mesh-delivery/modules/mesh-delivery-view/internal/view/view.go

// Package view derives the operator's rows from the delivery buckets.
package view

import (
	"encoding/json"
	"sort"
	"strings"
	"sync"
	"time"

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

// Entry is one change of a bucket's key, as the watch delivers it.
type Entry struct {
	Bucket  string
	Key     string
	Value   []byte
	At      time.Time
	Deleted bool
}

// Row is one module on one node.
type Row struct {
	Node     string    `json:"node"`
	Module   string    `json:"module"`
	Status   string    `json:"status"`
	Assigned string    `json:"assigned,omitempty"`
	Have     string    `json:"have,omitempty"`
	Phase    string    `json:"phase,omitempty"`
	Health   string    `json:"health,omitempty"`
	Since    time.Time `json:"since,omitzero"`
	Detail   string    `json:"detail,omitempty"`
	Waits    string    `json:"waits,omitempty"`
	// RequiredBy is what pulled the module onto the node, for an assignment that is pulled in.
	RequiredBy string `json:"required-by,omitempty"`
	Error      string `json:"error,omitempty"`
}

// NodeRow is a node's own facts as counts.
type NodeRow struct {
	Node    string    `json:"node"`
	Windows int       `json:"windows"`
	Busy    int       `json:"busy"`
	Strays  int       `json:"strays"`
	Since   time.Time `json:"since,omitzero"`
	Error   string    `json:"error,omitempty"`
}

// Snapshot is everything the view shows.
type Snapshot struct {
	Rows  []Row     `json:"rows"`
	Nodes []NodeRow `json:"nodes"`
	// Offline is every node and part instance whose presence is offline (the watchman's word).
	Offline []OfflineRow `json:"offline"`
	// Stuck is every key that stopped moving and has not moved again.
	Stuck []StuckRow `json:"stuck"`
	// Unreadable is every watchman value the view could not read, named by its key: never dropped silently.
	Unreadable []UnreadableRow `json:"unreadable"`
}

// OfflineRow is a node ("node") or a part's instance ("part") that is offline, since the presence entry's time.
type OfflineRow struct {
	Kind     string    `json:"kind"`
	Node     string    `json:"node,omitempty"`
	Part     string    `json:"part,omitempty"`
	Instance string    `json:"instance,omitempty"`
	Since    time.Time `json:"since,omitzero"`
}

// StuckRow is a key that stopped moving: its bucket and key, what it waited for and since when.
type StuckRow struct {
	Bucket string    `json:"bucket"`
	Key    string    `json:"key"`
	Waited string    `json:"waited,omitempty"`
	Since  time.Time `json:"since,omitzero"`
}

// UnreadableRow is a value of the watchman's buckets that did not decode, with the reader's error.
type UnreadableRow struct {
	Bucket string `json:"bucket"`
	Key    string `json:"key"`
	Error  string `json:"error"`
}

// Model holds the rows' inputs in memory and nothing else: a restart fills it again from the buckets' last values.
type Model struct {
	mu      sync.Mutex
	assigns map[string]assignment
	builds  map[string]build  // by <module>.build.<id>
	newest  map[string]string // module -> id
	haves   map[string]haveEntry
	nodes   map[string]nodeEntry
	// the watchman's buckets, by key
	presence map[string]presenceEntry
	stuck    map[string]stuckEntry
}

type presenceEntry struct {
	value delivery.Presence
	err   string
	at    time.Time
}

type stuckEntry struct {
	value delivery.Stuck
	err   string
}

// assignment holds only the fields of the controller's Assignment the view reads.
type assignment struct {
	Node       string    `json:"node"`
	Module     string    `json:"module"`
	Build      string    `json:"build"`
	Commit     string    `json:"commit"`
	Since      time.Time `json:"since"`
	Ended      time.Time `json:"ended"`
	State      string    `json:"state"`
	PulledIn   bool      `json:"pulled-in"`
	RequiredBy string    `json:"required-by"`
}

// open says whether the value is an assignment that stands for the module on its node, as the controller's
// Assignment.Stands does. A module pulled in by another stands too.
func (a assignment) open() bool { return a.Ended.IsZero() && a.State == "" }

// build holds only the fields of the controller's Build record the view reads. A record with no state and no failure
// is a done build.
type build struct {
	ID     string `json:"id"`
	Module string `json:"module"`
	State  string `json:"state"`
	Commit string `json:"commit"`
	Failed string `json:"failed"`
}

func (b build) done() bool { return b.State == "" && b.Failed == "" }

type haveEntry struct {
	have delivery.Have
	err  string
	at   time.Time
}

type nodeEntry struct {
	node delivery.HaveNode
	err  string
	at   time.Time
}

// New is an empty model.
func New() *Model {
	return &Model{assigns: map[string]assignment{}, builds: map[string]build{}, newest: map[string]string{},
		haves: map[string]haveEntry{}, nodes: map[string]nodeEntry{},
		presence: map[string]presenceEntry{}, stuck: map[string]stuckEntry{}}
}

// Apply takes one change of a key. A value it cannot read is kept as an error on its row, never guessed.
func (m *Model) Apply(e Entry) {
	m.mu.Lock()
	defer m.mu.Unlock()
	switch e.Bucket {
	case delivery.HaveBucket:
		node, module, isNode, err := delivery.ParseHaveKey(e.Key)
		if err != nil {
			return
		}
		if isNode {
			if e.Deleted {
				delete(m.nodes, node)
				return
			}
			n, err := delivery.DecodeHaveNode(e.Value)
			m.nodes[node] = nodeEntry{node: n, err: errText(err), at: e.At}
			return
		}
		key := node + "." + module
		if e.Deleted {
			delete(m.haves, key)
			return
		}
		h, err := delivery.DecodeHave(e.Value)
		m.haves[key] = haveEntry{have: h, err: errText(err), at: e.At}
	case delivery.PresenceBucket:
		if e.Deleted {
			delete(m.presence, e.Key)
			return
		}
		p, err := delivery.DecodePresence(e.Value)
		m.presence[e.Key] = presenceEntry{value: p, err: errText(err), at: e.At}
	case delivery.StuckBucket:
		if e.Deleted {
			delete(m.stuck, e.Key)
			return
		}
		st, err := delivery.DecodeStuck(e.Value)
		m.stuck[e.Key] = stuckEntry{value: st, err: errText(err)}
	case AssignmentsBucket:
		if e.Deleted {
			delete(m.assigns, e.Key)
			return
		}
		var a assignment
		if json.Unmarshal(e.Value, &a) != nil || a.Node == "" || a.Module == "" {
			return
		}
		m.assigns[a.Node+"."+a.Module] = a
	case BuildsBucket:
		module, rest, ok := strings.Cut(e.Key, ".")
		if !ok {
			return
		}
		switch {
		case rest == "newest":
			if e.Deleted {
				delete(m.newest, module)
				return
			}
			m.newest[module] = strings.TrimSpace(string(e.Value))
		case strings.HasPrefix(rest, "build."):
			if e.Deleted {
				delete(m.builds, e.Key)
				return
			}
			var b build
			if json.Unmarshal(e.Value, &b) != nil {
				return
			}
			m.builds[e.Key] = b
		}
	}
}

func errText(err error) string {
	if err == nil {
		return ""
	}
	return err.Error()
}

// Snapshot derives one row per module and node with an open assignment, a row marked left behind for each have with
// none, and a row per node's own key. Rows are ordered by node and module.
func (m *Model) Snapshot() Snapshot {
	m.mu.Lock()
	defer m.mu.Unlock()
	rows := []Row{}
	for key, a := range m.assigns {
		if !a.open() {
			continue
		}
		r := Row{Node: a.Node, Module: a.Module, Status: StatusAssigned, Assigned: a.Build}
		if a.PulledIn {
			r.RequiredBy = a.RequiredBy
		}
		var waits []string
		h, has := m.haves[key]
		switch {
		case !has:
			waits = append(waits, "no have yet")
		case h.err != "":
			r.Since, r.Error = h.at, h.err
		default:
			r.Have, r.Phase, r.Health = h.have.Build, string(h.have.Phase), string(h.have.Health)
			r.Since, r.Detail = h.at, h.have.Detail
			if a.Build != "" && h.have.Build != a.Build {
				waits = append(waits, "the node has not applied "+a.Build)
			}
		}
		if id := m.newest[a.Module]; id != "" && a.Build != "" && id != a.Build {
			if b, ok := m.builds[a.Module+".build."+id]; ok && b.done() {
				waits = append(waits, a.Module+" has a newer build "+id+" not yet assigned")
			}
		}
		r.Waits = strings.Join(waits, "; ")
		rows = append(rows, r)
	}
	for key, h := range m.haves {
		if a, ok := m.assigns[key]; ok && a.open() {
			continue
		}
		node, module, _, _ := delivery.ParseHaveKey(key)
		r := Row{Node: node, Module: module, Status: StatusLeftBehind, Since: h.at, Error: h.err}
		if h.err == "" {
			r.Have, r.Phase, r.Health, r.Detail = h.have.Build, string(h.have.Phase), string(h.have.Health), h.have.Detail
		}
		rows = append(rows, r)
	}
	sort.Slice(rows, func(i, j int) bool {
		if rows[i].Node != rows[j].Node {
			return rows[i].Node < rows[j].Node
		}
		return rows[i].Module < rows[j].Module
	})
	nodes := []NodeRow{}
	for name, n := range m.nodes {
		nodes = append(nodes, NodeRow{Node: name, Windows: count(n.node.Windows), Busy: count(n.node.Busy),
			Strays: count(n.node.Strays), Since: n.at, Error: n.err})
	}
	sort.Slice(nodes, func(i, j int) bool { return nodes[i].Node < nodes[j].Node })
	offline, stuck, unreadable := []OfflineRow{}, []StuckRow{}, []UnreadableRow{}
	for key, p := range m.presence {
		if p.err != "" {
			unreadable = append(unreadable, UnreadableRow{Bucket: delivery.PresenceBucket, Key: key, Error: p.err})
			continue
		}
		if p.value.Online {
			continue
		}
		node, part, instance, err := delivery.ParsePresenceKey(key)
		if err != nil {
			unreadable = append(unreadable, UnreadableRow{Bucket: delivery.PresenceBucket, Key: key, Error: err.Error()})
			continue
		}
		if part != "" {
			offline = append(offline, OfflineRow{Kind: "part", Part: part, Instance: instance, Since: p.at})
		} else {
			offline = append(offline, OfflineRow{Kind: "node", Node: node, Since: p.at})
		}
	}
	for key, st := range m.stuck {
		if st.err != "" {
			unreadable = append(unreadable, UnreadableRow{Bucket: delivery.StuckBucket, Key: key, Error: st.err})
			continue
		}
		if st.value.Moved {
			continue
		}
		stuck = append(stuck, StuckRow{Bucket: st.value.Bucket, Key: st.value.Key, Waited: st.value.Waited, Since: st.value.Since})
	}
	sort.Slice(offline, func(i, j int) bool {
		a, b := offline[i], offline[j]
		return a.Kind+a.Node+a.Part+a.Instance < b.Kind+b.Node+b.Part+b.Instance
	})
	sort.Slice(stuck, func(i, j int) bool {
		return stuck[i].Bucket+"."+stuck[i].Key < stuck[j].Bucket+"."+stuck[j].Key
	})
	sort.Slice(unreadable, func(i, j int) bool {
		return unreadable[i].Bucket+"."+unreadable[i].Key < unreadable[j].Bucket+"."+unreadable[j].Key
	})
	return Snapshot{Rows: rows, Nodes: nodes, Offline: offline, Stuck: stuck, Unreadable: unreadable}
}

// count is the number of items of a carried JSON array or object, and 0 for anything else.
func count(raw json.RawMessage) int {
	var list []json.RawMessage
	if json.Unmarshal(raw, &list) == nil {
		return len(list)
	}
	var obj map[string]json.RawMessage
	if json.Unmarshal(raw, &obj) == nil {
		return len(obj)
	}
	return 0
}

// Counted is the number of rows that are modules on the nodes, not left behind.
func (s Snapshot) Counted() int {
	n := 0
	for _, r := range s.Rows {
		if r.Status == StatusAssigned {
			n++
		}
	}
	return n
}

// The buckets the view reads besides the have bucket (which mesh-sdk's delivery package names).
const (
	AssignmentsBucket = "mesh-controller_assignments"
	BuildsBucket      = "mesh-controller_builds"
)

// The statuses of a row.
const (
	StatusAssigned   = "assigned"
	StatusLeftBehind = "left behind"
)