member-console/internal/workflows/schedule_dormancy.go

package workflows

import (
	"context"
	"errors"
	"fmt"
	"log/slog"

	"go.temporal.io/api/serviceerror"
	"go.temporal.io/sdk/client"
)

// DormantPauseNote marks a schedule pause as made by an unconfigured boot.
// ResumeDormantSchedule resumes ONLY pauses carrying this exact note, so an
// operator's deliberate pause (any other note, including one the system
// overwrote) survives restarts. Changing this string strands schedules
// paused under the old wording — treat it as frozen.
const DormantPauseNote = "paused by member-console: integration not configured or disabled this boot; a configured boot resumes it automatically"

const dormantResumeNote = "resumed by member-console: integration configured"

// PauseLeftoverSchedule pauses scheduleID if it exists and is running.
// Integrations call this when they boot unconfigured/disabled: Temporal
// schedules are durable, so a schedule ensured by an earlier configured
// boot would otherwise keep firing workflows into a worker whose provider
// client has no configuration, failing every attempt at every interval,
// indefinitely. An absent schedule is a no-op; an already-paused schedule
// is left untouched whether the pause was ours or an operator's.
func PauseLeftoverSchedule(ctx context.Context, c client.Client, scheduleID string, logger *slog.Logger) error {
	handle := c.ScheduleClient().GetHandle(ctx, scheduleID)
	desc, err := handle.Describe(ctx)
	if err != nil {
		if isScheduleAbsent(err) {
			return nil
		}
		return fmt.Errorf("describe schedule %s: %w", scheduleID, err)
	}
	if !shouldPauseLeftover(desc.Schedule.State) {
		return nil
	}
	if err := handle.Pause(ctx, client.SchedulePauseOptions{Note: DormantPauseNote}); err != nil {
		return fmt.Errorf("pause schedule %s: %w", scheduleID, err)
	}
	logger.Info("paused leftover schedule of unconfigured integration",
		slog.String("scheduleID", scheduleID))
	return nil
}

// ResumeDormantSchedule unpauses scheduleID only when its pause note is
// DormantPauseNote — a pause made by PauseLeftoverSchedule on an earlier
// unconfigured boot. Integrations call this after ensuring their schedule
// on a configured boot. A pause carrying any other note is treated as
// operator intent and left in place (logged, so a configured-but-silent
// integration is diagnosable).
func ResumeDormantSchedule(ctx context.Context, c client.Client, scheduleID string, logger *slog.Logger) error {
	handle := c.ScheduleClient().GetHandle(ctx, scheduleID)
	desc, err := handle.Describe(ctx)
	if err != nil {
		if isScheduleAbsent(err) {
			return nil
		}
		return fmt.Errorf("describe schedule %s: %w", scheduleID, err)
	}
	state := desc.Schedule.State
	if state == nil || !state.Paused {
		return nil
	}
	if !shouldResumeDormant(state) {
		logger.Info("schedule is paused with a non-dormancy note; leaving it paused",
			slog.String("scheduleID", scheduleID),
			slog.String("note", state.Note))
		return nil
	}
	if err := handle.Unpause(ctx, client.ScheduleUnpauseOptions{Note: dormantResumeNote}); err != nil {
		return fmt.Errorf("unpause schedule %s: %w", scheduleID, err)
	}
	logger.Info("resumed schedule paused by an earlier unconfigured boot",
		slog.String("scheduleID", scheduleID))
	return nil
}

// shouldPauseLeftover and shouldResumeDormant hold the decision logic as
// pure functions: the Schedule API has no testsuite fake (same constraint
// as the fedwiki ScheduleManager tests), so these are the reachable,
// meaningful units.
func shouldPauseLeftover(state *client.ScheduleState) bool {
	return state == nil || !state.Paused
}

func shouldResumeDormant(state *client.ScheduleState) bool {
	return state != nil && state.Paused && state.Note == DormantPauseNote
}

// isScheduleAbsent mirrors the NotFound-only classification the schedule
// managers use: only Temporal's *serviceerror.NotFound means "no schedule";
// anything else is surfaced rather than swallowed.
func isScheduleAbsent(err error) bool {
	var nf *serviceerror.NotFound
	return errors.As(err, &nf)
}