mesh-host/internal/apply/apply.go

// Package apply makes a machine match a declaration.
//
// Three properties, each following a recorded decision, and each of them the difference
// between this and a script that writes files:
//
//   - A failed step fails the apply (novox/hq ADR 0010). Not "logs and continues": a partial
//     apply that reports success is the mesh's most expensive shape.
//   - Every applier READS BACK. Setting a value is not evidence the value took.
//   - What was applied is recorded after it works, never before (ADR 0018). A failed apply
//     leaves the machine in whatever state it reached, and nothing must claim otherwise.
package apply

import (
	"context"
	"crypto/sha256"
	"errors"
	"fmt"
	"os"
	"os/exec"
	"path/filepath"
	"sort"
	"strconv"
	"strings"
	"time"

	"github.com/novox/mesh-host/internal/declaration"
	"github.com/novox/mesh-host/internal/store"
	"github.com/novox/mesh-host/internal/system"
)

// Runner executes a command. The real one is used everywhere outside unit tests; behaviour
// against a real system is tested alongside rather than mocked (novox/hq ADR 0017).
type Runner = system.Runner

// Outcome is what happened to one resource.
type Outcome struct {
	ID     string `json:"id"`
	Type   string `json:"type"`
	Target string `json:"target"`
	Action string `json:"action"` // created · updated · unchanged · removed
	Detail string `json:"detail,omitempty"`
}

// Report is what an apply did, in the order it did it.
type Report struct {
	Outcomes []Outcome `json:"outcomes"`
}

// Changed reports whether anything about the machine actually moved. An apply that changed
// nothing is the ordinary steady state, and saying so is not the same as saying it failed.
func (r Report) Changed() bool {
	for _, o := range r.Outcomes {
		if o.Action != "unchanged" {
			return true
		}
	}
	return false
}

// Error is a failure part-way through, carrying what had already been done.
//
// The outcomes matter as much as the message: the machine is in whatever state the apply
// reached, and the only honest thing to hand back is the list of what did happen.
type Error struct {
	Resource string
	Err      error
	Done     Report
}

func (e *Error) Error() string {
	return fmt.Sprintf("applying %q: %v\n\n%d resource(s) were applied before this and remain; "+
		"the machine is in whatever state that left it.", e.Resource, e.Err, len(e.Done.Outcomes))
}

func (e *Error) Unwrap() error { return e.Err }

// Apply makes the machine match the declaration, and returns what it did.
//
// Removal happens FIRST, and the order is not arbitrary. A resource that leaves a declaration
// while another arrives at the same path is an ordinary rename: removing afterwards would
// delete the file that had just been written. Removing first risks losing the old state if the
// apply then fails — a recovery concern, where the other is a correctness one.
func Apply(
	ctx context.Context,
	sys system.System,
	d *declaration.Declaration,
	known store.State,
	run Runner,
	log func(string),
) (Report, store.State, error) {
	if log == nil {
		log = func(string) {}
	}
	report := Report{}

	declared := map[string]bool{}
	for _, r := range d.Resources {
		declared[r.Identity()] = true
	}

	for _, orphan := range known.Orphans(declared) {
		action, detail, err := remove(ctx, sys, orphan, run)
		if err != nil {
			return report, known, &Error{Resource: orphan.ID, Err: err, Done: report}
		}
		known.Forget(orphan.ID)
		report.Outcomes = append(report.Outcomes, Outcome{
			ID: orphan.ID, Type: orphan.Type, Target: orphan.Target,
			Action: action, Detail: detail,
		})
		log(fmt.Sprintf("  %s %s (%s)", action, orphan.ID, orphan.Target))
	}

	for _, resource := range d.Resources {
		outcome, err := applyOne(ctx, sys, resource, run)
		if err != nil {
			return report, known, &Error{Resource: resource.Identity(), Err: err, Done: report}
		}

		// Only now. The record follows the fact, never leads it.
		known.Record(store.Applied{
			ID: resource.Identity(), Type: string(resource.Kind()),
			Target: outcome.Target, AppliedAt: time.Now().UTC(),
		})
		report.Outcomes = append(report.Outcomes, outcome)
		if outcome.Action != "unchanged" {
			log(fmt.Sprintf("  %s %s (%s)", outcome.Action, outcome.ID, outcome.Target))
		}
	}
	return report, known, nil
}

func applyOne(ctx context.Context, sys system.System, r declaration.Resource, run Runner) (Outcome, error) {
	switch res := r.(type) {
	case *declaration.Directory:
		return applyDirectory(res)
	case *declaration.File:
		return applyFile(res)
	case *declaration.Service:
		return applyService(ctx, sys, res, run)
	case *declaration.Package:
		return applyPackage(ctx, sys, res, run)
	case *declaration.Container:
		return applyContainer(ctx, res, run)
	case *declaration.Action:
		return applyAction(ctx, res, run)
	default:
		// Unreachable: the declaration refused this already. Present because "unreachable"
		// stops being true the moment someone adds a kind and forgets this switch.
		return Outcome{}, fmt.Errorf("no applier for type %q", r.Kind())
	}
}

// begin starts an outcome from any resource, so the three facts a report needs are read from
// the resource itself rather than restated by each applier.
func begin(r declaration.Resource) Outcome {
	return Outcome{ID: r.Identity(), Type: string(r.Kind()), Target: r.Target()}
}

func modeOf(spec string, fallback os.FileMode) (os.FileMode, error) {
	if spec == "" {
		return fallback, nil
	}
	parsed, err := strconv.ParseUint(spec, 8, 32)
	if err != nil {
		return 0, fmt.Errorf("mode %q: %w", spec, err)
	}
	return os.FileMode(parsed), nil
}

func applyDirectory(r *declaration.Directory) (Outcome, error) {
	out := begin(r)
	mode, err := modeOf(r.Mode, 0o755)
	if err != nil {
		return out, err
	}

	before, err := os.Stat(r.Path)
	existed := err == nil
	if err != nil && !errors.Is(err, os.ErrNotExist) {
		return out, err
	}
	if existed && !before.IsDir() {
		return out, fmt.Errorf("%s exists and is not a directory", r.Path)
	}

	if !existed {
		if err := os.MkdirAll(r.Path, mode); err != nil {
			return out, err
		}
	}
	// Set explicitly even when it existed: MkdirAll applies the mode only on creation, and a
	// permission set at creation is not a permission maintained — a lesson this repository
	// already paid for once, with world-readable environment files.
	if err := os.Chmod(r.Path, mode); err != nil {
		return out, err
	}

	// Read back.
	after, err := os.Stat(r.Path)
	if err != nil {
		return out, fmt.Errorf("made %s and cannot stat it: %w", r.Path, err)
	}
	if !after.IsDir() {
		return out, fmt.Errorf("%s is not a directory after applying", r.Path)
	}
	if after.Mode().Perm() != mode.Perm() {
		return out, fmt.Errorf("%s is mode %o after setting %o", r.Path, after.Mode().Perm(), mode.Perm())
	}

	out.Action = "unchanged"
	if !existed {
		out.Action = "created"
	} else if before.Mode().Perm() != mode.Perm() {
		out.Action = "updated"
		out.Detail = fmt.Sprintf("mode %o to %o", before.Mode().Perm(), mode.Perm())
	}
	return out, nil
}

func applyFile(r *declaration.File) (Outcome, error) {
	out := begin(r)
	mode, err := modeOf(r.Mode, 0o644)
	if err != nil {
		return out, err
	}

	existing, readErr := os.ReadFile(r.Path)
	existed := readErr == nil
	if readErr != nil && !errors.Is(readErr, os.ErrNotExist) {
		return out, readErr
	}

	var beforeMode os.FileMode
	if existed {
		if info, err := os.Stat(r.Path); err == nil {
			beforeMode = info.Mode().Perm()
		}
	}

	contentSame := existed && string(existing) == r.Content
	modeSame := existed && beforeMode == mode.Perm()

	if !contentSame {
		if err := os.MkdirAll(filepath.Dir(r.Path), 0o755); err != nil {
			return out, err
		}
		if err := writeAtomically(r.Path, []byte(r.Content), mode); err != nil {
			return out, err
		}
	} else if !modeSame {
		if err := os.Chmod(r.Path, mode); err != nil {
			return out, err
		}
	}

	// Read back — the file, not the call that wrote it.
	written, err := os.ReadFile(r.Path)
	if err != nil {
		return out, fmt.Errorf("wrote %s and cannot read it back: %w", r.Path, err)
	}
	if string(written) != r.Content {
		return out, fmt.Errorf("%s does not contain what was declared after writing it", r.Path)
	}
	info, err := os.Stat(r.Path)
	if err != nil {
		return out, err
	}
	if info.Mode().Perm() != mode.Perm() {
		return out, fmt.Errorf("%s is mode %o after setting %o", r.Path, info.Mode().Perm(), mode.Perm())
	}

	switch {
	case !existed:
		out.Action = "created"
	case !contentSame && !modeSame:
		out.Action = "updated"
		out.Detail = "content and mode"
	case !contentSame:
		out.Action = "updated"
		out.Detail = "content"
	case !modeSame:
		out.Action = "updated"
		out.Detail = fmt.Sprintf("mode %o to %o", beforeMode, mode.Perm())
	default:
		out.Action = "unchanged"
	}
	return out, nil
}

// writeAtomically writes through a temporary file in the same directory.
//
// A reader of a managed file must never see half of one. The mesh's own configuration is read
// by daemons that reload on change, so a torn write is a service reading a truncated config.
func writeAtomically(path string, content []byte, mode os.FileMode) error {
	tmp, err := os.CreateTemp(filepath.Dir(path), ".mesh-host-*")
	if err != nil {
		return err
	}
	defer os.Remove(tmp.Name())

	if _, err := tmp.Write(content); err != nil {
		tmp.Close()
		return err
	}
	if err := tmp.Sync(); err != nil {
		tmp.Close()
		return err
	}
	if err := tmp.Close(); err != nil {
		return err
	}
	if err := os.Chmod(tmp.Name(), mode); err != nil {
		return err
	}
	return os.Rename(tmp.Name(), path)
}

func applyService(ctx context.Context, sys system.System, r *declaration.Service, run Runner) (Outcome, error) {
	out := begin(r)
	var changes []string

	// Boot first. A unit asked to be running and enabled should survive this apply failing
	// half way in the more useful direction: enabled-and-stopped comes back at the next boot,
	// where running-and-disabled does not.
	if r.Boot != "" {
		bootBefore, err := sys.ServiceBoot(ctx, run, r.Unit)
		if err != nil {
			return out, err
		}
		if bootBefore != r.Boot {
			if err := sys.SetServiceBoot(ctx, run, r.Unit, r.Boot); err != nil {
				return out, fmt.Errorf("setting %s to %s at boot: %w", r.Unit, r.Boot, err)
			}
			bootAfter, err := sys.ServiceBoot(ctx, run, r.Unit)
			if err != nil {
				return out, err
			}
			if bootAfter != r.Boot {
				return out, fmt.Errorf(
					"%s was asked to be %s at boot and is %s", r.Unit, r.Boot, bootAfter)
			}
			changes = append(changes, "boot "+bootBefore+" to "+bootAfter)
		}
	}

	before, err := sys.ServiceState(ctx, run, r.Unit)
	if err != nil {
		return out, err
	}
	if before != r.State {
		if err := sys.SetServiceState(ctx, run, r.Unit, r.State); err != nil {
			return out, fmt.Errorf("setting %s to %s: %w", r.Unit, r.State, err)
		}

		// Read back. A service manager accepting a command says the transaction was accepted,
		// not that the unit is running — one that starts and immediately dies satisfies it.
		after, err := sys.ServiceState(ctx, run, r.Unit)
		if err != nil {
			return out, err
		}
		if after != r.State {
			return out, fmt.Errorf("%s was asked to be %s and is %s", r.Unit, r.State, after)
		}
		changes = append(changes, before+" to "+after)
	}

	if len(changes) == 0 {
		out.Action = "unchanged"
		out.Detail = before
		return out, nil
	}
	out.Action = "updated"
	out.Detail = strings.Join(changes, ", ")
	return out, nil
}

// remove undoes one resource the host applied and the declaration no longer names, and reports
// what it actually did.
//
// Only ever called for something in the store, which is what bounds it: the host is
// authoritative over its own footprint and inert everywhere else (novox/hq ADR 0005).
//
// It returns the action rather than assuming "removed", because for half the vocabulary the
// honest word is "forgotten". A host that reported a package removed when it left the package
// installed would be describing an effect it declined to have.
func remove(ctx context.Context, sys system.System, a store.Applied, run Runner) (string, string, error) {
	switch declaration.Type(a.Type) {
	case declaration.TypeFile, declaration.TypeDirectory:
		if err := os.RemoveAll(a.Target); err != nil {
			return "", "", err
		}
		if _, err := os.Stat(a.Target); !errors.Is(err, os.ErrNotExist) {
			return "", "", fmt.Errorf("%s is still there after removing it", a.Target)
		}
		return "removed", "no longer declared", nil

	case declaration.TypeService:
		// A unit that is no longer declared is stopped, not deleted. The host did not install
		// it and does not own the unit file — only the state it put the unit into.
		//
		// A unit that no longer EXISTS is already in the state removal is trying to reach, and
		// saying so matters: stopping it fails, and a failure here fails the whole apply. A
		// host holding a record of an uninstalled unit would then be unable to apply anything,
		// ever, with no way out but editing its state by hand. Removal is idempotent for the
		// same reason `os.RemoveAll` is.
		if _, err := sys.ServiceState(ctx, run, a.Target); err != nil {
			if strings.Contains(err.Error(), "does not exist on this machine") {
				return "forgotten", "the unit no longer exists", nil
			}
			return "", "", err
		}
		if err := sys.SetServiceState(ctx, run, a.Target, "stopped"); err != nil {
			return "", "", fmt.Errorf("stopping %s: %w", a.Target, err)
		}
		return "removed", "stopped; the unit file is not the host's to delete", nil

	case declaration.TypeContainer:
		// The host CREATED this one, so the host removes it. That is the line: it removes what
		// it made and leaves what it merely configured.
		if _, err := run(ctx, "docker", "rm", "-f", a.Target); err != nil {
			// Already gone is the state removal wants. Anything else is a real failure.
			if _, alive := containerState(ctx, a.Target, run); alive == nil {
				return "", "", fmt.Errorf("removing container %s: %w", a.Target, err)
			}
		}
		if _, err := containerState(ctx, a.Target, run); err == nil {
			return "", "", fmt.Errorf("container %s is still there after removing it", a.Target)
		}
		return "removed", "no longer declared", nil

	case declaration.TypePackage:
		// Deliberately not uninstalled, and this is a decision rather than an omission.
		//
		// The host cannot know what else on this machine needs the package. Uninstalling a
		// container runtime because a declaration changed would stop every container on the
		// node, and the machine may have had the package before the mesh ever saw it
		// (novox/hq research 012: adopted, not installed). Undeclaring says "the mesh no
		// longer requires this", which is not the same as "remove it".
		return "forgotten", "left installed; the host does not uninstall what it cannot know is unused", nil

	case declaration.TypeAction:
		// An action has no footprint the host can undo — it ran, and whatever it did belongs
		// to whatever it acted on.
		return "forgotten", "an action leaves nothing the host owns", nil

	default:
		return "", "", fmt.Errorf("no way to remove a %q", a.Type)
	}
}

// ExecRunner runs a real command, with stdin closed and output captured.
func ExecRunner(ctx context.Context, name string, args ...string) (string, error) {
	cmd := exec.CommandContext(ctx, name, args...)
	cmd.Stdin = nil
	out, err := cmd.Output()
	if err != nil {
		var exit *exec.ExitError
		if errors.As(err, &exit) {
			return string(out), fmt.Errorf("%s exited %d: %s",
				name, exit.ExitCode(), strings.TrimSpace(string(exit.Stderr)))
		}
		return string(out), fmt.Errorf("%s: %w", name, err)
	}
	return string(out), nil
}

// applyPackage installs a package the machine does not have.
//
// It never upgrades and never removes. "Present" is the whole of what a package resource
// asserts, because version is the package manager's business and the mesh does not have a
// second opinion about it (novox/hq ADR 0005 — the host depends on nothing, and that includes
// not becoming a second package manager).
func applyPackage(ctx context.Context, sys system.System, r *declaration.Package, run Runner) (Outcome, error) {
	out := begin(r)

	installed, err := sys.PackageInstalled(ctx, run, r.Package)
	if err != nil {
		return out, err
	}
	if installed {
		out.Action = "unchanged"
		out.Detail = "already installed"
		return out, nil
	}

	if err := sys.InstallPackage(ctx, run, r.Package); err != nil {
		return out, fmt.Errorf("installing %s: %w", r.Package, err)
	}

	// Read back. A package manager exiting zero says the transaction was accepted.
	installed, err = sys.PackageInstalled(ctx, run, r.Package)
	if err != nil {
		return out, err
	}
	if !installed {
		return out, fmt.Errorf(
			"%s was installed without error and the package database does not have it", r.Package)
	}

	out.Action = "created"
	return out, nil
}

// Labels the host puts on every container it creates.
//
// specLabel carries a digest of the declaration that made the container. It is what lets a
// reconcile answer "is this container the one the current declaration describes" without
// comparing every field the runtime reports — which cannot be done reliably, because a runtime
// normalises, defaults and reorders what it is given, and the differences that produces are
// indistinguishable from real drift.
const (
	specLabel = "mesh-host.spec"
	idLabel   = "mesh-host.id"
)

// containerSpec is the identity of a declared container: everything that, if changed, means
// the running container is no longer what was asked for.
func containerSpec(r *declaration.Container) string {
	keys := make([]string, 0, len(r.Env))
	for k := range r.Env {
		keys = append(keys, k)
	}
	sort.Strings(keys)

	var b strings.Builder
	b.WriteString(r.Image + "\n" + r.Name + "\n")
	for _, k := range keys {
		b.WriteString("env " + k + "=" + r.Env[k] + "\n")
	}
	for _, p := range r.Ports {
		b.WriteString("port " + p + "\n")
	}
	for _, v := range r.Volumes {
		b.WriteString("volume " + v + "\n")
	}
	for _, a := range r.Args {
		b.WriteString("arg " + a + "\n")
	}
	return fmt.Sprintf("%x", sha256.Sum256([]byte(b.String())))
}

// containerState reports whether a container is running and which spec made it.
// The error means the container does not exist.
func containerState(ctx context.Context, name string, run Runner) (state struct {
	Running bool
	Spec    string
}, err error) {
	out, err := run(ctx, "docker", "inspect", "--format",
		"{{.State.Running}}\t{{index .Config.Labels \""+specLabel+"\"}}", name)
	if err != nil {
		return state, fmt.Errorf("no container named %s", name)
	}
	running, spec, _ := strings.Cut(strings.TrimSpace(out), "\t")
	state.Running = running == "true"
	state.Spec = strings.TrimSpace(spec)
	return state, nil
}

// applyContainer makes the declared container the one that is running.
//
// There is no "update" for a container: a container's configuration is fixed when it is
// created, so any change is a replacement. Saying that plainly is better than a partial
// in-place update that leaves the running thing half-declared.
func applyContainer(ctx context.Context, r *declaration.Container, run Runner) (Outcome, error) {
	out := begin(r)
	want := containerSpec(r)

	cri, err := containerRuntime(ctx, run)
	if err != nil {
		return out, fmt.Errorf("%w, so nothing can be said about %q", err, r.Name)
	}

	before, err := containerState(ctx, r.Name, run)
	existed := err == nil

	switch {
	case existed && before.Spec == want && before.Running:
		out.Action = "unchanged"
		return out, nil
	case existed:
		if _, err := run(ctx, cri, "rm", "-f", r.Name); err != nil {
			return out, fmt.Errorf("replacing container %s: %w", r.Name, err)
		}
	}

	args := []string{"run", "--detach", "--name", r.Name, "--restart", "unless-stopped"}
	if r.Network != "" {
		args = append(args, "--network", r.Network)
	}
	args = append(args,
		"--label", specLabel+"="+want, "--label", idLabel+"="+r.ID)
	for _, k := range sortedKeys(r.Env) {
		args = append(args, "--env", k+"="+r.Env[k])
	}
	for _, p := range r.Ports {
		args = append(args, "--publish", p)
	}
	for _, v := range r.Volumes {
		args = append(args, "--volume", v)
	}
	args = append(args, r.Image)
	args = append(args, r.Args...)

	if _, err := run(ctx, cri, args...); err != nil {
		return out, fmt.Errorf("starting container %s: %w", r.Name, err)
	}

	// Read back. `docker run --detach` returning an id says the container was created, not
	// that it is still running — a container whose entrypoint exits immediately satisfies the
	// command exactly as one that came up does.
	after, err := containerState(ctx, r.Name, run)
	if err != nil {
		return out, fmt.Errorf("started container %s and it is not there: %w", r.Name, err)
	}
	if !after.Running {
		return out, fmt.Errorf(
			"container %s was started and is not running. It exited; ask the runtime for its "+
				"logs", r.Name)
	}
	if after.Spec != want {
		return out, fmt.Errorf("container %s is not the one that was declared after creating it", r.Name)
	}

	out.Action = "created"
	if existed {
		out.Action = "updated"
		out.Detail = "replaced; a container's configuration is fixed when it is created"
	}
	return out, nil
}

func sortedKeys(m map[string]string) []string {
	keys := make([]string, 0, len(m))
	for k := range m {
		keys = append(keys, k)
	}
	sort.Strings(keys)
	return keys
}

// applyAction runs something the bundle declared, and never learns what it means.
//
// Verify does double duty, and that is the design rather than a convenience: it is both the
// idempotency check and the read-back. Running it first is how the host knows whether there is
// anything to do — it does not know what a database is, so "is the database there" is a
// question only the declaration can ask. Running it again afterwards is how the host knows the
// command had the effect it claimed (novox/hq ADR 0005).
func applyAction(ctx context.Context, r *declaration.Action, run Runner) (Outcome, error) {
	out := begin(r)

	if _, err := runAction(ctx, r, r.Verify, run); err == nil {
		out.Action = "unchanged"
		out.Detail = "already true"
		return out, nil
	}

	if _, err := runAction(ctx, r, r.Command, run); err != nil {
		return out, fmt.Errorf("running the action: %w", err)
	}

	if _, err := runAction(ctx, r, r.Verify, run); err != nil {
		return out, fmt.Errorf(
			"the action ran without error and its own verify still fails: %w\n\n"+
				"The command reported success and the thing it was for did not happen, which "+
				"is exactly what verify exists to catch", err)
	}

	out.Action = "created"
	out.Detail = "verify was false and is now true"
	return out, nil
}

// runAction runs one of an action's command lines, on the machine or inside a container.
func runAction(ctx context.Context, r *declaration.Action, argv []string, run Runner) (string, error) {
	if len(argv) == 0 {
		return "", errors.New("no command")
	}
	if r.In != "" {
		return run(ctx, "docker", append([]string{"exec", r.In}, argv...)...)
	}
	return run(ctx, argv[0], argv[1:]...)
}

// Container runtimes the host knows how to ask.
//
// Two, because two exist on machines the mesh runs on. The list is short on purpose: each entry
// is a claim that its probe and its CLI have been checked, not that a binary of that name might
// work (novox/hq ADR 0005).
//
// The probe differs and the rest does not, which is what makes this a lookup rather than an
// interface. `docker info --format {{.ServerVersion}}` fails on podman — the field does not
// exist in its report — while `run`, `inspect --format` and `rm -f` are identical, including
// docker's own Go template syntax for reading state and labels.
var containerRuntimes = []struct {
	command string
	// probe asks the runtime for its version in the form THAT runtime understands. It must
	// prove the runtime is FUNCTIONING, never that a binary is on disk
	// (novox/hq 04-ISSUES/007).
	probe []string
}{
	{command: "docker", probe: []string{"info", "--format", "{{.ServerVersion}}"}},
	{command: "podman", probe: []string{"info", "--format", "{{.Version.Version}}"}},
}

// containerRuntime returns the runtime this machine actually has, or says there is none.
//
// Detected rather than declared, because a machine already carrying one keeps it: adoption
// takes over what is there rather than replacing it (novox/hq research 012). Which runtime a
// machine has is reported upward in the profile; what to install on a machine with none is the
// control plane's decision, not this one's.
func containerRuntime(ctx context.Context, run Runner) (string, error) {
	var tried []string
	for _, rt := range containerRuntimes {
		if _, err := run(ctx, rt.command, rt.probe...); err == nil {
			return rt.command, nil
		}
		tried = append(tried, rt.command)
	}
	return "", fmt.Errorf(
		"no container runtime answers on this machine (tried %s)", strings.Join(tried, ", "))
}