feat(monitor): persist health and recovery cooldown

Co-authored-by: GPT-6 <gpt-6@openai.com>
This commit is contained in:
ZeusinaandGPT-6 Bot committed 2026-10-10 08:04:48 +03:00
1 parent 5c540ce03b
commit 0c51fd0196
9 files changed
+497 -9

No files matched your search

+2 -2
View File
@@ -58,7 +58,7 @@ func populatedV2Store(t *testing.T) *sql.DB {
if err := migrate(db); err != nil {
t.Fatal(err)
}
if _, err := db.Exec(`DROP TABLE candidate_evidence; DROP TABLE strategy_candidates; PRAGMA user_version = 2;`); err != nil {
if _, err := db.Exec(`DROP TABLE monitor_state; DROP TABLE candidate_evidence; DROP TABLE strategy_candidates; PRAGMA user_version = 2;`); err != nil {
t.Fatal(err)
}
return db
@@ -99,7 +99,7 @@ func TestPopulatedV2CandidateMigration(t *testing.T) {
}
}
var version int
if err := db.QueryRow("PRAGMA user_version").Scan(&version); err != nil || version != 3 {
if err := db.QueryRow("PRAGMA user_version").Scan(&version); err != nil || version != 4 {
t.Fatalf("version=%d err=%v", version, err)
}
if before != legacySnapshot(t, db) {
+1 -1
View File
@@ -117,7 +117,7 @@ func TestPopulatedV1FingerprintMigration(t *testing.T) {
}
assertRowCounts(t, db, 3, 4)
var version int
if err := db.QueryRow("PRAGMA user_version").Scan(&version); err != nil || version != 3 {
if err := db.QueryRow("PRAGMA user_version").Scan(&version); err != nil || version != 4 {
t.Fatalf("version=%d err=%v", version, err)
}
for id, input := range map[int]string{7: "--lua-desync=fake", 9: args} {
+1 -1
View File
@@ -154,7 +154,7 @@ fi
if err := db.QueryRow(`PRAGMA user_version`).Scan(&version); err != nil {
t.Fatal(err)
}
if evidence != 2 || legacy != 0 || version != 3 {
if evidence != 2 || legacy != 0 || version != 4 {
t.Fatalf("evidence=%d legacy=%d version=%d", evidence, legacy, version)
}
for _, command := range [][]string{{"history"}, {"strategies", "youtube"}} {
+142
View File
@@ -0,0 +1,142 @@
package main
import (
"crypto/rand"
"database/sql"
"encoding/hex"
"fmt"
"strings"
"time"
)
// Policy foundation only; service_id identifies a service, not a rotating host
// or proxy address. No probe, scheduler or configuration changes happen here.
const monitorSchema = `
CREATE TABLE monitor_state (
service_id TEXT PRIMARY KEY NOT NULL CHECK (typeof(service_id) = 'text' AND length(trim(service_id)) > 0),
consecutive_failures INTEGER NOT NULL DEFAULT 0 CHECK (typeof(consecutive_failures) = 'integer' AND consecutive_failures >= 0),
last_probe_at INTEGER CHECK (last_probe_at IS NULL OR typeof(last_probe_at) = 'integer'),
recovery_running INTEGER NOT NULL DEFAULT 0 CHECK (typeof(recovery_running) = 'integer' AND recovery_running IN (0, 1)),
recovery_token TEXT CHECK (recovery_token IS NULL OR (typeof(recovery_token) = 'text' AND length(recovery_token) = 64)),
cooldown_until INTEGER CHECK (cooldown_until IS NULL OR typeof(cooldown_until) = 'integer'),
CHECK ((recovery_running = 0 AND recovery_token IS NULL) OR (recovery_running = 1 AND recovery_token IS NOT NULL)),
CHECK (last_probe_at IS NOT NULL OR consecutive_failures = 0)
);
PRAGMA user_version = 4;`
type monitorState struct {
ConsecutiveFailures int
LastProbeAt time.Time
RecoveryRunning bool
CooldownUntil time.Time
}
func (s monitorState) healthStatus() string {
if s.LastProbeAt.IsZero() {
return "UNKNOWN"
}
if s.ConsecutiveFailures > 0 {
return "UNHEALTHY"
}
return "HEALTHY"
}
func validateMonitorTime(serviceID string, now time.Time) error {
if strings.TrimSpace(serviceID) == "" || strings.TrimSpace(serviceID) != serviceID || strings.ContainsRune(serviceID, '\x00') {
return fmt.Errorf("invalid monitor service ID")
}
// Reject zero/overflowing timestamps rather than silently wrapping UnixNano.
if now.IsZero() || !time.Unix(0, now.UnixNano()).Equal(now) {
return fmt.Errorf("invalid monitor timestamp")
}
return nil
}
func recordMonitorProbe(db *sql.DB, serviceID string, success bool, now time.Time) error {
if err := validateMonitorTime(serviceID, now); err != nil {
return err
}
failures := 1
if success {
failures = 0
}
// One UPSERT serializes increments and leaves the recovery lock/cooldown intact.
_, err := db.Exec(`INSERT INTO monitor_state (service_id, consecutive_failures, last_probe_at)
VALUES (?, ?, ?) ON CONFLICT(service_id) DO UPDATE SET
consecutive_failures = CASE WHEN excluded.consecutive_failures = 0 THEN 0 ELSE monitor_state.consecutive_failures + 1 END,
last_probe_at = excluded.last_probe_at`, serviceID, failures, now.UnixNano())
return err
}
func readMonitorState(db *sql.DB, serviceID string) (monitorState, error) {
var state monitorState
var probe, cooldown sql.NullInt64
err := db.QueryRow(`SELECT consecutive_failures, last_probe_at, recovery_running, cooldown_until
FROM monitor_state WHERE service_id = ?`, serviceID).Scan(&state.ConsecutiveFailures, &probe, &state.RecoveryRunning, &cooldown)
if err == sql.ErrNoRows {
return state, nil // A service without recorded probes has unknown health.
}
if probe.Valid {
state.LastProbeAt = time.Unix(0, probe.Int64).UTC()
}
if cooldown.Valid {
state.CooldownUntil = time.Unix(0, cooldown.Int64).UTC()
}
return state, err
}
// A nonempty token identifies this acquisition only; an empty token means busy
// or ineligible. The caller must retain it for finalization, including after reopen.
func beginMonitorRecovery(db *sql.DB, serviceID string, threshold int, now time.Time) (string, error) {
if err := validateMonitorTime(serviceID, now); err != nil {
return "", err
}
if threshold <= 0 {
return "", fmt.Errorf("failure threshold must be positive")
}
var ownership [32]byte
if _, err := rand.Read(ownership[:]); err != nil {
return "", err
}
token := hex.EncodeToString(ownership[:])
// Compare-and-set in one SQLite write, including across independent processes.
result, err := db.Exec(`UPDATE monitor_state SET recovery_running = 1, recovery_token = ?
WHERE service_id = ? AND consecutive_failures >= ? AND recovery_running = 0
AND (cooldown_until IS NULL OR cooldown_until <= ?)`, token, serviceID, threshold, now.UnixNano())
if err != nil {
return "", err
}
count, err := result.RowsAffected()
if err != nil || count != 1 {
return "", err
}
return token, nil
}
// Call on either success or failure. A crash leaves the durable lock held;
// automatic crash reconciliation is deliberately not part of this foundation.
func finishMonitorRecovery(db *sql.DB, serviceID, token string, completed time.Time, cooldown time.Duration) error {
if err := validateMonitorTime(serviceID, completed); err != nil {
return err
}
if cooldown <= 0 {
return fmt.Errorf("recovery cooldown must be positive")
}
if len(token) != 64 {
return fmt.Errorf("invalid recovery ownership token")
}
expires := completed.Add(cooldown)
if err := validateMonitorTime(serviceID, expires); err != nil {
return err
}
result, err := db.Exec(`UPDATE monitor_state SET recovery_running = 0, recovery_token = NULL, cooldown_until = ?
WHERE service_id = ? AND recovery_running = 1 AND recovery_token = ?`, expires.UnixNano(), serviceID, token)
if err != nil {
return err
}
count, err := result.RowsAffected()
if err == nil && count != 1 {
return fmt.Errorf("no recovery owned by this token for service %q", serviceID)
}
return err
}
+325
View File
@@ -0,0 +1,325 @@
package main
import (
"database/sql"
"fmt"
"path/filepath"
"sync"
"testing"
"time"
)
var monitorTestTime = time.Date(2026, 10, 10, 12, 0, 0, 123, time.UTC)
func requireMonitorBegin(t *testing.T, db *sql.DB, id string, now time.Time, want bool) string {
t.Helper()
got, err := beginMonitorRecovery(db, id, 3, now)
if err != nil || (got != "") != want {
t.Fatalf("begin %s: got=%v want=%v err=%v", id, got, want, err)
}
return got
}
func TestMonitorProbeStreakAndThreshold(t *testing.T) {
db := testStore(t)
now := monitorTestTime
state, err := readMonitorState(db, "new-test-service")
if err != nil || state.healthStatus() != "UNKNOWN" {
t.Fatalf("initial=%+v err=%v", state, err)
}
requireMonitorBegin(t, db, "new-test-service", now, false)
for i := 1; i <= 3; i++ {
if err := recordMonitorProbe(db, "new-test-service", false, now); err != nil {
t.Fatal(err)
}
state, err := readMonitorState(db, "new-test-service")
if err != nil || state.ConsecutiveFailures != i || state.healthStatus() != "UNHEALTHY" || !state.LastProbeAt.Equal(now) {
t.Fatalf("probe %d: %+v err=%v", i, state, err)
}
if i < 3 {
requireMonitorBegin(t, db, "new-test-service", now, false)
}
}
requireMonitorBegin(t, db, "other-service", now, false)
token := requireMonitorBegin(t, db, "new-test-service", now, true)
requireMonitorBegin(t, db, "new-test-service", now, false)
if err := recordMonitorProbe(db, "new-test-service", true, now.Add(time.Minute)); err != nil {
t.Fatal(err)
}
state, err = readMonitorState(db, "new-test-service")
if err != nil || state.ConsecutiveFailures != 0 || state.healthStatus() != "HEALTHY" || !state.RecoveryRunning {
t.Fatalf("success=%+v err=%v", state, err)
}
if err := finishMonitorRecovery(db, "new-test-service", token, now.Add(2*time.Minute), 30*time.Minute); err != nil {
t.Fatal(err)
}
requireMonitorBegin(t, db, "new-test-service", now.Add(time.Hour), false)
if err := recordMonitorProbe(db, "new-test-service", false, now.Add(time.Hour)); err != nil {
t.Fatal(err)
}
state, err = readMonitorState(db, "new-test-service")
if err != nil || state.ConsecutiveFailures != 1 || state.CooldownUntil.IsZero() {
t.Fatalf("new streak=%+v err=%v", state, err)
}
}
func TestMonitorRecoveryDurableCooldownForEitherOutcome(t *testing.T) {
for _, outcome := range []string{"success", "failure"} {
t.Run(outcome, func(t *testing.T) {
path := filepath.Join(t.TempDir(), "monitor.db")
db, err := openStore(path)
if err != nil {
t.Fatal(err)
}
t.Cleanup(func() { db.Close() })
reopen := func() {
t.Helper()
if err := db.Close(); err != nil {
t.Fatal(err)
}
db, err = openStore(path)
if err != nil {
t.Fatal(err)
}
}
now := monitorTestTime
for i := 0; i < 3; i++ {
if err := recordMonitorProbe(db, "service", false, now); err != nil {
t.Fatal(err)
}
}
token := requireMonitorBegin(t, db, "service", now, true)
reopen()
requireMonitorBegin(t, db, "service", now.Add(time.Hour), false)
// Outcome deliberately does not affect cooldown; completion time does.
completed := now.Add(10 * time.Minute)
if err := finishMonitorRecovery(db, "service", token, completed, 30*time.Minute); err != nil {
t.Fatal(err)
}
if err := finishMonitorRecovery(db, "service", token, completed.Add(time.Minute), 30*time.Minute); err == nil {
t.Fatal("repeated finish accepted")
}
reopen()
state, err := readMonitorState(db, "service")
expires := completed.Add(30 * time.Minute)
if err != nil || state.RecoveryRunning || state.ConsecutiveFailures != 3 || !state.CooldownUntil.Equal(expires) {
t.Fatalf("reloaded=%+v err=%v", state, err)
}
requireMonitorBegin(t, db, "service", expires.Add(-time.Nanosecond), false)
requireMonitorBegin(t, db, "service", expires, true)
})
}
}
func TestMonitorStaleFinishCannotReleaseNewRecovery(t *testing.T) {
db := testStore(t)
now := monitorTestTime
for i := 0; i < 3; i++ {
if err := recordMonitorProbe(db, "service", false, now); err != nil {
t.Fatal(err)
}
}
tokenA := requireMonitorBegin(t, db, "service", now, true)
if err := finishMonitorRecovery(db, "service", tokenA, now, time.Minute); err != nil {
t.Fatal(err)
}
tokenB := requireMonitorBegin(t, db, "service", now.Add(time.Minute), true)
if tokenA == tokenB {
t.Fatal("new acquisition reused ownership token")
}
before, err := readMonitorState(db, "service")
if err != nil || !before.RecoveryRunning {
t.Fatalf("before stale finish: %+v err=%v", before, err)
}
for _, token := range []string{tokenA, ""} {
if err := finishMonitorRecovery(db, "service", token, now.Add(2*time.Minute), time.Hour); err == nil {
t.Fatal("unowned finish accepted")
}
after, err := readMonitorState(db, "service")
if err != nil || after != before {
t.Fatalf("unowned finish changed state: before=%+v after=%+v err=%v", before, after, err)
}
var storedToken string
if err := db.QueryRow("SELECT recovery_token FROM monitor_state WHERE service_id = ?", "service").Scan(&storedToken); err != nil || storedToken != tokenB {
t.Fatalf("new ownership changed: token=%q err=%v", storedToken, err)
}
requireMonitorBegin(t, db, "service", now.Add(24*time.Hour), false)
}
if err := finishMonitorRecovery(db, "service", tokenB, now.Add(2*time.Minute), time.Minute); err != nil {
t.Fatalf("current owner cannot finish: %v", err)
}
}
func TestMonitorConcurrentWrites(t *testing.T) {
path := filepath.Join(t.TempDir(), "monitor.db")
stores := make([]*sql.DB, 2)
for i := range stores {
db, err := openStore(path)
if err != nil {
t.Fatal(err)
}
defer db.Close()
stores[i] = db
}
const callers = 12
var wg sync.WaitGroup
for i := 0; i < callers; i++ {
wg.Add(1)
go func(db *sql.DB) {
defer wg.Done()
if err := recordMonitorProbe(db, "service", false, monitorTestTime); err != nil {
t.Error(err)
}
}(stores[i%2])
}
wg.Wait()
state, err := readMonitorState(stores[0], "service")
if err != nil || state.ConsecutiveFailures != callers {
t.Fatalf("increments lost: %+v err=%v", state, err)
}
results := make(chan bool, callers)
start := make(chan struct{})
for i := 0; i < callers; i++ {
wg.Add(1)
go func(db *sql.DB) {
defer wg.Done()
<-start
got, err := beginMonitorRecovery(db, "service", 3, monitorTestTime)
if err != nil {
t.Error(err)
}
results <- got != ""
}(stores[i%2])
}
close(start)
wg.Wait()
close(results)
winners := 0
for got := range results {
if got {
winners++
}
}
if winners != 1 {
t.Fatalf("recovery winners=%d", winners)
}
}
func TestMonitorValidationAndConstraints(t *testing.T) {
db := testStore(t)
for _, id := range []string{"", " ", " service", "service\x00"} {
if err := recordMonitorProbe(db, id, false, monitorTestTime); err == nil {
t.Fatalf("accepted ID %q", id)
}
}
for _, now := range []time.Time{{}, time.Date(2500, 1, 1, 0, 0, 0, 0, time.UTC)} {
if err := recordMonitorProbe(db, "service", false, now); err == nil {
t.Fatal("accepted invalid timestamp")
}
}
for _, threshold := range []int{0, -1} {
if _, err := beginMonitorRecovery(db, "service", threshold, monitorTestTime); err == nil {
t.Fatal("accepted invalid threshold")
}
}
for _, cooldown := range []time.Duration{0, -time.Minute} {
if err := finishMonitorRecovery(db, "service", "0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef", monitorTestTime, cooldown); err == nil {
t.Fatal("accepted invalid cooldown")
}
}
if err := finishMonitorRecovery(db, "missing", "0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef", monitorTestTime, time.Minute); err == nil {
t.Fatal("finished nonexistent recovery")
}
for _, statement := range []string{
`INSERT INTO monitor_state (service_id) VALUES ('')`,
`INSERT INTO monitor_state (service_id) VALUES (NULL)`,
`INSERT INTO monitor_state (service_id, consecutive_failures) VALUES ('x', -1)`,
`INSERT INTO monitor_state (service_id, consecutive_failures) VALUES ('x', 1)`,
`INSERT INTO monitor_state (service_id, recovery_running) VALUES ('x', 2)`,
`INSERT INTO monitor_state (service_id, recovery_running) VALUES ('x', 1)`,
`INSERT INTO monitor_state (service_id, recovery_token) VALUES ('x', 'bad')`,
`INSERT INTO monitor_state (service_id, recovery_token) VALUES ('x', '0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef')`,
`INSERT INTO monitor_state (service_id, last_probe_at) VALUES ('x', 'bad')`,
`INSERT INTO monitor_state (service_id, cooldown_until) VALUES ('x', 'bad')`,
} {
if _, err := db.Exec(statement); err == nil {
t.Fatalf("constraint accepted %s", statement)
}
}
}
func monitorMigrationSnapshot(t *testing.T, db *sql.DB) string {
t.Helper()
snapshot := legacySnapshot(t, db)
for _, table := range []string{"strategy_candidates", "candidate_evidence", "googlevideo_tune_runs"} {
rows, err := db.Query("SELECT * FROM " + table + " ORDER BY 1, 2, 3")
if err != nil {
t.Fatal(err)
}
columns, err := rows.Columns()
if err != nil {
rows.Close()
t.Fatal(err)
}
values := make([]interface{}, len(columns))
pointers := make([]interface{}, len(columns))
for i := range values {
pointers[i] = &values[i]
}
for rows.Next() {
if err := rows.Scan(pointers...); err != nil {
rows.Close()
t.Fatal(err)
}
snapshot += fmt.Sprintf("%s:%#v;", table, values)
}
err = rows.Err()
rows.Close()
if err != nil {
t.Fatal(err)
}
}
return snapshot
}
func TestMonitorV3MigrationPreservesHistory(t *testing.T) {
db := testStore(t)
saveProposal(t, db, "--lua-desync=fake")
if _, err := db.Exec(`INSERT INTO googlevideo_tune_runs (started_at, finished_at, state, error)
VALUES ('start', 'finish', 'FAILED', 'original failure');
DROP TABLE monitor_state; PRAGMA user_version = 3;`); err != nil {
t.Fatal(err)
}
before := monitorMigrationSnapshot(t, db)
for i := 0; i < 2; i++ {
if err := migrate(db); err != nil {
t.Fatal(err)
}
}
var version, count int
if err := db.QueryRow("PRAGMA user_version").Scan(&version); err != nil || version != 4 {
t.Fatalf("version=%d err=%v", version, err)
}
if err := db.QueryRow("SELECT count(*) FROM monitor_state").Scan(&count); err != nil || count != 0 {
t.Fatalf("monitor count=%d err=%v", count, err)
}
if after := monitorMigrationSnapshot(t, db); before != after {
t.Fatal("v3 history/candidate data changed")
}
}
func TestMonitorV3MigrationRollback(t *testing.T) {
db := testStore(t)
saveProposal(t, db, "--lua-desync=fake")
if _, err := db.Exec(`DROP TABLE monitor_state; CREATE TABLE monitor_state (conflict INTEGER); PRAGMA user_version = 3;`); err != nil {
t.Fatal(err)
}
before := monitorMigrationSnapshot(t, db)
if err := migrate(db); err == nil {
t.Fatal("expected conflicting v4 schema failure")
}
var version int
if err := db.QueryRow("PRAGMA user_version").Scan(&version); err != nil || version != 3 || before != monitorMigrationSnapshot(t, db) {
t.Fatalf("failed migration changed data/version: %d err=%v", version, err)
}
}
+6 -1
View File
@@ -114,7 +114,7 @@ func migrate(db *sql.DB) error {
if err = conn.QueryRowContext(ctx, "PRAGMA user_version").Scan(&version); err != nil {
return err
}
if version > 3 {
if version > 4 {
return fmt.Errorf("unsupported database schema version %d", version)
}
if version == 0 {
@@ -231,6 +231,11 @@ PRAGMA user_version = 1;`)
if _, err := conn.ExecContext(ctx, googlevideoRunSchema); err != nil {
return err
}
if version < 4 {
if _, err := conn.ExecContext(ctx, monitorSchema); err != nil {
return err
}
}
_, err = conn.ExecContext(ctx, "COMMIT")
return err
}
+2 -2
View File
@@ -224,7 +224,7 @@ func TestConcurrentFirstOpens(t *testing.T) {
}
for _, db := range stores {
var version, objects int
if err := db.QueryRow("PRAGMA user_version").Scan(&version); err != nil || version != 3 {
if err := db.QueryRow("PRAGMA user_version").Scan(&version); err != nil || version != 4 {
t.Fatalf("version=%d err=%v", version, err)
}
if err := db.QueryRow(`SELECT count(*) FROM sqlite_master WHERE
@@ -377,7 +377,7 @@ func TestSchemaConstraintsAndIncompleteEvidence(t *testing.T) {
func TestUnsupportedSchemaVersion(t *testing.T) {
db := testStore(t)
if _, err := db.Exec("PRAGMA user_version = 4"); err != nil {
if _, err := db.Exec("PRAGMA user_version = 5"); err != nil {
t.Fatal(err)
}
if err := migrate(db); err == nil || !strings.Contains(err.Error(), "unsupported") {