zitadel/internal/eventstore/handler/handler_projection.go

package handler

import (
	"context"
	"runtime/debug"
	"sort"
	"sync"
	"time"

	"github.com/caos/logging"

	"github.com/caos/zitadel/internal/eventstore"
)

const systemID = "system"

type ProjectionHandlerConfig struct {
	HandlerConfig
	ProjectionName   string
	RequeueEvery     time.Duration
	RetryFailedAfter time.Duration
}

//Update updates the projection with the given statements
type Update func(context.Context, []*Statement, Reduce) (unexecutedStmts []*Statement, err error)

//Reduce reduces the given event to a statement
//which is used to update the projection
type Reduce func(eventstore.Event) (*Statement, error)

//Lock is used for mutex handling if needed on the projection
type Lock func(context.Context, time.Duration, string) <-chan error

//Unlock releases the mutex of the projection
type Unlock func(string) error

//SearchQuery generates the search query to lookup for events
type SearchQuery func() (query *eventstore.SearchQueryBuilder, queryLimit uint64, err error)

type ProjectionHandler struct {
	Handler

	requeueAfter time.Duration
	shouldBulk   *time.Timer

	retryFailedAfter time.Duration
	shouldPush       *time.Timer
	pushSet          bool

	ProjectionName string

	lockMu sync.Mutex
	stmts  []*Statement
}

func NewProjectionHandler(config ProjectionHandlerConfig) *ProjectionHandler {
	h := &ProjectionHandler{
		Handler:        NewHandler(config.HandlerConfig),
		ProjectionName: config.ProjectionName,
		requeueAfter:   config.RequeueEvery,
		// first bulk is instant on startup
		shouldBulk:       time.NewTimer(0),
		shouldPush:       time.NewTimer(0),
		retryFailedAfter: config.RetryFailedAfter,
	}

	//unitialized timer
	//https://github.com/golang/go/issues/12721
	<-h.shouldPush.C

	if config.RequeueEvery <= 0 {
		if !h.shouldBulk.Stop() {
			<-h.shouldBulk.C
		}
		logging.WithFields("projection", h.ProjectionName).Info("starting handler without requeue")
		return h
	} else if config.RequeueEvery < 500*time.Millisecond {
		logging.WithFields("projection", h.ProjectionName).Fatal("requeue every must be greater 500ms or <= 0")
	}
	logging.WithFields("projection", h.ProjectionName).Info("starting handler")
	return h
}

func (h *ProjectionHandler) ResetShouldBulk() {
	if h.requeueAfter > 0 {
		h.shouldBulk.Reset(h.requeueAfter)
	}
}

func (h *ProjectionHandler) triggerShouldPush(after time.Duration) {
	if !h.pushSet {
		h.pushSet = true
		h.shouldPush.Reset(after)
	}
}

//Process waits for several conditions:
// if context is canceled the function gracefully shuts down
// if an event occures it reduces the event
// if the internal timer expires the handler will check
// for unprocessed events on eventstore
func (h *ProjectionHandler) Process(
	ctx context.Context,
	reduce Reduce,
	update Update,
	lock Lock,
	unlock Unlock,
	query SearchQuery,
) {
	//handle panic
	defer func() {
		cause := recover()
		logging.WithFields("projection", h.ProjectionName, "cause", cause, "stack", string(debug.Stack())).Error("projection handler paniced")
	}()

	execBulk := h.prepareExecuteBulk(query, reduce, update)
	for {
		select {
		case <-ctx.Done():
			if h.pushSet {
				h.push(context.Background(), update, reduce)
			}
			h.shutdown()
			return
		case event := <-h.Handler.EventQueue:
			if err := h.processEvent(ctx, event, reduce); err != nil {
				logging.WithFields("projection", h.ProjectionName).WithError(err).Warn("process failed")
				continue
			}
			h.triggerShouldPush(0)
		case <-h.shouldBulk.C:
			h.bulk(ctx, lock, execBulk, unlock)
			h.ResetShouldBulk()
		default:
			//lower prio select with push
			select {
			case <-ctx.Done():
				if h.pushSet {
					h.push(context.Background(), update, reduce)
				}
				h.shutdown()
				return
			case event := <-h.Handler.EventQueue:
				if err := h.processEvent(ctx, event, reduce); err != nil {
					logging.WithFields("projection", h.ProjectionName).WithError(err).Warn("process failed")
					continue
				}
				h.triggerShouldPush(0)
			case <-h.shouldBulk.C:
				h.bulk(ctx, lock, execBulk, unlock)
				h.ResetShouldBulk()
			case <-h.shouldPush.C:
				h.push(ctx, update, reduce)
				h.ResetShouldBulk()
			}
		}
	}
}

func (h *ProjectionHandler) processEvent(
	ctx context.Context,
	event eventstore.Event,
	reduce Reduce,
) error {
	stmt, err := reduce(event)
	if err != nil {
		logging.New().WithError(err).Warn("unable to process event")
		return err
	}

	h.lockMu.Lock()
	defer h.lockMu.Unlock()

	h.stmts = append(h.stmts, stmt)

	return nil
}

func (h *ProjectionHandler) bulk(
	ctx context.Context,
	lock Lock,
	executeBulk executeBulk,
	unlock Unlock,
) error {
	ctx, cancel := context.WithCancel(ctx)
	defer cancel()

	errs := lock(ctx, h.requeueAfter, systemID)
	//wait until projection is locked
	if err, ok := <-errs; err != nil || !ok {
		logging.WithFields("projection", h.ProjectionName).OnError(err).Warn("initial lock failed")
		return err
	}
	go h.cancelOnErr(ctx, errs, cancel)

	execErr := executeBulk(ctx)
	logging.WithFields("projection", h.ProjectionName).OnError(execErr).Warn("unable to execute")

	unlockErr := unlock(systemID)
	logging.WithFields("projection", h.ProjectionName).OnError(unlockErr).Warn("unable to unlock")

	if execErr != nil {
		return execErr
	}

	return unlockErr
}

func (h *ProjectionHandler) cancelOnErr(ctx context.Context, errs <-chan error, cancel func()) {
	for {
		select {
		case err := <-errs:
			if err != nil {
				logging.WithFields("projection", h.ProjectionName).WithError(err).Warn("bulk canceled")
				cancel()
				return
			}
		case <-ctx.Done():
			cancel()
			return
		}

	}
}

type executeBulk func(ctx context.Context) error

func (h *ProjectionHandler) prepareExecuteBulk(
	query SearchQuery,
	reduce Reduce,
	update Update,
) executeBulk {
	return func(ctx context.Context) error {
		for {
			select {
			case <-ctx.Done():
				return nil
			default:
				hasLimitExeeded, err := h.fetchBulkStmts(ctx, query, reduce)
				if err != nil || len(h.stmts) == 0 {
					logging.WithFields("projection", h.ProjectionName).OnError(err).Warn("unable to fetch stmts")
					return err
				}

				if err = h.push(ctx, update, reduce); err != nil {
					return err
				}

				if !hasLimitExeeded {
					return nil
				}
			}
		}
	}
}

func (h *ProjectionHandler) fetchBulkStmts(
	ctx context.Context,
	query SearchQuery,
	reduce Reduce,
) (limitExeeded bool, err error) {
	eventQuery, eventsLimit, err := query()
	if err != nil {
		logging.WithFields("projection", h.ProjectionName).WithError(err).Warn("unable to create event query")
		return false, err
	}

	events, err := h.Eventstore.Filter(ctx, eventQuery)
	if err != nil {
		logging.WithFields("projection", h.ProjectionName).WithError(err).Info("Unable to bulk fetch events")
		return false, err
	}

	for _, event := range events {
		if err = h.processEvent(ctx, event, reduce); err != nil {
			logging.WithFields("projection", h.ProjectionName, "sequence", event.Sequence(), "instanceID", event.Aggregate().InstanceID).WithError(err).Warn("unable to process event in bulk")
			return false, err
		}
	}

	return len(events) == int(eventsLimit), nil
}

func (h *ProjectionHandler) push(
	ctx context.Context,
	update Update,
	reduce Reduce,
) (err error) {
	h.lockMu.Lock()
	defer h.lockMu.Unlock()

	sort.Slice(h.stmts, func(i, j int) bool {
		return h.stmts[i].Sequence < h.stmts[j].Sequence
	})

	h.stmts, err = update(ctx, h.stmts, reduce)
	h.pushSet = len(h.stmts) > 0

	if h.pushSet {
		h.triggerShouldPush(h.retryFailedAfter)
		return nil
	}

	h.shouldPush.Stop()

	return err
}

func (h *ProjectionHandler) shutdown() {
	h.lockMu.Lock()
	defer h.lockMu.Unlock()
	h.Sub.Unsubscribe()
	if !h.shouldBulk.Stop() {
		<-h.shouldBulk.C
	}
	if !h.shouldPush.Stop() {
		<-h.shouldPush.C
	}
	logging.New().Info("stop processing")
}