terraform-provider-stackitprivatepreview/stackit/internal/wait/postgresflexalpha/wait.go

package postgresflexalpha

import (
	"context"
	"errors"
	"fmt"
	"math"
	"net/http"
	"time"

	"github.com/hashicorp/terraform-plugin-log/tflog"

	postgresflex "tf-provider.git.onstackit.cloud/stackit-dev-tools/terraform-provider-stackitprivatepreview/pkg_gen/postgresflexalpha"

	"github.com/stackitcloud/stackit-sdk-go/core/oapierror"
	"github.com/stackitcloud/stackit-sdk-go/core/wait"
)

// "READY" "PENDING" "PROGRESSING" "FAILURE" "UNKNOWN" "TERMINATING"
const (
	InstanceStateEmpty       = ""
	InstanceStateProgressing = "PROGRESSING"
	InstanceStateSuccess     = "READY"
	InstanceStateFailed      = "FAILURE"
	InstanceStateTerminating = "TERMINATING"
	InstanceStateUnknown     = "UNKNOWN"
	InstanceStatePending     = "PENDING"
)

// APIClientInstanceInterface Interface needed for tests
type APIClientInstanceInterface interface {
	GetInstanceRequestExecute(ctx context.Context, projectId, region, instanceId string) (
		*postgresflex.GetInstanceResponse,
		error,
	)

	ListUsersRequestExecute(
		ctx context.Context,
		projectId string,
		region string,
		instanceId string,
	) (*postgresflex.ListUserResponse, error)
}

// APIClientUserInterface Interface needed for tests
type APIClientUserInterface interface {
	GetUserRequestExecute(ctx context.Context, projectId, region, instanceId string, userId int32) (
		*postgresflex.GetUserResponse,
		error,
	)

	GetDatabaseRequestExecute(
		ctx context.Context,
		projectId string,
		region string,
		instanceId string,
		databaseId int32,
	) (*postgresflex.GetDatabaseResponse, error)
}

// CreateInstanceWaitHandler will wait for instance creation
func CreateInstanceWaitHandler(
	ctx context.Context, a APIClientInstanceInterface, projectId, region,
	instanceId string,
) *wait.AsyncActionHandler[postgresflex.GetInstanceResponse] {
	instanceCreated := false
	var instanceGetResponse *postgresflex.GetInstanceResponse
	maxWait := time.Minute * 45
	startTime := time.Now()
	extendedTimeout := 0

	handler := wait.New(
		func() (waitFinished bool, response *postgresflex.GetInstanceResponse, err error) {
			if !instanceCreated {
				s, err := a.GetInstanceRequestExecute(ctx, projectId, region, instanceId)
				if err != nil {
					return false, nil, err
				}
				if s == nil || s.Id == nil || *s.Id != instanceId || s.Status == nil {
					return false, nil, nil
				}
				tflog.Debug(
					ctx, "waiting for instance ready", map[string]interface{}{
						"status": *s.Status,
					},
				)
				switch *s.Status {
				default:
					return true, s, fmt.Errorf("instance with id %s has unexpected status %s", instanceId, *s.Status)
				case InstanceStateEmpty:
					return false, nil, nil
				case InstanceStatePending:
					return false, nil, nil
				case InstanceStateUnknown:
					return false, nil, nil
				case InstanceStateProgressing:
					if time.Since(startTime) < maxWait {
						return false, nil, nil
					}
					tflog.Warn(
						ctx,
						fmt.Sprintf(
							"Wait handler still got status %s after %v for instance: %s",
							InstanceStateProgressing,
							maxWait,
							instanceId,
						),
					)
					if extendedTimeout < 3 {
						maxWait += time.Minute * 5
						extendedTimeout++
						if *s.Network.AccessScope == "SNA" {
							ready := true
							if s.Network == nil || s.Network.InstanceAddress == nil {
								tflog.Warn(ctx, "Waiting for instance_address")
								ready = false
							}
							if s.Network.RouterAddress == nil {
								tflog.Warn(ctx, "Waiting for router_address")
								ready = false
							}
							if !ready {
								return false, nil, nil
							}
						}
						if s.IsDeletable == nil {
							tflog.Warn(ctx, "Waiting for is_deletable")
							return false, nil, nil
						}
					}

					instanceCreated = true
					instanceGetResponse = s
				case InstanceStateSuccess:
					if s.Network != nil && s.Network.AccessScope != nil && *s.Network.AccessScope == "SNA" {
						if s.Network.InstanceAddress == nil {
							tflog.Warn(ctx, "Waiting for instance_address")
							return false, nil, nil
						}
						if s.Network.RouterAddress == nil {
							tflog.Warn(ctx, "Waiting for router_address")
							return false, nil, nil
						}
					}
					instanceCreated = true
					instanceGetResponse = s
				case InstanceStateFailed:
					tflog.Warn(ctx, fmt.Sprintf("Wait handler got status FAILURE for instance: %s", instanceId))
					return false, nil, nil
					// API responds with FAILURE for some seconds and then the instance goes to READY
					// return true, s, fmt.Errorf("create failed for instance with id %s", instanceId)
				}
			}

			tflog.Info(ctx, "Waiting for instance (calling list users")
			// 	// User operations aren't available right after an instance is deemed successful
			//			// To check if they are, perform a users request
			_, err = a.ListUsersRequestExecute(ctx, projectId, region, instanceId)
			if err == nil {
				return true, instanceGetResponse, nil
			}
			oapiErr, ok := err.(*oapierror.GenericOpenAPIError) // nolint:errorlint //complaining that error.As should be used to catch wrapped errors, but this error should not be wrapped
			if !ok {
				return false, nil, err
			}
			// TODO: refactor and cooperate with api guys to mitigate
			if oapiErr.StatusCode < 500 {
				return true, instanceGetResponse, fmt.Errorf(
					"users request after instance creation returned %d status code",
					oapiErr.StatusCode,
				)
			}
			return false, nil, nil
		},
	)
	// Sleep before wait is set because sometimes API returns 404 right after creation request
	handler.SetTimeout(90 * time.Minute).SetSleepBeforeWait(30 * time.Second)
	return handler
}

// PartialUpdateInstanceWaitHandler will wait for instance update
func PartialUpdateInstanceWaitHandler(
	ctx context.Context, a APIClientInstanceInterface, projectId, region,
	instanceId string,
) *wait.AsyncActionHandler[postgresflex.GetInstanceResponse] {
	handler := wait.New(
		func() (waitFinished bool, response *postgresflex.GetInstanceResponse, err error) {
			s, err := a.GetInstanceRequestExecute(ctx, projectId, region, instanceId)
			if err != nil {
				return false, nil, err
			}
			if s == nil || s.Id == nil || *s.Id != instanceId || s.Status == nil {
				return false, nil, nil
			}
			switch *s.Status {
			default:
				return true, s, fmt.Errorf("instance with id %s has unexpected status %s", instanceId, *s.Status)
			case InstanceStateEmpty:
				return false, nil, nil
			case InstanceStatePending:
				return false, nil, nil
			case InstanceStateProgressing:
				return false, nil, nil
			case InstanceStateSuccess:
				return true, s, nil
			case InstanceStateTerminating:
				return false, nil, nil
			case InstanceStateUnknown:
				return false, nil, nil
			case InstanceStateFailed:
				return true, s, fmt.Errorf("update got status FAILURE for instance with id %s", instanceId)
			}
		},
	)
	handler.SetTimeout(45 * time.Minute).SetSleepBeforeWait(30 * time.Second)
	return handler
}

// GetUserByIdWaitHandler will wait for instance creation
func GetUserByIdWaitHandler(
	ctx context.Context,
	a APIClientUserInterface,
	projectId, instanceId, region string,
	userId int64,
) *wait.AsyncActionHandler[postgresflex.GetUserResponse] {
	handler := wait.New(
		func() (waitFinished bool, response *postgresflex.GetUserResponse, err error) {
			if userId > math.MaxInt32 {
				return false, nil, fmt.Errorf("userId value is too big for int32")
			}
			userId32 := int32(userId) //nolint:gosec // we need to convert databaseId to int32 because API expects int32
			s, err := a.GetUserRequestExecute(ctx, projectId, region, instanceId, userId32)
			if err != nil {
				var oapiErr *oapierror.GenericOpenAPIError
				ok := errors.As(err, &oapiErr)
				if !ok {
					return false, nil, fmt.Errorf("could not convert error to oapierror.GenericOpenAPIError")
				}
				switch oapiErr.StatusCode {
				case http.StatusBadGateway, http.StatusGatewayTimeout, http.StatusServiceUnavailable:
				case http.StatusNotFound:
					tflog.Warn(
						ctx, "api responded with status", map[string]interface{}{
							"status": oapiErr.StatusCode,
						},
					)
					return false, nil, nil
				default:
					return false, nil, err
				}
			}
			return true, s, nil
		},
	)
	return handler
}

// GetDatabaseByIdWaitHandler will wait for instance creation
func GetDatabaseByIdWaitHandler(
	ctx context.Context,
	a APIClientUserInterface,
	projectId, instanceId, region string,
	databaseId int64,
) *wait.AsyncActionHandler[postgresflex.GetDatabaseResponse] {
	handler := wait.New(
		func() (waitFinished bool, response *postgresflex.GetDatabaseResponse, err error) {
			dbId32 := int32(databaseId) //nolint:gosec // we need to convert databaseId to int32 because API expects int32
			s, err := a.GetDatabaseRequestExecute(ctx, projectId, region, instanceId, dbId32)
			if err != nil {
				var oapiErr *oapierror.GenericOpenAPIError
				ok := errors.As(err, &oapiErr)
				if !ok {
					return false, nil, fmt.Errorf("could not convert error to oapierror.GenericOpenAPIError")
				}
				switch oapiErr.StatusCode {
				case http.StatusBadGateway, http.StatusGatewayTimeout, http.StatusServiceUnavailable:
					tflog.Warn(
						ctx, "api responded with 50[2,3,4] status", map[string]interface{}{
							"status": oapiErr.StatusCode,
						},
					)
					return false, nil, nil
				case http.StatusNotFound:
					tflog.Warn(
						ctx, "api responded with 404 status", map[string]interface{}{
							"status": oapiErr.StatusCode,
						},
					)
					return false, nil, nil
				default:
					return false, nil, err
				}
			}
			return true, s, nil
		},
	)
	return handler
}