//go:generate mapstructure-to-hcl2 -type Config package jdcloud import ( "fmt" "log" "strings" "sync" "time" "github.com/hashicorp/packer-plugin-sdk/common" "github.com/hashicorp/packer-plugin-sdk/multistep" "github.com/hashicorp/packer-plugin-sdk/template/interpolate" vm "github.com/jdcloud-api/jdcloud-sdk-go/services/vm/client" vpc "github.com/jdcloud-api/jdcloud-sdk-go/services/vpc/client" ) const ( FINE = 0 CONNECT_FAILED = "Client.Timeout exceeded" VM_PENDING = "pending" VM_RUNNING = "running" VM_STARTING = "starting" VM_STOPPING = "stopping" VM_STOPPED = "stopped" READY = "ready" BUILDER_ID = "hashicorp.jdcloud" ) var ( VmClient *vm.VmClient VpcClient *vpc.VpcClient Region string ) type Config struct { JDCloudCredentialConfig `mapstructure:",squash"` JDCloudInstanceSpecConfig `mapstructure:",squash"` common.PackerConfig `mapstructure:",squash"` ctx interpolate.Context } type Builder struct { config Config runner multistep.Runner } func Retry(timeout time.Duration, f RetryFunc) error { // These are used to pull the error out of the function; need a mutex to // avoid a data race. var resultErr error var resultErrMu sync.Mutex c := &StateChangeConf{ Pending: []string{"retryableerror"}, Target: []string{"success"}, Timeout: timeout, MinTimeout: 500 * time.Millisecond, Refresh: func() (interface{}, string, error) { rerr := f() resultErrMu.Lock() defer resultErrMu.Unlock() if rerr == nil { resultErr = nil return 42, "success", nil } resultErr = rerr.Err if rerr.Retryable { return 42, "retryableerror", nil } return nil, "quit", rerr.Err }, } _, waitErr := c.WaitForState() // Need to acquire the lock here to be able to avoid race using resultErr as // the return value resultErrMu.Lock() defer resultErrMu.Unlock() // resultErr may be nil because the wait timed out and resultErr was never // set; this is still an error if resultErr == nil { return waitErr } // resultErr takes precedence over waitErr if both are set because it is // more likely to be useful return resultErr } // RetryFunc is the function retried until it succeeds. type RetryFunc func() *RetryError // RetryError is the required return type of RetryFunc. It forces client code // to choose whether or not a given error is retryable. type RetryError struct { Err error Retryable bool } // RetryableError is a helper to create a RetryError that's retryable from a // given error. func RetryableError(err error) *RetryError { if err == nil { return nil } return &RetryError{Err: err, Retryable: true} } // NonRetryableError is a helper to create a RetryError that's _not_ retryable // from a given error. func NonRetryableError(err error) *RetryError { if err == nil { return nil } return &RetryError{Err: err, Retryable: false} } // WaitForState watches an object and waits for it to achieve the state // specified in the configuration using the specified Refresh() func, // waiting the number of seconds specified in the timeout configuration. // // If the Refresh function returns an error, exit immediately with that error. // // If the Refresh function returns a state other than the Target state or one // listed in Pending, return immediately with an error. // // If the Timeout is exceeded before reaching the Target state, return an // error. // // Otherwise, the result is the result of the first call to the Refresh function to // reach the target state. func (conf *StateChangeConf) WaitForState() (interface{}, error) { log.Printf("[DEBUG] Waiting for state to become: %s", conf.Target) notfoundTick := 0 targetOccurence := 0 // Set a default for times to check for not found if conf.NotFoundChecks == 0 { conf.NotFoundChecks = 20 } if conf.ContinuousTargetOccurence == 0 { conf.ContinuousTargetOccurence = 1 } type Result struct { Result interface{} State string Error error Done bool } // Read every result from the refresh loop, waiting for a positive result.Done. resCh := make(chan Result, 1) // cancellation channel for the refresh loop cancelCh := make(chan struct{}) result := Result{} go func() { defer close(resCh) time.Sleep(conf.Delay) // start with 0 delay for the first loop var wait time.Duration for { // store the last result resCh <- result // wait and watch for cancellation select { case <-cancelCh: return case <-time.After(wait): // first round had no wait if wait == 0 { wait = 100 * time.Millisecond } } res, currentState, err := conf.Refresh() result = Result{ Result: res, State: currentState, Error: err, } if err != nil { resCh <- result return } // If we're waiting for the absence of a thing, then return if res == nil && len(conf.Target) == 0 { targetOccurence++ if conf.ContinuousTargetOccurence == targetOccurence { result.Done = true resCh <- result return } continue } if res == nil { // If we didn't find the resource, check if we have been // not finding it for awhile, and if so, report an error. notfoundTick++ if notfoundTick > conf.NotFoundChecks { result.Error = &NotFoundError{ LastError: err, Retries: notfoundTick, } resCh <- result return } } else { // Reset the counter for when a resource isn't found notfoundTick = 0 found := false for _, allowed := range conf.Target { if currentState == allowed { found = true targetOccurence++ if conf.ContinuousTargetOccurence == targetOccurence { result.Done = true resCh <- result return } continue } } for _, allowed := range conf.Pending { if currentState == allowed { found = true targetOccurence = 0 break } } if !found && len(conf.Pending) > 0 { result.Error = &UnexpectedStateError{ LastError: err, State: result.State, ExpectedState: conf.Target, } resCh <- result return } } // Wait between refreshes using exponential backoff, except when // waiting for the target state to reoccur. if targetOccurence == 0 { wait *= 2 } // If a poll interval has been specified, choose that interval. // Otherwise bound the default value. if conf.PollInterval > 0 && conf.PollInterval < 180*time.Second { wait = conf.PollInterval } else { if wait < conf.MinTimeout { wait = conf.MinTimeout } else if wait > 10*time.Second { wait = 10 * time.Second } } log.Printf("[TRACE] Waiting %s before next try", wait) } }() // store the last value result from the refresh loop lastResult := Result{} timeout := time.After(conf.Timeout) for { select { case r, ok := <-resCh: // channel closed, so return the last result if !ok { return lastResult.Result, lastResult.Error } // we reached the intended state if r.Done { return r.Result, r.Error } // still waiting, store the last result lastResult = r case <-timeout: log.Printf("[WARN] WaitForState timeout after %s", conf.Timeout) log.Printf("[WARN] WaitForState starting %s refresh grace period", 30*time.Second) // cancel the goroutine and start our grace period timer close(cancelCh) timeout := time.After(30 * time.Second) // we need a for loop and a label to break on, because we may have // an extra response value to read, but still want to wait for the // channel to close. forSelect: for { select { case r, ok := <-resCh: if r.Done { // the last refresh loop reached the desired state return r.Result, r.Error } if !ok { // the goroutine returned break forSelect } // target state not reached, save the result for the // TimeoutError and wait for the channel to close lastResult = r case <-timeout: log.Println("[ERROR] WaitForState exceeded refresh grace period") break forSelect } } return nil, &TimeoutError{ LastError: lastResult.Error, LastState: lastResult.State, Timeout: conf.Timeout, ExpectedState: conf.Target, } } } } type StateChangeConf struct { Delay time.Duration // Wait this time before starting checks Pending []string // States that are "allowed" and will continue trying Refresh StateRefreshFunc // Refreshes the current state Target []string // Target state Timeout time.Duration // The amount of time to wait before timeout MinTimeout time.Duration // Smallest time to wait before refreshes PollInterval time.Duration // Override MinTimeout/backoff and only poll this often NotFoundChecks int // Number of times to allow not found // This is to work around inconsistent APIs ContinuousTargetOccurence int // Number of times the Target state has to occur continuously } type NotFoundError struct { LastError error LastRequest interface{} LastResponse interface{} Message string Retries int } func (e *NotFoundError) Error() string { if e.Message != "" { return e.Message } if e.Retries > 0 { return fmt.Sprintf("couldn't find resource (%d retries)", e.Retries) } return "couldn't find resource" } // UnexpectedStateError is returned when Refresh returns a state that's neither in Target nor Pending type UnexpectedStateError struct { LastError error State string ExpectedState []string } func (e *UnexpectedStateError) Error() string { return fmt.Sprintf( "unexpected state '%s', wanted target '%s'. last error: %s", e.State, strings.Join(e.ExpectedState, ", "), e.LastError, ) } // TimeoutError is returned when WaitForState times out type TimeoutError struct { LastError error LastState string Timeout time.Duration ExpectedState []string } func (e *TimeoutError) Error() string { expectedState := "resource to be gone" if len(e.ExpectedState) > 0 { expectedState = fmt.Sprintf("state to become '%s'", strings.Join(e.ExpectedState, ", ")) } extraInfo := make([]string, 0) if e.LastState != "" { extraInfo = append(extraInfo, fmt.Sprintf("last state: '%s'", e.LastState)) } if e.Timeout > 0 { extraInfo = append(extraInfo, fmt.Sprintf("timeout: %s", e.Timeout.String())) } suffix := "" if len(extraInfo) > 0 { suffix = fmt.Sprintf(" (%s)", strings.Join(extraInfo, ", ")) } if e.LastError != nil { return fmt.Sprintf("timeout while waiting for %s%s: %s", expectedState, suffix, e.LastError) } return fmt.Sprintf("timeout while waiting for %s%s", expectedState, suffix) } type StateRefreshFunc func() (result interface{}, state string, err error)