018-10-16 19:11:09 -07:00

269 lines
8.4 KiB
Go

package terraform
import (
"fmt"
"log"
"github.com/hashicorp/terraform/configs"
"github.com/hashicorp/terraform/dag"
"github.com/hashicorp/terraform/states"
)
// GraphNodeDestroyerCBD must be implemented by nodes that might be
// create-before-destroy destroyers.
type GraphNodeDestroyerCBD interface {
GraphNodeDestroyer
// CreateBeforeDestroy returns true if this node represents a node
// that is doing a CBD.
CreateBeforeDestroy() bool
// ModifyCreateBeforeDestroy is called when the CBD state of a node
// is changed dynamically. This can return an error if this isn't
// allowed.
ModifyCreateBeforeDestroy(bool) error
}
// GraphNodeAttachDestroyer is implemented by applyable nodes that have a
// companion destroy node. This allows the creation node to look up the status
// of the destroy node and determine if it needs to depose the existing state,
// or replace it.
// If a node is not marked as create-before-destroy in the configuration, but a
// dependency forces that status, only the destroy node will be aware of that
// status.
type GraphNodeAttachDestroyer interface {
// AttachDestroyNode takes a destroy node and saves a reference to that
// node in the receiver, so it can later check the status of
// CreateBeforeDestroy().
AttachDestroyNode(n GraphNodeDestroyerCBD)
}
// CBDEdgeTransformer modifies the edges of CBD nodes that went through
// the DestroyEdgeTransformer to have the right dependencies. There are
// two real tasks here:
//
// 1. With CBD, the destroy edge is inverted: the destroy depends on
// the creation.
//
// 2. A_d must depend on resources that depend on A. This is to enable
// the destroy to only happen once nodes that depend on A successfully
// update to A. Example: adding a web server updates the load balancer
// before deleting the old web server.
//
type CBDEdgeTransformer struct {
// Module and State are only needed to look up dependencies in
// any way possible. Either can be nil if not availabile.
Config *configs.Config
State *states.State
// If configuration is present then Schemas is required in order to
// obtain schema information from providers and provisioners so we can
// properly resolve implicit dependencies.
Schemas *Schemas
}
func (t *CBDEdgeTransformer) Transform(g *Graph) error {
// Go through and reverse any destroy edges
destroyMap := make(map[string][]dag.Vertex)
for _, v := range g.Vertices() {
dn, ok := v.(GraphNodeDestroyerCBD)
if !ok {
continue
}
if !dn.CreateBeforeDestroy() {
// If there are no CBD ancestors (dependent nodes), then we
// do nothing here.
if !t.hasCBDAncestor(g, v) {
continue
}
// If this isn't naturally a CBD node, this means that an ancestor is
// and we need to auto-upgrade this node to CBD. We do this because
// a CBD node depending on non-CBD will result in cycles. To avoid this,
// we always attempt to upgrade it.
log.Printf("[TRACE] CBDEdgeTransformer: forcing create_before_destroy on for %q (%T)", dag.VertexName(v), v)
if err := dn.ModifyCreateBeforeDestroy(true); err != nil {
return fmt.Errorf(
"%s: must have create before destroy enabled because "+
"a dependent resource has CBD enabled. However, when "+
"attempting to automatically do this, an error occurred: %s",
dag.VertexName(v), err)
}
}
// Find the destroy edge. There should only be one.
for _, e := range g.EdgesTo(v) {
// Not a destroy edge, ignore it
de, ok := e.(*DestroyEdge)
if !ok {
continue
}
log.Printf("[TRACE] CBDEdgeTransformer: inverting edge: %s => %s",
dag.VertexName(de.Source()), dag.VertexName(de.Target()))
// Found it! Invert.
g.RemoveEdge(de)
g.Connect(&DestroyEdge{S: de.Target(), T: de.Source()})
}
// If the address has an index, we strip that. Our depMap creation
// graph doesn't expand counts so we don't currently get _exact_
// dependencies. One day when we limit dependencies more exactly
// this will have to change. We have a test case covering this
// (depNonCBDCountBoth) so it'll be caught.
addr := dn.DestroyAddr()
key := addr.ContainingResource().String()
// Add this to the list of nodes that we need to fix up
// the edges for (step 2 above in the docs).
destroyMap[key] = append(destroyMap[key], v)
}
// If we have no CBD nodes, then our work here is done
if len(destroyMap) == 0 {
return nil
}
// We have CBD nodes. We now have to move on to the much more difficult
// task of connecting dependencies of the creation side of the destroy
// to the destruction node. The easiest way to explain this is an example:
//
// Given a pre-destroy dependence of: A => B
// And A has CBD set.
//
// The resulting graph should be: A => B => A_d
//
// They key here is that B happens before A is destroyed. This is to
// facilitate the primary purpose for CBD: making sure that downstreams
// are properly updated to avoid downtime before the resource is destroyed.
//
// We can't trust that the resource being destroyed or anything that
// depends on it is actually in our current graph so we make a new
// graph in order to determine those dependencies and add them in.
log.Printf("[TRACE] CBDEdgeTransformer: building graph to find dependencies...")
depMap, err := t.depMap(destroyMap)
if err != nil {
return err
}
// We now have the mapping of resource addresses to the destroy
// nodes they need to depend on. We now go through our own vertices to
// find any matching these addresses and make the connection.
for _, v := range g.Vertices() {
// We're looking for creators
rn, ok := v.(GraphNodeCreator)
if !ok {
continue
}
// Get the address
addr := rn.CreateAddr()
// If the address has an index, we strip that. Our depMap creation
// graph doesn't expand counts so we don't currently get _exact_
// dependencies. One day when we limit dependencies more exactly
// this will have to change. We have a test case covering this
// (depNonCBDCount) so it'll be caught.
key := addr.ContainingResource().String()
// If there is nothing this resource should depend on, ignore it
dns, ok := depMap[key]
if !ok {
continue
}
// We have nodes! Make the connection
for _, dn := range dns {
log.Printf("[TRACE] CBDEdgeTransformer: destroy depends on dependence: %s => %s",
dag.VertexName(dn), dag.VertexName(v))
g.Connect(dag.BasicEdge(dn, v))
}
}
return nil
}
func (t *CBDEdgeTransformer) depMap(destroyMap map[string][]dag.Vertex) (map[string][]dag.Vertex, error) {
// Build the graph of our config, this ensures that all resources
// are present in the graph.
g, diags := (&BasicGraphBuilder{
Steps: []GraphTransformer{
&FlatConfigTransformer{Config: t.Config},
&AttachResourceConfigTransformer{Config: t.Config},
&AttachStateTransformer{State: t.State},
&AttachSchemaTransformer{Schemas: t.Schemas},
&ReferenceTransformer{},
},
Name: "CBDEdgeTransformer",
}).Build(nil)
if diags.HasErrors() {
return nil, diags.Err()
}
// Using this graph, build the list of destroy nodes that each resource
// address should depend on. For example, when we find B, we map the
// address of B to A_d in the "depMap" variable below.
depMap := make(map[string][]dag.Vertex)
for _, v := range g.Vertices() {
// We're looking for resources.
rn, ok := v.(GraphNodeResource)
if !ok {
continue
}
// Get the address
addr := rn.ResourceAddr()
key := addr.String()
// Get the destroy nodes that are destroying this resource.
// If there aren't any, then we don't need to worry about
// any connections.
dns, ok := destroyMap[key]
if !ok {
continue
}
// Get the nodes that depend on this on. In the example above:
// finding B in A => B.
for _, v := range g.UpEdges(v).List() {
// We're looking for resources.
rn, ok := v.(GraphNodeResource)
if !ok {
continue
}
// Keep track of the destroy nodes that this address
// needs to depend on.
key := rn.ResourceAddr().String()
depMap[key] = append(depMap[key], dns...)
}
}
return depMap, nil
}
// hasCBDAncestor returns true if any ancestor (node that depends on this)
// has CBD set.
func (t *CBDEdgeTransformer) hasCBDAncestor(g *Graph, v dag.Vertex) bool {
s, _ := g.Ancestors(v)
if s == nil {
return true
}
for _, v := range s.List() {
dn, ok := v.(GraphNodeDestroyerCBD)
if !ok {
continue
}
if dn.CreateBeforeDestroy() {
// some ancestor is CreateBeforeDestroy, so we need to follow suit
return true
}
}
return false
}