mesh-sdk/go/views/writer.go

package views

import (
	"context"
	"encoding/json"
	"errors"
	"fmt"
	"slices"
	"sync"
	"time"

	"github.com/nats-io/nats.go"
	"github.com/nats-io/nats.go/jetstream"

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

// The finding writer (design 47 G §461, §485, §503): the one writer every observer uses to keep its findings in its own
// state `findings` and to say them as events (§383, §384). It names no seat and no module: the observer hands in the
// subjects of `finding-written`, `finding-cleared`, `findings-stated` and `findings-asked`, which the generated
// contract of the seat that defines them makes from its module name and node (§461, §439).

// The time constants of the writer (design 47 G §383, §485).
const (
	// PublishRetryFor is how long an event the bus did not acknowledge is sent again before it is logged as an error.
	PublishRetryFor = time.Minute
	// ObserveRetryEvery is how often a failing observe is called again until a set is said.
	ObserveRetryEvery = time.Minute
)

// FindingSubject is what a finding is about (the `Subject` of design 47 G §439's finding).
type FindingSubject struct {
	Scope   string   `json:"scope"`
	ID      string   `json:"id"`
	Machine string   `json:"machine,omitempty"`
	Also    []string `json:"also,omitempty"`
}

// FindingEvidence is one observation, as it was said.
type FindingEvidence struct {
	At   time.Time `json:"at"`
	Said string    `json:"said"`
}

// FindingAction is one answer the operator may give from a notification.
type FindingAction struct {
	Label     string            `json:"label"`
	Verb      string            `json:"verb"`
	Machine   string            `json:"machine,omitempty"`
	Arguments map[string]string `json:"arguments,omitempty"`
	Level     string            `json:"level,omitempty"`
}

// FindingBody is one finding as an observer says it: the layout whole of design 47 G §439's list, the body of
// `finding-written`, of each entry of `findings-stated` and the value of its key `<Scope>.<ID>.<Kind>` in the state `findings`.
type FindingBody struct {
	Kind        string            `json:"kind"`
	Scope       string            `json:"scope"`
	ID          string            `json:"id"`
	At          time.Time         `json:"at"`
	Since       time.Time         `json:"since"`
	Severity    string            `json:"severity"`
	Summary     string            `json:"summary"`
	Headline    string            `json:"headline,omitempty"`
	Explanation string            `json:"explanation,omitempty"`
	Subject     FindingSubject    `json:"subject"`
	Evidence    []FindingEvidence `json:"evidence,omitempty"`
	Source      string            `json:"source"`
	Observer    string            `json:"observer"`
	Owner       string            `json:"owner,omitempty"`
	WaitingOn   string            `json:"waiting-on,omitempty"`
	Actions     []FindingAction   `json:"actions,omitempty"`
	Needs       string            `json:"needs,omitempty"`
	Resolver    string            `json:"resolver,omitempty"`
	Ref         string            `json:"ref,omitempty"`
	Run         string            `json:"run,omitempty"`
}

// Key is the finding's key in the state `findings`: `<scope>.<id>.<kind>` (novox/hq ADR 0564 §2).
func (f FindingBody) Key() string { return FindingKey(f.Scope, f.ID, f.Kind) }

// FindingKey is the key of the finding of scope, id and kind.
func FindingKey(scope, id, kind string) string { return scope + "." + id + "." + kind }

// FindingCleared is the body of `finding-cleared`: an observer no longer observes the finding.
type FindingCleared struct {
	Kind  string    `json:"kind"`
	Scope string    `json:"scope"`
	ID    string    `json:"id"`
	At    time.Time `json:"at"`
	Why   string    `json:"why"`
}

// PendingKey is a key the observer cannot judge yet since its start: its `since` and its bound (a duration such as 5m).
type PendingKey struct {
	Key   string    `json:"key"`
	Since time.Time `json:"since"`
	Bound string    `json:"bound"`
}

// FindingsStated is the body of `findings-stated`: every finding the observer observes now, and what it cannot judge yet.
type FindingsStated struct {
	At       time.Time     `json:"at"`
	Asked    string        `json:"asked,omitempty"`
	Findings []FindingBody `json:"findings"`
	Pending  []PendingKey  `json:"pending"`
}

// FindingsAsked is the body of `findings-asked`: the keeper asks every observer, or the one named, to state its findings.
type FindingsAsked struct {
	Ask    string    `json:"ask"`
	At     time.Time `json:"at"`
	By     string    `json:"by"`
	Module string    `json:"module,omitempty"`
}

// Observed is what an observer's own truth says now: the findings it observes and the keys it cannot judge yet.
type Observed struct {
	Findings []FindingBody
	Pending  []PendingKey
}

// FindingState is the observer's own state `findings` (stdio.State and process.State both are one).
type FindingState interface {
	Get(key string) (*stdio.StateEntry, error)
	Put(key string, value any) (uint64, error)
	Delete(key string) error
	Keys() ([]string, error)
}

// WriterSubjects are the subjects of the four events, as the seat's generated contract makes them for the observer's
// module and node.
type WriterSubjects struct {
	Written, Cleared, Stated, Asked string
}

// WriterConfig makes a Writer.
type WriterConfig struct {
	// Module is the observer's module name: it fills the observer of a finding and tells an ask for another module.
	Module string
	// State is the observer's own state `findings`.
	State FindingState
	// JS is the bus connection's JetStream; Conn, when given, lets the writer hear the connection made again.
	JS   jetstream.JetStream
	Conn *nats.Conn
	// Subjects are the four events' subjects.
	Subjects WriterSubjects
	// Observe answers the findings the observer observes now and its pending keys, or an error. It is called under the
	// writer's one lock, so it must not call Write or Clear.
	Observe func(ctx context.Context) (Observed, error)
	// Send puts a set larger than one bus message in the observer's sent store and answers what to publish instead
	// (stateobjects.Sender.Send); nil sends the body as it is.
	Send func(label string, body []byte) ([]byte, error)
	// Logf is the observer's journal.
	Logf func(string, ...any)
	// Now is the clock, Wait pauses for d and answers false when ctx ended; nil are the real ones.
	Now  func() time.Time
	Wait func(ctx context.Context, d time.Duration) bool
	// Publish sends one message and answers when the bus acknowledged it, and Listen makes a consumer that delivers only
	// the asks published after it is made; nil are JS's.
	Publish func(ctx context.Context, subject string, data []byte, id string) error
	Listen  func(ctx context.Context, subject string, handle func(data []byte)) (stop func(), err error)
}

// Writer keeps an observer's findings and says them (design 47 G §461, §485). Its acts, Write, Clear and a statement,
// run one at a time in the order they are asked.
type Writer struct {
	cfg  WriterConfig
	mu   sync.Mutex // the one lock of the acts
	seq  uint64
	ctl  sync.Mutex // guards the consumer, the asks and the loop's context
	stop func()
	asks []string
	want bool
	wake chan struct{}
	run  context.Context
	end  context.CancelFunc
}

// NewWriter is the writer for cfg.
func NewWriter(cfg WriterConfig) (*Writer, error) {
	switch {
	case cfg.Module == "":
		return nil, errors.New("a finding writer needs its module's name")
	case cfg.State == nil:
		return nil, errors.New("a finding writer needs the state `findings`")
	case cfg.Observe == nil:
		return nil, errors.New("a finding writer needs the observe function")
	case cfg.Subjects.Written == "" || cfg.Subjects.Cleared == "" || cfg.Subjects.Stated == "" || cfg.Subjects.Asked == "":
		return nil, errors.New("a finding writer needs the subjects of finding-written, finding-cleared, findings-stated and findings-asked")
	case cfg.JS == nil && (cfg.Publish == nil || cfg.Listen == nil):
		return nil, errors.New("a finding writer needs the bus connection's JetStream")
	}
	if cfg.Now == nil {
		cfg.Now = func() time.Time { return time.Now().UTC() }
	}
	if cfg.Wait == nil {
		cfg.Wait = func(ctx context.Context, d time.Duration) bool {
			t := time.NewTimer(d)
			defer t.Stop()
			select {
			case <-ctx.Done():
				return false
			case <-t.C:
				return true
			}
		}
	}
	if cfg.Logf == nil {
		cfg.Logf = func(string, ...any) {}
	}
	if cfg.Publish == nil {
		cfg.Publish = jetStreamPublish(cfg.JS)
	}
	if cfg.Listen == nil {
		cfg.Listen = jetStreamListen(cfg.JS)
	}
	return &Writer{cfg: cfg, wake: make(chan struct{}, 1)}, nil
}

func jetStreamPublish(js jetstream.JetStream) func(context.Context, string, []byte, string) error {
	return func(ctx context.Context, subject string, data []byte, id string) error {
		pub, cancel := context.WithTimeout(ctx, 10*time.Second)
		defer cancel()
		_, err := js.PublishMsg(pub, &nats.Msg{Subject: subject, Data: data}, jetstream.WithMsgID(id))
		return err
	}
}

// jetStreamListen makes an ephemeral consumer of its own on the stream that holds subject, from the next message on.
func jetStreamListen(js jetstream.JetStream) func(context.Context, string, func([]byte)) (func(), error) {
	return func(ctx context.Context, subject string, handle func([]byte)) (func(), error) {
		stream, err := js.StreamNameBySubject(ctx, subject)
		if err != nil {
			return nil, fmt.Errorf("the stream of %s: %w", subject, err)
		}
		cons, err := js.CreateConsumer(ctx, stream, jetstream.ConsumerConfig{
			FilterSubject: subject, DeliverPolicy: jetstream.DeliverNewPolicy, AckPolicy: jetstream.AckNonePolicy,
			InactiveThreshold: 10 * time.Minute,
		})
		if err != nil {
			return nil, fmt.Errorf("a consumer of %s: %w", subject, err)
		}
		cc, err := cons.Consume(func(m jetstream.Msg) { handle(m.Data()) })
		if err != nil {
			return nil, fmt.Errorf("consuming %s: %w", subject, err)
		}
		return cc.Stop, nil
	}
}

// Start makes the consumer of `findings-asked` and then states the findings now, and again on each ask and each time the
// connection is made again (design 47 G §384, §485). It returns once the consumer exists; the statements run in the
// writer's own goroutine until ctx ends or Close.
func (w *Writer) Start(ctx context.Context) error {
	w.ctl.Lock()
	defer w.ctl.Unlock()
	if w.run != nil {
		return errors.New("the finding writer is started")
	}
	w.run, w.end = context.WithCancel(ctx)
	if err := w.listen(); err != nil {
		w.end()
		w.run = nil
		return err
	}
	w.asks, w.want = append(w.asks, ""), true
	if w.cfg.Conn != nil {
		prev := w.cfg.Conn.Opts.ReconnectedCB
		w.cfg.Conn.SetReconnectHandler(func(nc *nats.Conn) {
			if prev != nil {
				prev(nc)
			}
			w.Reconnected()
		})
	}
	go w.loop(w.run)
	w.signal()
	return nil
}

// Close ends the writer's consumer and its statements.
func (w *Writer) Close() {
	w.ctl.Lock()
	defer w.ctl.Unlock()
	if w.stop != nil {
		w.stop()
		w.stop = nil
	}
	if w.end != nil {
		w.end()
	}
}

func (w *Writer) listen() error {
	stop, err := w.cfg.Listen(w.run, w.cfg.Subjects.Asked, w.heard)
	if err != nil {
		return err
	}
	if w.stop != nil {
		w.stop()
	}
	w.stop = stop
	return nil
}

// Reconnected is told that the bus connection was made again: the writer makes its consumer again, from the newest ask,
// and states its findings again (design 47 G §485).
func (w *Writer) Reconnected() {
	w.ctl.Lock()
	defer w.ctl.Unlock()
	if w.run == nil || w.run.Err() != nil {
		return
	}
	if err := w.listen(); err != nil {
		w.cfg.Logf("the finding writer could not make its consumer of %s again: %v", w.cfg.Subjects.Asked, err)
	}
	w.asks, w.want = append(w.asks, ""), true
	w.signal()
}

func (w *Writer) signal() {
	select {
	case w.wake <- struct{}{}:
	default:
	}
}

// heard is an ask: answered when it names no module or this one.
func (w *Writer) heard(data []byte) {
	var a FindingsAsked
	if err := json.Unmarshal(data, &a); err != nil {
		w.cfg.Logf("an ask for findings could not be read: %v", err)
		return
	}
	if a.Module != "" && a.Module != w.cfg.Module {
		return
	}
	w.ctl.Lock()
	w.asks, w.want = append(w.asks, a.Ask), true
	w.ctl.Unlock()
	w.signal()
}

func (w *Writer) loop(ctx context.Context) {
	for {
		select {
		case <-ctx.Done():
			return
		case <-w.wake:
		}
		w.ctl.Lock()
		asks, want := w.asks, w.want
		w.asks, w.want = nil, false
		w.ctl.Unlock()
		if !want {
			continue
		}
		asked := ""
		if len(asks) > 0 {
			asked = asks[len(asks)-1]
		}
		for {
			err := w.State(ctx, asked)
			if err == nil {
				break
			}
			w.cfg.Logf("the findings could not be stated, no set said, trying again in %s: %v", ObserveRetryEvery, err)
			if !w.cfg.Wait(ctx, ObserveRetryEvery) {
				return
			}
		}
	}
}

func (w *Writer) now() time.Time { return w.cfg.Now().UTC() }

// unsaid is a finding as compared to say whether what it says changed: not the time it was last seen, nor what saw it.
func unsaid(f FindingBody) string {
	f.At, f.Run = time.Time{}, ""
	raw, _ := json.Marshal(f)
	return string(raw)
}

func (w *Writer) read(key string) (*FindingBody, error) {
	e, err := w.cfg.State.Get(key)
	if err != nil {
		return nil, err
	}
	if e == nil {
		return nil, nil
	}
	var old FindingBody
	if err := json.Unmarshal(e.Value, &old); err != nil {
		return nil, fmt.Errorf("the finding %s in the state findings: %w", key, err)
	}
	return &old, nil
}

// complete fills what the observer leaves to the writer and keeps the key's `since` where the key exists (design 47 G §384).
func (w *Writer) complete(f FindingBody) (FindingBody, *FindingBody, error) {
	if f.Kind == "" || f.Scope == "" || f.ID == "" {
		return f, nil, fmt.Errorf("a finding needs its kind, scope and id: %q.%q.%q", f.Scope, f.ID, f.Kind)
	}
	old, err := w.read(f.Key())
	if err != nil {
		return f, nil, err
	}
	if f.Observer == "" {
		f.Observer = w.cfg.Module
	}
	if f.At.IsZero() {
		f.At = w.now()
	}
	f.At = f.At.UTC()
	switch {
	case old != nil && !old.Since.IsZero():
		f.Since = old.Since
	case f.Since.IsZero():
		f.Since = f.At
	}
	f.Since = f.Since.UTC()
	return f, old, nil
}

// Write keeps the finding in its key and says `finding-written` only when what the finding says changed (design 47 G
// §383, §461). The key's `since` stays where the key exists.
func (w *Writer) Write(ctx context.Context, f FindingBody) error {
	w.mu.Lock()
	defer w.mu.Unlock()
	f, old, err := w.complete(f)
	if err != nil {
		return err
	}
	if _, err := w.cfg.State.Put(f.Key(), f); err != nil {
		return fmt.Errorf("writing the finding %s: %w", f.Key(), err)
	}
	if old != nil && unsaid(*old) == unsaid(f) {
		return nil
	}
	raw, _ := json.Marshal(f)
	w.say(ctx, w.cfg.Subjects.Written, "finding-written", f.Key(), raw)
	return nil
}

// Clear deletes the finding's key and says `finding-cleared`; a key that does not exist is deleted of nothing and is said
// all the same, so a keeper holding the finding clears it (design 47 G §485).
func (w *Writer) Clear(ctx context.Context, scope, id, kind, why string) error {
	w.mu.Lock()
	defer w.mu.Unlock()
	key := FindingKey(scope, id, kind)
	if err := w.cfg.State.Delete(key); err != nil {
		return fmt.Errorf("clearing the finding %s: %w", key, err)
	}
	raw, _ := json.Marshal(FindingCleared{Kind: kind, Scope: scope, ID: id, At: w.now(), Why: why})
	w.say(ctx, w.cfg.Subjects.Cleared, "finding-cleared", key, raw)
	return nil
}

// State judges every finding anew with observe, writes each stated finding's key, deletes each key neither stated nor
// pending, and says `findings-stated` (design 47 G §384, §461); asked is the ask it answers, none at a start. An error of
// observe says no set.
func (w *Writer) State(ctx context.Context, asked string) error {
	w.mu.Lock()
	defer w.mu.Unlock()
	obs, err := w.cfg.Observe(ctx)
	if err != nil {
		return fmt.Errorf("observe: %w", err)
	}
	keep := map[string]bool{}
	body := FindingsStated{At: w.now(), Asked: asked, Findings: []FindingBody{}, Pending: []PendingKey{}}
	for _, f := range obs.Findings {
		f, _, err := w.complete(f)
		if err != nil {
			return err
		}
		if _, err := w.cfg.State.Put(f.Key(), f); err != nil {
			return fmt.Errorf("writing the finding %s: %w", f.Key(), err)
		}
		keep[f.Key()] = true
		body.Findings = append(body.Findings, f)
	}
	for _, p := range obs.Pending {
		keep[p.Key] = true
		body.Pending = append(body.Pending, p)
	}
	keys, err := w.cfg.State.Keys()
	if err != nil {
		return fmt.Errorf("the keys of the state findings: %w", err)
	}
	slices.Sort(keys)
	for _, k := range keys {
		if !keep[k] {
			if err := w.cfg.State.Delete(k); err != nil {
				return fmt.Errorf("deleting the finding %s: %w", k, err)
			}
		}
	}
	raw, _ := json.Marshal(body)
	if w.cfg.Send != nil {
		if raw, err = w.cfg.Send("findings-stated."+w.cfg.Module, raw); err != nil {
			w.cfg.Logf("the findings-stated of %s could not be sent, the next statement covers it: %v", w.cfg.Module, err)
			return nil
		}
	}
	w.say(ctx, w.cfg.Subjects.Stated, "findings-stated", w.cfg.Module, raw)
	return nil
}

// say publishes one event awaiting the bus's acknowledgement, sent again for PublishRetryFor when unacknowledged, then
// written in the journal as an error naming what it said (design 47 G §383); the next statement covers it.
func (w *Writer) say(ctx context.Context, subject, event, what string, data []byte) {
	w.seq++
	id := fmt.Sprintf("%s-%s-%d-%d", w.cfg.Module, event, w.now().UnixNano(), w.seq)
	start := w.now()
	for delay := time.Second; ; delay = min(delay*2, 10*time.Second) {
		err := w.cfg.Publish(ctx, subject, data, id)
		if err == nil {
			return
		}
		if w.now().Sub(start) >= PublishRetryFor || ctx.Err() != nil || !w.cfg.Wait(ctx, delay) {
			w.cfg.Logf("error: %s for %s was not acknowledged by the bus, the next statement covers it: %v", event, what, err)
			return
		}
	}
}