Implementing more advanced feature: audit and more

This commit is contained in:
Martino Ferrari
2026-06-19 14:17:46 +02:00
parent 8f6dbcba49
commit 901b87d407
31 changed files with 1669 additions and 569 deletions
+65
View File
@@ -0,0 +1,65 @@
// Package audit records system-affecting actions (signal writes by users or the
// control-logic engine, interface and control-logic mutations) to an append-only
// SQLite log that audit staff can query later. It is enabled via [audit] in the
// config; when disabled a no-op Recorder is used so call sites need no guards.
package audit
import "time"
// Actor types distinguish human-initiated actions from automated ones.
const (
ActorUser = "user" // action attributed to an authenticated end-user
ActorSystem = "system" // action performed by the control-logic engine
)
// Outcomes record whether the action succeeded.
const (
OutcomeOK = "ok"
OutcomeError = "error"
)
// Event is a single audit record. Optional fields are omitted from the database
// when empty. Time defaults to the current time when zero.
type Event struct {
Time time.Time `json:"time"`
Actor string `json:"actor"` // username, or graph name for system actions
ActorType string `json:"actorType"` // ActorUser | ActorSystem
Action string `json:"action"` // e.g. "signal.write", "interface.update"
DS string `json:"ds,omitempty"` // data source (signal writes)
Signal string `json:"signal,omitempty"` // signal / target name
Value string `json:"value,omitempty"` // serialised written value
Detail string `json:"detail,omitempty"` // free-form context (id, graph, trigger)
IP string `json:"ip,omitempty"` // client address (user actions)
Outcome string `json:"outcome"` // OutcomeOK | OutcomeError
Error string `json:"error,omitempty"` // message when Outcome==OutcomeError
}
// Filter selects a subset of events for Query. Zero-valued fields are ignored.
type Filter struct {
Start time.Time
End time.Time
Actor string
Action string
DS string
Signal string
Limit int // <=0 means a default cap is applied
}
// Recorder appends events and answers queries. Implementations must be safe for
// concurrent use and Record must never block the caller for long.
type Recorder interface {
Record(Event)
Query(Filter) ([]Event, error)
Close() error
}
// nopRecorder is used when auditing is disabled. It discards every event.
type nopRecorder struct{}
func (nopRecorder) Record(Event) {}
func (nopRecorder) Query(Filter) ([]Event, error) { return []Event{}, nil }
func (nopRecorder) Close() error { return nil }
// Nop returns a Recorder that discards everything. Call sites can hold a Recorder
// unconditionally and call Record without checking whether auditing is enabled.
func Nop() Recorder { return nopRecorder{} }
+183
View File
@@ -0,0 +1,183 @@
package audit
import (
"database/sql"
"fmt"
"log/slog"
"strings"
"sync"
"time"
_ "modernc.org/sqlite" // pure-Go SQLite driver (no CGo, keeps the single static binary)
)
// defaultQueryLimit caps how many rows Query returns when the filter does not
// specify a smaller limit, protecting the API and UI from unbounded result sets.
const defaultQueryLimit = 1000
// writeBuffer is the depth of the async insert queue. When full, Record falls
// back to a synchronous insert so events are never silently dropped.
const writeBuffer = 4096
// sqliteRecorder appends events to a SQLite database. Inserts are normally
// handled by a background goroutine so Record does not block the calling write
// path; the queue has a synchronous fallback to guarantee durability under load.
type sqliteRecorder struct {
db *sql.DB
log *slog.Logger
ch chan Event
wg sync.WaitGroup
closed chan struct{}
once sync.Once
}
// NewSQLite opens (creating if needed) the audit database at path and starts the
// background writer. The returned Recorder must be Closed on shutdown.
func NewSQLite(path string, log *slog.Logger) (Recorder, error) {
// WAL + a busy timeout let the async writer and synchronous query/fallback
// paths share the file without "database is locked" errors.
dsn := fmt.Sprintf("file:%s?_pragma=busy_timeout(5000)&_pragma=journal_mode(WAL)&_pragma=synchronous(NORMAL)", path)
db, err := sql.Open("sqlite", dsn)
if err != nil {
return nil, fmt.Errorf("open audit db: %w", err)
}
if _, err := db.Exec(schema); err != nil {
_ = db.Close()
return nil, fmt.Errorf("init audit schema: %w", err)
}
r := &sqliteRecorder{
db: db,
log: log,
ch: make(chan Event, writeBuffer),
closed: make(chan struct{}),
}
r.wg.Add(1)
go r.writeLoop()
return r, nil
}
const schema = `
CREATE TABLE IF NOT EXISTS audit_log (
id INTEGER PRIMARY KEY AUTOINCREMENT,
ts_ns INTEGER NOT NULL,
actor TEXT NOT NULL,
actor_type TEXT NOT NULL,
action TEXT NOT NULL,
ds TEXT NOT NULL DEFAULT '',
signal TEXT NOT NULL DEFAULT '',
value TEXT NOT NULL DEFAULT '',
detail TEXT NOT NULL DEFAULT '',
ip TEXT NOT NULL DEFAULT '',
outcome TEXT NOT NULL DEFAULT 'ok',
error TEXT NOT NULL DEFAULT ''
);
CREATE INDEX IF NOT EXISTS idx_audit_ts ON audit_log(ts_ns);
CREATE INDEX IF NOT EXISTS idx_audit_actor ON audit_log(actor);
CREATE INDEX IF NOT EXISTS idx_audit_action ON audit_log(action);
`
func (r *sqliteRecorder) Record(e Event) {
if e.Time.IsZero() {
e.Time = time.Now()
}
if e.Outcome == "" {
e.Outcome = OutcomeOK
}
select {
case <-r.closed:
// Recorder is shutting down; best-effort synchronous insert.
r.insert(e)
case r.ch <- e:
default:
// Queue full: write synchronously rather than drop an audit record.
r.insert(e)
}
}
func (r *sqliteRecorder) writeLoop() {
defer r.wg.Done()
for e := range r.ch {
r.insert(e)
}
}
func (r *sqliteRecorder) insert(e Event) {
_, err := r.db.Exec(
`INSERT INTO audit_log (ts_ns, actor, actor_type, action, ds, signal, value, detail, ip, outcome, error)
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,
e.Time.UnixNano(), e.Actor, e.ActorType, e.Action,
e.DS, e.Signal, e.Value, e.Detail, e.IP, e.Outcome, e.Error,
)
if err != nil {
r.log.Error("audit: insert failed", "action", e.Action, "err", err)
}
}
func (r *sqliteRecorder) Query(f Filter) ([]Event, error) {
var where []string
var args []any
if !f.Start.IsZero() {
where = append(where, "ts_ns >= ?")
args = append(args, f.Start.UnixNano())
}
if !f.End.IsZero() {
where = append(where, "ts_ns <= ?")
args = append(args, f.End.UnixNano())
}
if f.Actor != "" {
where = append(where, "actor = ?")
args = append(args, f.Actor)
}
if f.Action != "" {
where = append(where, "action = ?")
args = append(args, f.Action)
}
if f.DS != "" {
where = append(where, "ds = ?")
args = append(args, f.DS)
}
if f.Signal != "" {
where = append(where, "signal LIKE ?")
args = append(args, "%"+f.Signal+"%")
}
limit := f.Limit
if limit <= 0 || limit > defaultQueryLimit {
limit = defaultQueryLimit
}
q := "SELECT ts_ns, actor, actor_type, action, ds, signal, value, detail, ip, outcome, error FROM audit_log"
if len(where) > 0 {
q += " WHERE " + strings.Join(where, " AND ")
}
q += " ORDER BY ts_ns DESC LIMIT ?"
args = append(args, limit)
rows, err := r.db.Query(q, args...)
if err != nil {
return nil, fmt.Errorf("query audit log: %w", err)
}
defer rows.Close()
out := []Event{}
for rows.Next() {
var e Event
var tsNs int64
if err := rows.Scan(&tsNs, &e.Actor, &e.ActorType, &e.Action,
&e.DS, &e.Signal, &e.Value, &e.Detail, &e.IP, &e.Outcome, &e.Error); err != nil {
return nil, fmt.Errorf("scan audit row: %w", err)
}
e.Time = time.Unix(0, tsNs)
out = append(out, e)
}
return out, rows.Err()
}
func (r *sqliteRecorder) Close() error {
r.once.Do(func() {
close(r.closed)
close(r.ch)
})
r.wg.Wait()
return r.db.Close()
}
+68
View File
@@ -0,0 +1,68 @@
package audit
import (
"log/slog"
"path/filepath"
"testing"
"time"
)
func TestSQLiteRoundTrip(t *testing.T) {
path := filepath.Join(t.TempDir(), "audit.db")
rec, err := NewSQLite(path, slog.Default())
if err != nil {
t.Fatal("NewSQLite:", err)
}
defer rec.Close()
base := time.Now()
rec.Record(Event{Time: base, Actor: "alice", ActorType: ActorUser, Action: "signal.write", DS: "epics", Signal: "PV:A", Value: "1.5"})
rec.Record(Event{Time: base.Add(time.Second), Actor: "flow1", ActorType: ActorSystem, Action: "signal.write", DS: "epics", Signal: "PV:B", Value: "0"})
rec.Record(Event{Time: base.Add(2 * time.Second), Actor: "bob", ActorType: ActorUser, Action: "interface.update", Detail: "panel-1", Outcome: OutcomeError, Error: "denied"})
// Flush the async writer.
if err := rec.Close(); err != nil {
t.Fatal("Close:", err)
}
rec, err = NewSQLite(path, slog.Default())
if err != nil {
t.Fatal("reopen:", err)
}
defer rec.Close()
all, err := rec.Query(Filter{})
if err != nil {
t.Fatal("Query:", err)
}
if len(all) != 3 {
t.Fatalf("got %d events, want 3", len(all))
}
// Newest first.
if all[0].Actor != "bob" {
t.Errorf("first actor = %q, want bob", all[0].Actor)
}
byActor, err := rec.Query(Filter{Actor: "alice"})
if err != nil {
t.Fatal("Query actor:", err)
}
if len(byActor) != 1 || byActor[0].Signal != "PV:A" {
t.Errorf("actor filter = %+v, want one PV:A event", byActor)
}
byAction, err := rec.Query(Filter{Action: "signal.write"})
if err != nil {
t.Fatal("Query action:", err)
}
if len(byAction) != 2 {
t.Errorf("action filter returned %d, want 2", len(byAction))
}
since, err := rec.Query(Filter{Start: base.Add(1500 * time.Millisecond)})
if err != nil {
t.Fatal("Query start:", err)
}
if len(since) != 1 || since[0].Actor != "bob" {
t.Errorf("time filter = %+v, want one bob event", since)
}
}