mesh-catalog/modules/claude-code/cmd/claude-code/mesh.go

package main

// What the mesh says about its own tools and machines, as the agent's instructions and the guard on its
// shell need it (novox/hq ADR 0245): every seat verb and module tool that says what shell command it
// replaces, and the mesh's machines by every name and address a session could reach one by.
//
// Read from the bus at start and every few minutes (issue 1126 B13; the controller that answered this before is
// not placed, novox/hq ADR 0590): the seats' verbs from the seats' contract (mesh-contracts), the modules'
// own tools from the manifests the indexer keeps (`mesh-delivery-indexer.modules` and `.manifests`), and the
// machines and where each module runs from the seat mesh-module-assigner's `assignments`. Kept in the module's state so
// a render never waits for the bus, and rendered whenever it changes. Nothing here is written by hand: a verb
// that gains a `replaces` is in the next render's table and the guard's next refusal.

import (
	"context"
	"encoding/json"
	"errors"
	"fmt"
	assignerclient "git.novox.be/novox/mesh-contracts/go/clients/meshmoduleassignerclient"
	mesh "git.novox.be/novox/mesh-contracts/go/seats/meshmoduleassigner"
	catalogclient "git.novox.be/novox/mesh-delivery/contracts/go/clients/meshcatalogclient"
	catalog "git.novox.be/novox/mesh-delivery/contracts/go/seats/meshcatalog"
	"net"
	"os"
	"path/filepath"
	"sort"
	"strings"
	"sync"
	"time"

	"git.novox.be/novox/mesh-sdk/go/catalogue"
	"git.novox.be/novox/mesh-sdk/go/seatdef"
	"git.novox.be/novox/mesh-sdk/go/seatlist"
)

// Replacement is one verb or tool and the shell commands it is the mesh's way to do.
type Replacement struct {
	// Address is how it is called: `<seat>.<verb>`, or `<module>.<tool>`; NodeScoped says a machine goes
	// before it, `<node>/`.
	Address    string   `json:"address"`
	NodeScoped bool     `json:"node_scoped,omitempty"`
	Seat       bool     `json:"seat,omitempty"`
	Does       string   `json:"does,omitempty"`
	Replaces   []string `json:"replaces"`
	// On is where a module's tool runs; empty for a seat's verb, which every holder serves.
	On []string `json:"on,omitempty"`
}

// Machine is one of the mesh's machines and the names and addresses it is reached by.
type Machine struct {
	Name      string   `json:"name"`
	Domains   []string `json:"domains,omitempty"`
	Addresses []string `json:"addresses,omitempty"`
}

// MeshTools is what the bus said, as kept in the module's state.
type MeshTools struct {
	Replacements []Replacement `json:"replacements"`
	Machines     []Machine     `json:"machines"`
	// Asked is when the bus was last read whole.
	Asked string `json:"asked,omitempty"`
	// Refused is how often the guard here named each tool instead of a work-around, from its record: the
	// table's order. Never kept — read from the record at each render.
	Refused map[string]int `json:"-"`
}

// RefusedCounts are how often the guard's record named each tool, over its last lines.
func RefusedCounts(log string) map[string]int {
	out := map[string]int{}
	for _, e := range ReadGuardLog(log, 2000) {
		for _, t := range e.Tools {
			out[t]++
		}
	}
	return out
}

// InternalSuffix is the mesh's own names' domain: every machine and service is `<name>.internal`
// (novox/hq ADR 0194), and a session reaches none of them by ssh.
const InternalSuffix = ".internal"

func (p Paths) meshTools() string { return filepath.Join(p.State, "mesh-tools.json") }

// ReadMeshTools is what was kept, or nil.
func ReadMeshTools(p Paths) *MeshTools {
	var m MeshTools
	if !readJSON(p.meshTools(), &m) {
		return nil
	}
	return &m
}

// The states read, each declared in the manifest's reads.
const (
	AssignmentsRead = (assignerModule + "." + mesh.StateAssignments)
	ModulesRead     = indexerModule + "." + catalog.StateModules
	ManifestsRead   = indexerModule + "." + catalog.StateManifests
)

// States reads the module states the manifest declares in its reads: a state's keys, and one key's value.
type States interface {
	Keys(read string) ([]string, error)
	Get(read, key string) ([]byte, bool, error)
}

// ModuleTools is one module's own tools that replace a command, as its manifest says (`replaces`), and where it runs.
type ModuleTools struct {
	Module   string
	On       []string
	Replaces map[string][]string
}

// assignments is every node's assignments, from the seat mesh-module-assigner's state.
func assignments(st States) (map[string]mesh.Assignments, error) {
	keys, err := st.Keys(AssignmentsRead)
	if err != nil {
		return nil, fmt.Errorf("%s: %w", AssignmentsRead, err)
	}
	read := assignerclient.NewKeys()
	for _, k := range keys {
		v, found, err := st.Get(AssignmentsRead, k)
		if err != nil {
			return nil, fmt.Errorf("%s %s: %w", AssignmentsRead, k, err)
		}
		if found {
			_, _ = read.Put(k, v, 0, false) // a key that does not read places nothing
		}
	}
	return read.All(), nil
}

// runsOn is, by module, the nodes assignments puts it on, sorted.
func runsOn(placed map[string]mesh.Assignments) map[string][]string {
	out := map[string][]string{}
	for node, p := range placed {
		for _, m := range p.Modules {
			out[m.Module] = append(out[m.Module], node)
		}
	}
	for _, nodes := range out {
		sort.Strings(nodes)
	}
	return out
}

// NodesRunning is the nodes assignments puts module on.
func NodesRunning(st States, module string) ([]string, error) {
	placed, err := assignments(st)
	if err != nil {
		return nil, err
	}
	return runsOn(placed)[module], nil
}

// modulesTools is every placed module's own tools that replace a command, by module, from the manifest the indexer
// keeps of it. A module whose manifest cannot be read is left out and said in the error.
func modulesTools(st States, on map[string][]string) ([]ModuleTools, error) {
	var out []ModuleTools
	var problems []error
	modules := make([]string, 0, len(on))
	for m := range on {
		modules = append(modules, m)
	}
	sort.Strings(modules)
	for _, module := range modules {
		key, err := catalogclient.ModulesKey(module)
		if err != nil {
			continue
		}
		raw, found, err := st.Get(ModulesRead, key)
		if err != nil || !found {
			if err != nil {
				problems = append(problems, fmt.Errorf("%s %s: %w", ModulesRead, module, err))
			}
			continue
		}
		m, err := catalogclient.DecodeModule(raw)
		if err != nil || m.Manifest == "" {
			continue
		}
		mkey, err := catalogclient.ManifestsKey(m.Manifest)
		if err != nil {
			continue
		}
		raw, found, err = st.Get(ManifestsRead, mkey)
		if err != nil || !found {
			if err != nil {
				problems = append(problems, fmt.Errorf("%s of %s: %w", ManifestsRead, module, err))
			}
			continue
		}
		kept, err := catalogclient.DecodeManifest(raw)
		if err != nil {
			problems = append(problems, fmt.Errorf("the manifest of %s: %w", module, err))
			continue
		}
		var man struct {
			Replaces map[string][]string `json:"replaces"`
		}
		if json.Unmarshal([]byte(kept.Manifest), &man) != nil || len(man.Replaces) == 0 {
			continue
		}
		out = append(out, ModuleTools{Module: module, On: on[module], Replaces: man.Replaces})
	}
	return out, errors.Join(problems...)
}

// firstSentence is what a verb does, short: its description to the first full stop, at most 70 characters.
func firstSentence(s string) string {
	s = strings.TrimSpace(s)
	if i := strings.Index(s, ". "); i >= 0 {
		s = s[:i]
	}
	s = strings.TrimSuffix(s, ".")
	if i := strings.IndexAny(s, ":;("); i > 20 {
		s = strings.TrimSpace(s[:i])
	}
	if r := []rune(s); len(r) > 70 {
		s = strings.TrimSpace(string(r[:69])) + "…"
	}
	return s
}

// ReplacementsOf is every seat verb and module tool that says it replaces a command: the seats' verbs first, in the
// seats' order, then the modules' own tools by module and tool.
func ReplacementsOf(defined []seatdef.Definition, modules []ModuleTools) []Replacement {
	var out []Replacement
	for _, s := range defined {
		for _, v := range s.Verbs {
			if len(v.Replaces) == 0 {
				continue
			}
			out = append(out, Replacement{Address: s.Seat + "." + v.Name, NodeScoped: s.Scope == "node", Seat: true,
				Does: firstSentence(v.Description), Replaces: v.Replaces})
		}
	}
	sorted := append([]ModuleTools(nil), modules...)
	sort.SliceStable(sorted, func(i, j int) bool { return sorted[i].Module < sorted[j].Module })
	for _, m := range sorted {
		names := make([]string, 0, len(m.Replaces))
		for t := range m.Replaces {
			names = append(names, t)
		}
		sort.Strings(names)
		for _, t := range names {
			if len(m.Replaces[t]) == 0 {
				continue
			}
			out = append(out, Replacement{Address: m.Module + "." + t, NodeScoped: true, Replaces: m.Replaces[t], On: m.On,
				Does: strings.ReplaceAll(strings.TrimPrefix(t, strings.ReplaceAll(m.Module, "-", "_")+"_"), "_", " ")})
		}
	}
	return out
}

// lookup resolves one name, briefly: a name that does not resolve here is no address.
func lookup(name string) []string {
	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
	defer cancel()
	addrs, _ := net.DefaultResolver.LookupHost(ctx, name)
	return addrs
}

// ReadMeshToolsFrom reads the bus now: the seats' verbs, the placed modules' own tools, and every node assignments names
// with its domains and addresses. A part that cannot be read is left out and said in the error; what was read is kept.
func ReadMeshToolsFrom(st States, defined []seatdef.Definition, resolve func(string) []string) (MeshTools, error) {
	var m MeshTools
	placed, err := assignments(st)
	if err != nil {
		m.Replacements = ReplacementsOf(defined, nil)
		return m, err
	}
	modules, merr := modulesTools(st, runsOn(placed))
	m.Replacements = ReplacementsOf(defined, modules)
	for node, p := range placed {
		machine := Machine{Name: node, Domains: []string{node + InternalSuffix}}
		if d := strings.ToLower(strings.TrimSpace(p.Decided.PublicDomain)); d != "" {
			machine.Domains = append(machine.Domains, d)
		}
		seen := map[string]bool{}
		for _, name := range append([]string{node}, machine.Domains...) {
			for _, a := range resolve(name) {
				if !seen[a] && !net.ParseIP(a).IsLoopback() {
					seen[a] = true
					machine.Addresses = append(machine.Addresses, a)
				}
			}
		}
		sort.Strings(machine.Addresses)
		m.Machines = append(m.Machines, machine)
	}
	sort.Slice(m.Machines, func(i, j int) bool { return m.Machines[i].Name < m.Machines[j].Name })
	return m, merr
}

// KeepMeshTools writes what was asked to the module's state when it differs from what was kept, and
// answers whether it did. When was asked is not a difference.
func KeepMeshTools(p Paths, m MeshTools) bool {
	if was := ReadMeshTools(p); was != nil {
		a, b := *was, m
		a.Asked, b.Asked = "", ""
		x, _ := json.Marshal(a)
		y, _ := json.Marshal(b)
		if string(x) == string(y) {
			return false
		}
	}
	m.Asked = time.Now().UTC().Format(time.RFC3339)
	raw, _ := indented(m)
	tmp := p.meshTools() + ".tmp"
	if os.WriteFile(tmp, raw, 0o600) != nil {
		return false
	}
	return os.Rename(tmp, p.meshTools()) == nil
}

// GuardDataOf is what the guard judges with on this machine.
func GuardDataOf(p Paths, m *MeshTools) GuardData {
	d := GuardData{Node: p.Node, Log: filepath.Join(p.State, "guard.log"), Machines: []Machine{}, Replacements: []Replacement{}}
	if p.Confined() {
		// The session's account cannot write the module's state, and must not rewrite its own record (ADR 0266).
		d.Agent, d.Log = p.Agent, JournalRecord+GuardJournalID
	}
	if m != nil {
		d.Machines, d.Replacements = m.Machines, m.Replacements
	}
	return d
}

// guardFiles is the guard to place: this module's own executable and what it judges with. None where the
// executable cannot be named — then no hook is written, rather than one that cannot run.
func guardFiles(p Paths, m *MeshTools) *GuardFiles {
	exe, err := os.Executable()
	if err != nil {
		return nil
	}
	if resolved, err := filepath.EvalSymlinks(exe); err == nil {
		exe = resolved
	}
	return &GuardFiles{Exe: exe, Data: GuardDataOf(p, m), WaitsDir: WaitsDirPath(p)}
}

// MeshToolsEvery is how often the machines' addresses are resolved again; the tools and the assignments are read again
// when their followed states change (a views event), not on this timer.
const MeshToolsEvery = 10 * time.Minute

// MeshToolsFirst is how long the first read waits for the first views to be answered.
const MeshToolsFirst = 2 * time.Minute

// refreshMeshTools reads the pictures once the first views are answered, and again whenever one changes (and the
// machines' addresses every few minutes), and renders when what it said changed.
func refreshMeshTools(p Paths) {
	select {
	case <-meshViews.Ready():
	case <-time.After(MeshToolsFirst):
		say("the first views of the tools and the assignments were not answered in %s; reading what is held", MeshToolsFirst)
	}
	startSeatList(seatlist.StdioBus())
	for {
		m, err := ReadMeshToolsFrom(viewStates{meshViews}, definedSeats(), lookup)
		if err != nil {
			say("reading what each tool replaces and which machines the mesh has: %v", err)
		}
		if (err == nil || len(m.Replacements) > 0) && KeepMeshTools(p, m) {
			if _, err := RenderNow(p, writeManaged); err != nil {
				say("rendering what the bus said of the tools: %v", err)
			} else {
				say("the mesh's tools changed: %d replace a command, %d machines; rendered", len(m.Replacements), len(m.Machines))
			}
		}
		select {
		case <-meshChanged:
			time.Sleep(2 * time.Second) // a burst of events is one read
			select {
			case <-meshChanged:
			default:
			}
		case <-time.After(MeshToolsEvery):
		}
	}
}

// GuardStatus is what claude_code_guard answers.
func GuardStatus(p Paths, last int) map[string]any {
	m := ReadMeshTools(p)
	d := GuardDataOf(p, m)
	asked := "never: the bus has not been read whole yet"
	if m != nil {
		asked = m.Asked
	}
	return map[string]any{
		"refuses": []string{
			"ssh, scp, sftp, rsync, mosh or autossh to a mesh machine: its name, a name under its domains, any *" + InternalSuffix +
				" name or one of its addresses, read as ssh reads it (ssh -G); a jump through one too. An ssh login as `" +
				ForgeUser + "` is the forge's account and passes",
			"writing /etc/hosts or /etc/resolv.conf, by the shell or the agent's Edit and Write",
			"HOSTALIASES, named anywhere in a command",
			"any command naming " + OverrideVar,
			"where the machine names an agent account of its own (novox/hq ADR 0266): every command and file edit of a " +
				"session that does not run as it, whatever it is, and no override lifts it",
		},
		"agent_account": map[bool]string{true: p.Agent, false: "none of its own: agents run as the operator's account"}[p.Confined()],
		"override": OverrideVar + "=<why>, set in the operator's own shell before the session starts; read from the session's " +
			"environment as it was started, recorded with why, and not honoured when it cannot be recorded",
		"hook":         GuardCommand(),
		"asked":        asked,
		"machines":     d.Machines,
		"replacements": d.Replacements,
		"instead_of":   InsteadOf(m),
		"record":       ReadGuardLog(d.Log, last),
	}
}

// seatList is the seat list this module holds: the definitions of the seats, read from the seat mesh-catalog and
// followed on its events (novox/hq ADR 0635 §1, ADR 0636), the foundation's bootstrapped ones until it answers.
var seatList *seatlist.Reader

// startSeatList starts the reader once the runtime serves; a change of the list is told like any followed state, so the
// instructions' table is read again.
func startSeatList(bus seatlist.Bus) {
	r, err := seatlist.Start(seatlist.Config{Bus: bus, Bootstrap: seatlist.Bootstrapped(),
		OnChange: func([]catalogue.Seat) {
			select {
			case meshChanged <- struct{}{}:
			default:
			}
		}})
	if err != nil {
		say("following the seat list: %v", err)
	}
	seatListMu.Lock()
	seatList = r
	seatListMu.Unlock()
}

var seatListMu sync.Mutex

// definedSeats are the held definitions in reading order, mesh seats first, then node seats, each by name; empty until
// the reader has started.
func definedSeats() []seatdef.Definition {
	seatListMu.Lock()
	r := seatList
	seatListMu.Unlock()
	if r == nil {
		return nil
	}
	held := r.Definitions()
	var out []seatdef.Definition
	for _, s := range r.Seats() {
		if d, ok := held[s.Name]; ok {
			out = append(out, d)
		}
	}
	return out
}