terraform/state/state.go

243 lines
6.5 KiB
Go
Raw Normal View History

2015-02-21 20:52:55 +01:00
package state
import (
"bytes"
"context"
"encoding/json"
"errors"
"fmt"
"math/rand"
"os"
"os/user"
"strings"
"text/template"
"time"
uuid "github.com/hashicorp/go-uuid"
2015-02-21 20:52:55 +01:00
"github.com/hashicorp/terraform/terraform"
"github.com/hashicorp/terraform/version"
2015-02-21 20:52:55 +01:00
)
var rngSource *rand.Rand
func init() {
rngSource = rand.New(rand.NewSource(time.Now().UnixNano()))
}
// State is the collection of all state interfaces.
type State interface {
StateReader
StateWriter
StateRefresher
StatePersister
Locker
}
2015-02-21 20:52:55 +01:00
// StateReader is the interface for things that can return a state. Retrieving
// the state here must not error. Loading the state fresh (an operation that
// can likely error) should be implemented by RefreshState. If a state hasn't
// been loaded yet, it is okay for State to return nil.
state: more robust handling of state Serial Previously we relied on a constellation of coincidences for everything to work out correctly with state serials. In particular, callers needed to be very careful about mutating states (or not) because many different bits of code shared pointers to the same objects. Here we move to a model where all of the state managers always use distinct instances of state, copied when WriteState is called. This means that they are truly a snapshot of the state as it was at that call, even if the caller goes on mutating the state that was passed. We also adjust the handling of serials so that the state managers ignore any serials in incoming states and instead just treat each Persist as the next version after what was most recently Refreshed. (An exception exists for when nothing has been refreshed, e.g. because we are writing a state to a location for the first time. In that case we _do_ trust the caller, since the given state is either a new state or it's a copy of something we're migrating from elsewhere with its state and lineage intact.) The intent here is to allow the rest of Terraform to not worry about serials and state identity, and instead just treat the state as a mutable structure. We'll just snapshot it occasionally, when WriteState is called, and deal with serials _only_ at persist time. This is intended as a more robust version of #15423, which was a quick hotfix to an issue that resulted from our previous slopping handling of state serials but arguably makes the problem worse by depending on an additional coincidental behavior of the local backend's apply implementation.
2017-07-05 21:34:30 +02:00
//
// Each caller of this function must get a distinct copy of the state, and
// it must also be distinct from any instance cached inside the reader, to
// ensure that mutations of the returned state will not affect the values
// returned to other callers.
2015-02-21 20:52:55 +01:00
type StateReader interface {
State() *terraform.State
}
// StateWriter is the interface that must be implemented by something that
// can write a state. Writing the state can be cached or in-memory, as
// full persistence should be implemented by StatePersister.
state: more robust handling of state Serial Previously we relied on a constellation of coincidences for everything to work out correctly with state serials. In particular, callers needed to be very careful about mutating states (or not) because many different bits of code shared pointers to the same objects. Here we move to a model where all of the state managers always use distinct instances of state, copied when WriteState is called. This means that they are truly a snapshot of the state as it was at that call, even if the caller goes on mutating the state that was passed. We also adjust the handling of serials so that the state managers ignore any serials in incoming states and instead just treat each Persist as the next version after what was most recently Refreshed. (An exception exists for when nothing has been refreshed, e.g. because we are writing a state to a location for the first time. In that case we _do_ trust the caller, since the given state is either a new state or it's a copy of something we're migrating from elsewhere with its state and lineage intact.) The intent here is to allow the rest of Terraform to not worry about serials and state identity, and instead just treat the state as a mutable structure. We'll just snapshot it occasionally, when WriteState is called, and deal with serials _only_ at persist time. This is intended as a more robust version of #15423, which was a quick hotfix to an issue that resulted from our previous slopping handling of state serials but arguably makes the problem worse by depending on an additional coincidental behavior of the local backend's apply implementation.
2017-07-05 21:34:30 +02:00
//
// Implementors that cache the state in memory _must_ take a copy of it
// before returning, since the caller may continue to modify it once
// control returns. The caller must ensure that the state instance is not
// concurrently modified _during_ the call, or behavior is undefined.
//
// If an object implements StatePersister in conjunction with StateReader
// then these methods must coordinate such that a subsequent read returns
// a copy of the most recent write, even if it has not yet been persisted.
2015-02-21 20:52:55 +01:00
type StateWriter interface {
WriteState(*terraform.State) error
}
// StateRefresher is the interface that is implemented by something that
// can load a state. This might be refreshing it from a remote location or
// it might simply be reloading it from disk.
type StateRefresher interface {
RefreshState() error
}
// StatePersister is implemented to truly persist a state. Whereas StateWriter
// is allowed to perhaps be caching in memory, PersistState must write the
// state to some durable storage.
state: more robust handling of state Serial Previously we relied on a constellation of coincidences for everything to work out correctly with state serials. In particular, callers needed to be very careful about mutating states (or not) because many different bits of code shared pointers to the same objects. Here we move to a model where all of the state managers always use distinct instances of state, copied when WriteState is called. This means that they are truly a snapshot of the state as it was at that call, even if the caller goes on mutating the state that was passed. We also adjust the handling of serials so that the state managers ignore any serials in incoming states and instead just treat each Persist as the next version after what was most recently Refreshed. (An exception exists for when nothing has been refreshed, e.g. because we are writing a state to a location for the first time. In that case we _do_ trust the caller, since the given state is either a new state or it's a copy of something we're migrating from elsewhere with its state and lineage intact.) The intent here is to allow the rest of Terraform to not worry about serials and state identity, and instead just treat the state as a mutable structure. We'll just snapshot it occasionally, when WriteState is called, and deal with serials _only_ at persist time. This is intended as a more robust version of #15423, which was a quick hotfix to an issue that resulted from our previous slopping handling of state serials but arguably makes the problem worse by depending on an additional coincidental behavior of the local backend's apply implementation.
2017-07-05 21:34:30 +02:00
//
// If an object implements StatePersister in conjunction with StateReader
// and/or StateRefresher then these methods must coordinate such that
// subsequent reads after a persist return an updated value.
2015-02-21 20:52:55 +01:00
type StatePersister interface {
PersistState() error
}
// Locker is implemented to lock state during command execution.
// The info parameter can be recorded with the lock, but the
// implementation should not depend in its value. The string returned by Lock
// is an ID corresponding to the lock acquired, and must be passed to Unlock to
// ensure that the correct lock is being released.
//
// Lock and Unlock may return an error value of type LockError which in turn
// can contain the LockInfo of a conflicting lock.
type Locker interface {
Lock(info *LockInfo) (string, error)
Unlock(id string) error
}
2017-04-03 20:26:05 +02:00
// test hook to verify that LockWithContext has attempted a lock
var postLockHook func()
// Lock the state, using the provided context for timeout and cancellation.
// This backs off slightly to an upper limit.
func LockWithContext(ctx context.Context, s State, info *LockInfo) (string, error) {
delay := time.Second
maxDelay := 16 * time.Second
for {
id, err := s.Lock(info)
if err == nil {
return id, nil
}
le, ok := err.(*LockError)
if !ok {
// not a lock error, so we can't retry
return "", err
}
if le == nil || le.Info == nil || le.Info.ID == "" {
// If we dont' have a complete LockError, there's something wrong with the lock
return "", err
}
2017-04-03 20:26:05 +02:00
if postLockHook != nil {
postLockHook()
}
// there's an existing lock, wait and try again
select {
case <-ctx.Done():
// return the last lock error with the info
return "", err
case <-time.After(delay):
if delay < maxDelay {
delay *= 2
}
}
}
}
// Generate a LockInfo structure, populating the required fields.
func NewLockInfo() *LockInfo {
// this doesn't need to be cryptographically secure, just unique.
// Using math/rand alleviates the need to check handle the read error.
// Use a uuid format to match other IDs used throughout Terraform.
buf := make([]byte, 16)
rngSource.Read(buf)
id, err := uuid.FormatUUID(buf)
if err != nil {
// this of course shouldn't happen
panic(err)
}
// don't error out on user and hostname, as we don't require them
userName := ""
if userInfo, err := user.Current(); err == nil {
userName = userInfo.Username
}
host, _ := os.Hostname()
info := &LockInfo{
ID: id,
Who: fmt.Sprintf("%s@%s", userName, host),
Version: version.Version,
Created: time.Now().UTC(),
}
return info
}
// LockInfo stores lock metadata.
//
// Only Operation and Info are required to be set by the caller of Lock.
type LockInfo struct {
// Unique ID for the lock. NewLockInfo provides a random ID, but this may
// be overridden by the lock implementation. The final value if ID will be
// returned by the call to Lock.
ID string
// Terraform operation, provided by the caller.
Operation string
// Extra information to store with the lock, provided by the caller.
Info string
// user@hostname when available
Who string
// Terraform version
Version string
// Time that the lock was taken.
Created time.Time
// Path to the state file when applicable. Set by the Lock implementation.
Path string
}
// Err returns the lock info formatted in an error
func (l *LockInfo) Err() error {
return errors.New(l.String())
}
// Marshal returns a string json representation of the LockInfo
func (l *LockInfo) Marshal() []byte {
js, err := json.Marshal(l)
if err != nil {
panic(err)
}
return js
}
// String return a multi-line string representation of LockInfo
func (l *LockInfo) String() string {
tmpl := `Lock Info:
ID: {{.ID}}
Path: {{.Path}}
Operation: {{.Operation}}
Who: {{.Who}}
Version: {{.Version}}
Created: {{.Created}}
Info: {{.Info}}
`
t := template.Must(template.New("LockInfo").Parse(tmpl))
var out bytes.Buffer
if err := t.Execute(&out, l); err != nil {
panic(err)
}
return out.String()
}
type LockError struct {
Info *LockInfo
Err error
}
func (e *LockError) Error() string {
var out []string
if e.Err != nil {
out = append(out, e.Err.Error())
}
if e.Info != nil {
out = append(out, e.Info.String())
}
return strings.Join(out, "\n")
}