terraform/states/resource.go

247 lines
7.8 KiB
Go
Raw Normal View History

states: New package with modern models for Terraform state Our previous state models in the "terraform" package had a few limitations that are addressed here: - Instance attributes were stored as map[string]string with dot-separated keys representing traversals through a data structure. Now that we have a full type system, it's preferable to store it as a real data structure. - The existing state structures skipped over the "resource" concept and went straight to resource instance, requiring heuristics to decide whether a particular resource should appear as a single object or as a list of objects when used in configuration expressions. - Related to the previous point, the state models also used incorrect terminology where "ResourceState" was really a resource instance state and "InstanceState" was really the state of a particular remote object associated with an instance. These new models use the correct names for each of these, introducing the idea of a "ResourceInstanceObject" as the local record of a remote object associated with an instance. This is a first pass at fleshing out a new model for state. Undoubtedly there will be further iterations of this as we work on integrating these new models into the "terraform" package. These new model types no longer serve double-duty as a description of the JSON state file format, since they are for in-memory use only. A subsequent commit will introduce a separate package that deals with persisting state to files and reloading those files later.
2018-06-08 02:27:57 +02:00
package states
import (
"fmt"
"math/rand"
"time"
"github.com/hashicorp/terraform/addrs"
)
// Resource represents the state of a resource.
type Resource struct {
// Addr is the module-relative address for the resource this state object
// belongs to.
Addr addrs.Resource
// EachMode is the multi-instance mode currently in use for this resource,
// or NoEach if this is a single-instance resource. This dictates what
// type of value is returned when accessing this resource via expressions
// in the Terraform language.
EachMode EachMode
// Instances contains the potentially-multiple instances associated with
// this resource. This map can contain a mixture of different key types,
// but only the ones of InstanceKeyType are considered current.
Instances map[addrs.InstanceKey]*ResourceInstance
// ProviderConfig is the absolute address for the provider configuration that
// most recently managed this resource. This is used to connect a resource
// with a provider configuration when the resource configuration block is
// not available, such as if it has been removed from configuration
// altogether.
ProviderConfig addrs.AbsProviderConfig
}
// Instance returns the state for the instance with the given key, or nil
// if no such instance is tracked within the state.
func (rs *Resource) Instance(key addrs.InstanceKey) *ResourceInstance {
return rs.Instances[key]
}
// CreateInstance creates an instance and adds it to the resource
func (rs *Resource) CreateInstance(key addrs.InstanceKey) *ResourceInstance {
is := NewResourceInstance()
rs.Instances[key] = is
return is
}
states: New package with modern models for Terraform state Our previous state models in the "terraform" package had a few limitations that are addressed here: - Instance attributes were stored as map[string]string with dot-separated keys representing traversals through a data structure. Now that we have a full type system, it's preferable to store it as a real data structure. - The existing state structures skipped over the "resource" concept and went straight to resource instance, requiring heuristics to decide whether a particular resource should appear as a single object or as a list of objects when used in configuration expressions. - Related to the previous point, the state models also used incorrect terminology where "ResourceState" was really a resource instance state and "InstanceState" was really the state of a particular remote object associated with an instance. These new models use the correct names for each of these, introducing the idea of a "ResourceInstanceObject" as the local record of a remote object associated with an instance. This is a first pass at fleshing out a new model for state. Undoubtedly there will be further iterations of this as we work on integrating these new models into the "terraform" package. These new model types no longer serve double-duty as a description of the JSON state file format, since they are for in-memory use only. A subsequent commit will introduce a separate package that deals with persisting state to files and reloading those files later.
2018-06-08 02:27:57 +02:00
// EnsureInstance returns the state for the instance with the given key,
// creating a new empty state for it if one doesn't already exist.
//
// Because this may create and save a new state, it is considered to be
// a write operation.
func (rs *Resource) EnsureInstance(key addrs.InstanceKey) *ResourceInstance {
ret := rs.Instance(key)
if ret == nil {
ret = NewResourceInstance()
rs.Instances[key] = ret
}
return ret
}
// ResourceInstance represents the state of a particular instance of a resource.
type ResourceInstance struct {
// Current, if non-nil, is the remote object that is currently represented
// by the corresponding resource instance.
Current *ResourceInstanceObjectSrc
states: New package with modern models for Terraform state Our previous state models in the "terraform" package had a few limitations that are addressed here: - Instance attributes were stored as map[string]string with dot-separated keys representing traversals through a data structure. Now that we have a full type system, it's preferable to store it as a real data structure. - The existing state structures skipped over the "resource" concept and went straight to resource instance, requiring heuristics to decide whether a particular resource should appear as a single object or as a list of objects when used in configuration expressions. - Related to the previous point, the state models also used incorrect terminology where "ResourceState" was really a resource instance state and "InstanceState" was really the state of a particular remote object associated with an instance. These new models use the correct names for each of these, introducing the idea of a "ResourceInstanceObject" as the local record of a remote object associated with an instance. This is a first pass at fleshing out a new model for state. Undoubtedly there will be further iterations of this as we work on integrating these new models into the "terraform" package. These new model types no longer serve double-duty as a description of the JSON state file format, since they are for in-memory use only. A subsequent commit will introduce a separate package that deals with persisting state to files and reloading those files later.
2018-06-08 02:27:57 +02:00
// Deposed, if len > 0, contains any remote objects that were previously
// represented by the corresponding resource instance but have been
// replaced and are pending destruction due to the create_before_destroy
// lifecycle mode.
Deposed map[DeposedKey]*ResourceInstanceObjectSrc
states: New package with modern models for Terraform state Our previous state models in the "terraform" package had a few limitations that are addressed here: - Instance attributes were stored as map[string]string with dot-separated keys representing traversals through a data structure. Now that we have a full type system, it's preferable to store it as a real data structure. - The existing state structures skipped over the "resource" concept and went straight to resource instance, requiring heuristics to decide whether a particular resource should appear as a single object or as a list of objects when used in configuration expressions. - Related to the previous point, the state models also used incorrect terminology where "ResourceState" was really a resource instance state and "InstanceState" was really the state of a particular remote object associated with an instance. These new models use the correct names for each of these, introducing the idea of a "ResourceInstanceObject" as the local record of a remote object associated with an instance. This is a first pass at fleshing out a new model for state. Undoubtedly there will be further iterations of this as we work on integrating these new models into the "terraform" package. These new model types no longer serve double-duty as a description of the JSON state file format, since they are for in-memory use only. A subsequent commit will introduce a separate package that deals with persisting state to files and reloading those files later.
2018-06-08 02:27:57 +02:00
}
// NewResourceInstance constructs and returns a new ResourceInstance, ready to
// use.
func NewResourceInstance() *ResourceInstance {
return &ResourceInstance{
Deposed: map[DeposedKey]*ResourceInstanceObjectSrc{},
states: New package with modern models for Terraform state Our previous state models in the "terraform" package had a few limitations that are addressed here: - Instance attributes were stored as map[string]string with dot-separated keys representing traversals through a data structure. Now that we have a full type system, it's preferable to store it as a real data structure. - The existing state structures skipped over the "resource" concept and went straight to resource instance, requiring heuristics to decide whether a particular resource should appear as a single object or as a list of objects when used in configuration expressions. - Related to the previous point, the state models also used incorrect terminology where "ResourceState" was really a resource instance state and "InstanceState" was really the state of a particular remote object associated with an instance. These new models use the correct names for each of these, introducing the idea of a "ResourceInstanceObject" as the local record of a remote object associated with an instance. This is a first pass at fleshing out a new model for state. Undoubtedly there will be further iterations of this as we work on integrating these new models into the "terraform" package. These new model types no longer serve double-duty as a description of the JSON state file format, since they are for in-memory use only. A subsequent commit will introduce a separate package that deals with persisting state to files and reloading those files later.
2018-06-08 02:27:57 +02:00
}
}
// HasCurrent returns true if this resource instance has a "current"-generation
// object. Most instances do, but this can briefly be false during a
// create-before-destroy replace operation when the current has been deposed
// but its replacement has not yet been created.
func (i *ResourceInstance) HasCurrent() bool {
return i != nil && i.Current != nil
}
// HasDeposed returns true if this resource instance has a deposed object
// with the given key.
func (i *ResourceInstance) HasDeposed(key DeposedKey) bool {
return i != nil && i.Deposed[key] != nil
}
// HasAnyDeposed returns true if this resource instance has one or more
// deposed objects.
func (i *ResourceInstance) HasAnyDeposed() bool {
return i != nil && len(i.Deposed) > 0
}
// HasObjects returns true if this resource has any objects at all, whether
// current or deposed.
func (i *ResourceInstance) HasObjects() bool {
return i.Current != nil || len(i.Deposed) != 0
}
// deposeCurrentObject is part of the real implementation of
// SyncState.DeposeResourceInstanceObject. The exported method uses a lock
// to ensure that we can safely allocate an unused deposed key without
// collision.
core: Be more explicit in how we handle create_before_destroy Previously our handling of create_before_destroy -- and of deposed objects in particular -- was rather "implicit" and spread over various different subsystems. We'd quietly just destroy every deposed object during a destroy operation, without any user-visible plan to do so. Here we make things more explicit by tracking each deposed object individually by its pseudorandomly-allocated key. There are two different mechanisms at play here, building on the same concepts: - During a replace operation with create_before_destroy, we *pre-allocate* a DeposedKey to use for the prior object in the "apply" node and then pass that exact id to the destroy node, ensuring that we only destroy the single object we planned to destroy. In the happy path here the user never actually sees the allocated deposed key because we use it and then immediately destroy it within the same operation. However, that destroy may fail, which brings us to the second mechanism: - If any deposed objects are already present in state during _plan_, we insert a destroy change for them into the plan so that it's explicit to the user that we are going to destroy these additional objects, and then create an individual graph node for each one in DiffTransformer. The main motivation here is to be more careful in how we handle these destroys so that from a user's standpoint we never destroy something without the user knowing about it ahead of time. However, this new organization also hopefully makes the code itself a little easier to follow because the connection between the create and destroy steps of a Replace is reprseented in a single place (in DiffTransformer) and deposed instances each have their own explicit graph node rather than being secretly handled as part of the main instance-level graph node.
2018-09-20 21:30:52 +02:00
func (i *ResourceInstance) deposeCurrentObject(forceKey DeposedKey) DeposedKey {
states: New package with modern models for Terraform state Our previous state models in the "terraform" package had a few limitations that are addressed here: - Instance attributes were stored as map[string]string with dot-separated keys representing traversals through a data structure. Now that we have a full type system, it's preferable to store it as a real data structure. - The existing state structures skipped over the "resource" concept and went straight to resource instance, requiring heuristics to decide whether a particular resource should appear as a single object or as a list of objects when used in configuration expressions. - Related to the previous point, the state models also used incorrect terminology where "ResourceState" was really a resource instance state and "InstanceState" was really the state of a particular remote object associated with an instance. These new models use the correct names for each of these, introducing the idea of a "ResourceInstanceObject" as the local record of a remote object associated with an instance. This is a first pass at fleshing out a new model for state. Undoubtedly there will be further iterations of this as we work on integrating these new models into the "terraform" package. These new model types no longer serve double-duty as a description of the JSON state file format, since they are for in-memory use only. A subsequent commit will introduce a separate package that deals with persisting state to files and reloading those files later.
2018-06-08 02:27:57 +02:00
if !i.HasCurrent() {
return NotDeposed
}
core: Be more explicit in how we handle create_before_destroy Previously our handling of create_before_destroy -- and of deposed objects in particular -- was rather "implicit" and spread over various different subsystems. We'd quietly just destroy every deposed object during a destroy operation, without any user-visible plan to do so. Here we make things more explicit by tracking each deposed object individually by its pseudorandomly-allocated key. There are two different mechanisms at play here, building on the same concepts: - During a replace operation with create_before_destroy, we *pre-allocate* a DeposedKey to use for the prior object in the "apply" node and then pass that exact id to the destroy node, ensuring that we only destroy the single object we planned to destroy. In the happy path here the user never actually sees the allocated deposed key because we use it and then immediately destroy it within the same operation. However, that destroy may fail, which brings us to the second mechanism: - If any deposed objects are already present in state during _plan_, we insert a destroy change for them into the plan so that it's explicit to the user that we are going to destroy these additional objects, and then create an individual graph node for each one in DiffTransformer. The main motivation here is to be more careful in how we handle these destroys so that from a user's standpoint we never destroy something without the user knowing about it ahead of time. However, this new organization also hopefully makes the code itself a little easier to follow because the connection between the create and destroy steps of a Replace is reprseented in a single place (in DiffTransformer) and deposed instances each have their own explicit graph node rather than being secretly handled as part of the main instance-level graph node.
2018-09-20 21:30:52 +02:00
key := forceKey
if key == NotDeposed {
key = i.findUnusedDeposedKey()
} else {
if _, exists := i.Deposed[key]; exists {
panic(fmt.Sprintf("forced key %s is already in use", forceKey))
}
}
states: New package with modern models for Terraform state Our previous state models in the "terraform" package had a few limitations that are addressed here: - Instance attributes were stored as map[string]string with dot-separated keys representing traversals through a data structure. Now that we have a full type system, it's preferable to store it as a real data structure. - The existing state structures skipped over the "resource" concept and went straight to resource instance, requiring heuristics to decide whether a particular resource should appear as a single object or as a list of objects when used in configuration expressions. - Related to the previous point, the state models also used incorrect terminology where "ResourceState" was really a resource instance state and "InstanceState" was really the state of a particular remote object associated with an instance. These new models use the correct names for each of these, introducing the idea of a "ResourceInstanceObject" as the local record of a remote object associated with an instance. This is a first pass at fleshing out a new model for state. Undoubtedly there will be further iterations of this as we work on integrating these new models into the "terraform" package. These new model types no longer serve double-duty as a description of the JSON state file format, since they are for in-memory use only. A subsequent commit will introduce a separate package that deals with persisting state to files and reloading those files later.
2018-06-08 02:27:57 +02:00
i.Deposed[key] = i.Current
i.Current = nil
return key
}
// GetGeneration retrieves the object of the given generation from the
// ResourceInstance, or returns nil if there is no such object.
//
// If the given generation is nil or invalid, this method will panic.
func (i *ResourceInstance) GetGeneration(gen Generation) *ResourceInstanceObjectSrc {
states: New package with modern models for Terraform state Our previous state models in the "terraform" package had a few limitations that are addressed here: - Instance attributes were stored as map[string]string with dot-separated keys representing traversals through a data structure. Now that we have a full type system, it's preferable to store it as a real data structure. - The existing state structures skipped over the "resource" concept and went straight to resource instance, requiring heuristics to decide whether a particular resource should appear as a single object or as a list of objects when used in configuration expressions. - Related to the previous point, the state models also used incorrect terminology where "ResourceState" was really a resource instance state and "InstanceState" was really the state of a particular remote object associated with an instance. These new models use the correct names for each of these, introducing the idea of a "ResourceInstanceObject" as the local record of a remote object associated with an instance. This is a first pass at fleshing out a new model for state. Undoubtedly there will be further iterations of this as we work on integrating these new models into the "terraform" package. These new model types no longer serve double-duty as a description of the JSON state file format, since they are for in-memory use only. A subsequent commit will introduce a separate package that deals with persisting state to files and reloading those files later.
2018-06-08 02:27:57 +02:00
if gen == CurrentGen {
return i.Current
}
if dk, ok := gen.(DeposedKey); ok {
return i.Deposed[dk]
}
if gen == nil {
panic(fmt.Sprintf("get with nil Generation"))
}
// Should never fall out here, since the above covers all possible
// Generation values.
panic(fmt.Sprintf("get invalid Generation %#v", gen))
}
core: Be more explicit in how we handle create_before_destroy Previously our handling of create_before_destroy -- and of deposed objects in particular -- was rather "implicit" and spread over various different subsystems. We'd quietly just destroy every deposed object during a destroy operation, without any user-visible plan to do so. Here we make things more explicit by tracking each deposed object individually by its pseudorandomly-allocated key. There are two different mechanisms at play here, building on the same concepts: - During a replace operation with create_before_destroy, we *pre-allocate* a DeposedKey to use for the prior object in the "apply" node and then pass that exact id to the destroy node, ensuring that we only destroy the single object we planned to destroy. In the happy path here the user never actually sees the allocated deposed key because we use it and then immediately destroy it within the same operation. However, that destroy may fail, which brings us to the second mechanism: - If any deposed objects are already present in state during _plan_, we insert a destroy change for them into the plan so that it's explicit to the user that we are going to destroy these additional objects, and then create an individual graph node for each one in DiffTransformer. The main motivation here is to be more careful in how we handle these destroys so that from a user's standpoint we never destroy something without the user knowing about it ahead of time. However, this new organization also hopefully makes the code itself a little easier to follow because the connection between the create and destroy steps of a Replace is reprseented in a single place (in DiffTransformer) and deposed instances each have their own explicit graph node rather than being secretly handled as part of the main instance-level graph node.
2018-09-20 21:30:52 +02:00
// FindUnusedDeposedKey generates a unique DeposedKey that is guaranteed not to
// already be in use for this instance at the time of the call.
//
// Note that the validity of this result may change if new deposed keys are
// allocated before it is used. To avoid this risk, instead use the
// DeposeResourceInstanceObject method on the SyncState wrapper type, which
// allocates a key and uses it atomically.
func (i *ResourceInstance) FindUnusedDeposedKey() DeposedKey {
return i.findUnusedDeposedKey()
}
states: New package with modern models for Terraform state Our previous state models in the "terraform" package had a few limitations that are addressed here: - Instance attributes were stored as map[string]string with dot-separated keys representing traversals through a data structure. Now that we have a full type system, it's preferable to store it as a real data structure. - The existing state structures skipped over the "resource" concept and went straight to resource instance, requiring heuristics to decide whether a particular resource should appear as a single object or as a list of objects when used in configuration expressions. - Related to the previous point, the state models also used incorrect terminology where "ResourceState" was really a resource instance state and "InstanceState" was really the state of a particular remote object associated with an instance. These new models use the correct names for each of these, introducing the idea of a "ResourceInstanceObject" as the local record of a remote object associated with an instance. This is a first pass at fleshing out a new model for state. Undoubtedly there will be further iterations of this as we work on integrating these new models into the "terraform" package. These new model types no longer serve double-duty as a description of the JSON state file format, since they are for in-memory use only. A subsequent commit will introduce a separate package that deals with persisting state to files and reloading those files later.
2018-06-08 02:27:57 +02:00
// findUnusedDeposedKey generates a unique DeposedKey that is guaranteed not to
// already be in use for this instance.
func (i *ResourceInstance) findUnusedDeposedKey() DeposedKey {
for {
key := NewDeposedKey()
if _, exists := i.Deposed[key]; !exists {
return key
}
// Spin until we find a unique one. This shouldn't take long, because
// we have a 32-bit keyspace and there's rarely more than one deposed
// instance.
}
}
// EachMode specifies the multi-instance mode for a resource.
type EachMode rune
const (
NoEach EachMode = 0
EachList EachMode = 'L'
EachMap EachMode = 'M'
)
//go:generate go run golang.org/x/tools/cmd/stringer -type EachMode
states: New package with modern models for Terraform state Our previous state models in the "terraform" package had a few limitations that are addressed here: - Instance attributes were stored as map[string]string with dot-separated keys representing traversals through a data structure. Now that we have a full type system, it's preferable to store it as a real data structure. - The existing state structures skipped over the "resource" concept and went straight to resource instance, requiring heuristics to decide whether a particular resource should appear as a single object or as a list of objects when used in configuration expressions. - Related to the previous point, the state models also used incorrect terminology where "ResourceState" was really a resource instance state and "InstanceState" was really the state of a particular remote object associated with an instance. These new models use the correct names for each of these, introducing the idea of a "ResourceInstanceObject" as the local record of a remote object associated with an instance. This is a first pass at fleshing out a new model for state. Undoubtedly there will be further iterations of this as we work on integrating these new models into the "terraform" package. These new model types no longer serve double-duty as a description of the JSON state file format, since they are for in-memory use only. A subsequent commit will introduce a separate package that deals with persisting state to files and reloading those files later.
2018-06-08 02:27:57 +02:00
func eachModeForInstanceKey(key addrs.InstanceKey) EachMode {
switch key.(type) {
case addrs.IntKey:
return EachList
case addrs.StringKey:
return EachMap
default:
if key == addrs.NoKey {
return NoEach
}
panic(fmt.Sprintf("don't know an each mode for instance key %#v", key))
}
}
// DeposedKey is a 8-character hex string used to uniquely identify deposed
// instance objects in the state.
type DeposedKey string
// NotDeposed is a special invalid value of DeposedKey that is used to represent
// the absense of a deposed key. It must not be used as an actual deposed key.
const NotDeposed = DeposedKey("")
var deposedKeyRand = rand.New(rand.NewSource(time.Now().UnixNano()))
// NewDeposedKey generates a pseudo-random deposed key. Because of the short
// length of these keys, uniqueness is not a natural consequence and so the
// caller should test to see if the generated key is already in use and generate
// another if so, until a unique key is found.
func NewDeposedKey() DeposedKey {
v := deposedKeyRand.Uint32()
return DeposedKey(fmt.Sprintf("%08x", v))
}
func (k DeposedKey) String() string {
return string(k)
}
func (k DeposedKey) GoString() string {
ks := string(k)
switch {
case ks == "":
return "states.NotDeposed"
default:
return fmt.Sprintf("states.DeposedKey(%s)", ks)
}
}
// Generation is a helper method to convert a DeposedKey into a Generation.
// If the reciever is anything other than NotDeposed then the result is
// just the same value as a Generation. If the receiver is NotDeposed then
// the result is CurrentGen.
func (k DeposedKey) Generation() Generation {
if k == NotDeposed {
return CurrentGen
}
return k
}
states: New package with modern models for Terraform state Our previous state models in the "terraform" package had a few limitations that are addressed here: - Instance attributes were stored as map[string]string with dot-separated keys representing traversals through a data structure. Now that we have a full type system, it's preferable to store it as a real data structure. - The existing state structures skipped over the "resource" concept and went straight to resource instance, requiring heuristics to decide whether a particular resource should appear as a single object or as a list of objects when used in configuration expressions. - Related to the previous point, the state models also used incorrect terminology where "ResourceState" was really a resource instance state and "InstanceState" was really the state of a particular remote object associated with an instance. These new models use the correct names for each of these, introducing the idea of a "ResourceInstanceObject" as the local record of a remote object associated with an instance. This is a first pass at fleshing out a new model for state. Undoubtedly there will be further iterations of this as we work on integrating these new models into the "terraform" package. These new model types no longer serve double-duty as a description of the JSON state file format, since they are for in-memory use only. A subsequent commit will introduce a separate package that deals with persisting state to files and reloading those files later.
2018-06-08 02:27:57 +02:00
// generation is an implementation of Generation.
func (k DeposedKey) generation() {}