mesh-catalog/modules/dnsmasq/cmd/dnsmasq-tools/tools.go

package main

import (
	"context"
	"fmt"
	"slices"
	"strings"
	"time"

	stdio "git.novox.be/novox/mesh-sdk/go"
)

// Stale is how old the watch's last look may be before the health tool says it stopped looking.
const Stale = 2 * time.Minute

// Health is the module's health answer (novox/hq ADR 0240): healthy only when the resolver answers its
// real question on every listen address, over UDP and TCP, and the watch is looking.
type Health struct {
	Healthy bool   `json:"healthy"`
	Why     string `json:"why,omitempty"`
	Look    Look   `json:"look"`
	// Watch is what the watch last saw and did; absent when it has written nothing yet.
	Watch *State `json:"watch,omitempty"`
}

// Health asks the resolver now, and reads what the watch did.
func (m *Machine) Health(ctx context.Context) Health {
	l := m.Look(ctx)
	h := Health{Healthy: l.Healthy, Look: l}
	var why []string
	if !l.Healthy {
		why = append(why, "the resolver does not answer as it must: "+l.Why())
	}
	s, had, err := m.ReadState()
	switch {
	case err != nil:
		h.Healthy = false
		why = append(why, "what the watch saw cannot be read: "+err.Error())
	case !had:
		h.Healthy = false
		why = append(why, "the watch has written nothing yet, so nothing restarts the resolver when it stops answering")
	default:
		h.Watch = &s
		if age := m.Now().Sub(s.Last.At); age > Stale {
			h.Healthy = false
			why = append(why, fmt.Sprintf("the watch last looked %s ago, so nothing restarts the resolver when it "+
				"stops answering", age.Round(time.Second)))
		}
		if !l.Healthy {
			why = append(why, restartWords(s))
		}
		if s.Unit != "" && s.Unit != "active" {
			why = append(why, "the unit is "+s.Unit)
		}
	}
	h.Why = strings.Join(nonEmpty(why), "; ")
	return h
}

// restartWords is what the watch did about it, in words.
func restartWords(s State) string {
	if len(s.Restarts) == 0 {
		return "the watch has not restarted it" + orEmpty("; ", s.Held)
	}
	r := s.Restarts[len(s.Restarts)-1]
	words := fmt.Sprintf("the watch restarted it at %s", r.At.Format("15:04:05 UTC"))
	switch {
	case r.Error != "":
		words += ", and the restart failed: " + r.Error
	case r.Helped != nil && !*r.Helped:
		words += ", which did not help"
	case r.Helped == nil:
		words += ", and its next looks decide whether that helped"
	}
	if s.Held != "" {
		words += "; " + s.Held
	}
	return words
}

func nonEmpty(s []string) []string {
	var out []string
	for _, x := range s {
		if x != "" {
			out = append(out, x)
		}
	}
	return out
}

func str(description string) map[string]any {
	return map[string]any{"type": "string", "description": description}
}

func arg(a map[string]any, k string) string {
	v, _ := a[k].(string)
	return strings.TrimSpace(v)
}

// Tools are the module's tools.
func Tools(m *Machine) []stdio.Tool {
	return []stdio.Tool{
		{Name: "dnsmasq_health", Description: "Whether this machine's mesh resolver answers: its real question — this " +
			"machine's own name — asked now on every address its configuration listens on, over UDP and TCP, each " +
			"answer with its code, addresses and time; and what the module's watch last saw, the restarts it made " +
			"within the hour, whether each helped, and why it holds back from another. Healthy only when every " +
			"question was answered rightly and the watch is looking (novox/hq ADR 0240, issue 348). (r)",
			Input: map[string]any{},
			Run: func(map[string]any) (any, error) {
				return m.Health(context.Background()), nil
			}},
		{Name: "dnsmasq_names", Description: "The mesh's machine names this machine's resolver answers, each with its " +
			"private address, as the roster wrote them (" + NodesFile + "). (r)",
			Input: map[string]any{},
			Run: func(map[string]any) (any, error) {
				text, err := m.read(NodesFile)
				if err != nil {
					return nil, fmt.Errorf("the roster %s cannot be read: %w", NodesFile, err)
				}
				return map[string]any{"names": AnsweredNames(text)}, nil
			}},
		{Name: "dnsmasq_resolve", Description: "Ask this machine's resolver one name's IPv4 address, directly, on one " +
			"address it listens on (default every one) over udp or tcp (default both): the answer code, the addresses " +
			"and the time each took, or why none came — refused when nothing listens there. Replaces dig @address. (r)",
			Input: map[string]any{"name": str("the name to ask, e.g. plex.anchor.internal"),
				"address": str("one listen address to ask (default: every one in its configuration)"),
				"over":    str("udp or tcp (default: both)")},
			Run: func(a map[string]any) (any, error) {
				return m.Resolve(context.Background(), arg(a, "name"), arg(a, "address"), arg(a, "over"))
			}},
	}
}

// Resolve asks name on one listen address or every one, over one transport or both.
func (m *Machine) Resolve(ctx context.Context, name, address, over string) (any, error) {
	if name == "" {
		return nil, fmt.Errorf("name is required")
	}
	config, err := m.read(ConfigFile)
	if err != nil {
		return nil, fmt.Errorf("its configuration %s cannot be read: %w", ConfigFile, err)
	}
	addresses := ListenAddresses(config)
	if address != "" {
		if !slices.Contains(addresses, address) {
			return nil, fmt.Errorf("address %q is not one the resolver listens on (%s); netcheck_dns asks any resolver",
				address, strings.Join(addresses, ", "))
		}
		addresses = []string{address}
	}
	overs := []string{"udp", "tcp"}
	switch over {
	case "":
	case "udp", "tcp":
		overs = []string{over}
	default:
		return nil, fmt.Errorf("over is udp or tcp, not %q", over)
	}
	var answers []Answer
	for _, a := range addresses {
		for _, o := range overs {
			answers = append(answers, m.Ask(ctx, a+":"+m.Port, o, name, AskWithin))
		}
	}
	return map[string]any{"name": name, "answers": answers}, nil
}