nektos/act · error

--health-start-interval cannot be negative

Error message

--health-start-interval cannot be negative

What it means

Error [90]: Before copying a tar stream into the container at destPath, act first sends a synthetic tar directory entry to '/' via CopyToContainer to create the destination directory. That mkdir-by-tar API call failed. Causes: container not running, daemon error during upload, or filesystem errors in the container (read-only rootfs, full disk).

Source

Thrown at pkg/container/docker_cli.go:598

	} else if haveHealthSettings {
		var probe []string
		if copts.healthCmd != "" {
			probe = []string{"CMD-SHELL", copts.healthCmd}
		}
		if copts.healthInterval < 0 {
			return nil, errors.New("--health-interval cannot be negative")
		}
		if copts.healthTimeout < 0 {
			return nil, errors.New("--health-timeout cannot be negative")
		}
		if copts.healthRetries < 0 {
			return nil, errors.New("--health-retries cannot be negative")
		}
		if copts.healthStartPeriod < 0 {
			return nil, errors.New("--health-start-period cannot be negative")
		}
		if copts.healthStartInterval < 0 {
			return nil, errors.New("--health-start-interval cannot be negative")
		}

		healthConfig = &container.HealthConfig{
			Test:          probe,
			Interval:      copts.healthInterval,
			Timeout:       copts.healthTimeout,
			StartPeriod:   copts.healthStartPeriod,
			StartInterval: copts.healthStartInterval,
			Retries:       copts.healthRetries,
		}
	}

	deviceRequests := copts.gpus.Value()
	if len(cdiDeviceNames) > 0 {
		cdiDeviceRequest := container.DeviceRequest{
			Driver:    "cdi",
			DeviceIDs: cdiDeviceNames,
		}

View on GitHub (pinned to 4f41128141)

Solutions

1. Check container state: docker ps -a --filter name=act — restart/freed? then inspect logs for the crash cause
2. Remove '--read-only' from container options if present
3. Check host disk space: df -h (overlay2 lives on host)
4. Re-run after cleaning leftover containers: docker rm -f $(docker ps -aq --filter name=act-)

Example fix

# before
act --container-options "--read-only"
# after
act --container-options ""

// Verify container is running and writable before copying
info, _ := cli.ContainerInspect(ctx, id)
if !info.State.Running { return errors.New("container not running") }
if info.HostConfig != nil && info.HostConfig.ReadonlyRootfs { return errors.New("rootfs read-only") }

if err != nil && strings.Contains(err.Error(), "failed to mkdir to copy content") {
	// inspect container state + logs; readonly rootfs and dead containers are the usual causes
}

When it happens

Trigger: cr.cli.CopyToContainer with the TypeDir tar header failing: target container exited/paused between exec and copy, container root filesystem is read-only (readonly rootfs in options), disk full inside container, or API/stream error to daemon.

Common situations: Actions like actions/checkout or file-copy steps against a job container that crashed earlier; --container-options with --read-only; overlay2 disk exhaustion on the host; container in 'Dead' state after OOM.

Related errors

• conflicting options: cannot specify both --link and per-netw
• --no-healthcheck conflicts with --health-* options
• conflicting options: cannot specify both --restart and --rm
• conflicting options: cannot attach both user-defined and non
• conflicting options: cannot specify both --network-alias and

AI-assisted analysis of nektos/act@4f41128141 (2026-08-15). Data as JSON: /api/errors/45acec902c7dc5d6. Report an issue: GitHub.