Files
yovision/Bell/internal/store/alert_postgres.go
T
QiuSW 8208118904
Harness governance / validate (pull_request) Has been cancelled
feat(bell): add alert acknowledgement vertical slice
2026-08-11 17:01:53 +08:00

356 lines
16 KiB
Go

package store
import (
"bytes"
"context"
"crypto/sha256"
"database/sql"
"encoding/json"
"errors"
"fmt"
"strings"
"time"
"github.com/oklog/ulid/v2"
"yovision/bell/internal/alert"
)
func (p *Postgres) AlertReady(ctx context.Context) error {
var version int64
if err := p.db.QueryRowContext(ctx, `SELECT COALESCE(MAX(version),0) FROM bell.schema_migrations`).Scan(&version); err != nil || version < 6 {
return errors.New("postgres Bell schema migration v6 is required for alerts")
}
for _, table := range []string{"rule_versions", "event_rule_sweeps", "rule_evaluations", "alerts", "alert_events", "alert_transitions", "alert_command_receipts"} {
var selectAllowed, insertAllowed, updateAllowed, deleteAllowed, truncateAllowed bool
if err := p.db.QueryRowContext(ctx, `SELECT
has_table_privilege(current_user,$1,'SELECT'), has_table_privilege(current_user,$1,'INSERT'),
has_table_privilege(current_user,$1,'UPDATE'), has_table_privilege(current_user,$1,'DELETE'),
has_table_privilege(current_user,$1,'TRUNCATE')`, "bell."+table).Scan(
&selectAllowed, &insertAllowed, &updateAllowed, &deleteAllowed, &truncateAllowed,
); err != nil || !selectAllowed || !insertAllowed || updateAllowed || deleteAllowed || truncateAllowed {
return fmt.Errorf("Bell runtime alert privileges violate append-only boundary for %s", table)
}
}
return nil
}
func (p *Postgres) PublishRule(ctx context.Context, rule alert.RuleSpec) (bool, error) {
if err := rule.Validate(); err != nil {
return false, err
}
digest, err := rule.Digest()
if err != nil {
return false, err
}
tx, err := p.db.BeginTx(ctx, nil)
if err != nil {
return false, errors.New("begin Bell rule publish")
}
defer tx.Rollback()
if _, err := tx.ExecContext(ctx, `SELECT pg_advisory_xact_lock(hashtextextended($1,0))`, fmt.Sprintf("rule:%d:%s", rule.TenantID, rule.RuleKey)); err != nil {
return false, errors.New("lock Bell rule key")
}
var existing string
err = tx.QueryRowContext(ctx, `SELECT id FROM bell.rule_versions WHERE tenant_id=$1 AND rule_key=$2 AND config_hash=$3`, rule.TenantID, rule.RuleKey, digest[:]).Scan(&existing)
if err == nil {
return false, tx.Commit()
}
if !errors.Is(err, sql.ErrNoRows) {
return false, errors.New("read Bell rule version")
}
var version int
if err := tx.QueryRowContext(ctx, `SELECT COALESCE(MAX(version),0)+1 FROM bell.rule_versions WHERE tenant_id=$1 AND rule_key=$2`, rule.TenantID, rule.RuleKey).Scan(&version); err != nil {
return false, errors.New("allocate Bell rule version")
}
if _, err := tx.ExecContext(ctx, `INSERT INTO bell.rule_versions(
id,tenant_id,site_id,rule_key,version,display_name,event_kind,minimum_severity,enabled,effective_from,config_hash
) VALUES($1,$2,$3,$4,$5,$6,$7,$8,$9,$10,$11)`, newPrefixedID("ruv_"), rule.TenantID, rule.SiteID,
rule.RuleKey, version, rule.DisplayName, rule.EventKind, rule.MinimumSeverity, rule.Enabled, rule.EffectiveFrom, digest[:]); err != nil {
return false, fmt.Errorf("insert Bell rule version: %w", err)
}
if err := tx.Commit(); err != nil {
return false, errors.New("commit Bell rule version")
}
return true, nil
}
func (p *Postgres) EvaluateNext(ctx context.Context) (bool, error) {
tx, err := p.db.BeginTx(ctx, nil)
if err != nil {
return false, errors.New("begin Bell rule evaluation")
}
defer tx.Rollback()
var eventID, kind, severity string
var tenantID, siteID, deviceID int64
var occurredAt time.Time
err = tx.QueryRowContext(ctx, `SELECT e.id,e.tenant_id,e.site_id,e.device_id,e.kind,e.severity,e.occurred_at
FROM bell.events e WHERE NOT EXISTS(SELECT 1 FROM bell.event_rule_sweeps s WHERE s.event_id=e.id)
ORDER BY e.created_at,e.id LIMIT 1`).Scan(&eventID, &tenantID, &siteID, &deviceID, &kind, &severity, &occurredAt)
if errors.Is(err, sql.ErrNoRows) {
return false, nil
}
if err != nil {
return false, errors.New("select Bell event for rules")
}
if _, err := tx.ExecContext(ctx, `SELECT pg_advisory_xact_lock(hashtextextended($1,0))`, "event-rule:"+eventID); err != nil {
return false, errors.New("lock Bell event evaluation")
}
result, err := tx.ExecContext(ctx, `INSERT INTO bell.event_rule_sweeps(event_id) VALUES($1) ON CONFLICT DO NOTHING`, eventID)
if err != nil {
return false, errors.New("claim Bell event evaluation")
}
rowsAffected, _ := result.RowsAffected()
if rowsAffected == 0 {
return false, tx.Commit()
}
rows, err := tx.QueryContext(ctx, `SELECT id,rule_key,version,display_name,event_kind,minimum_severity,enabled
FROM (SELECT DISTINCT ON (rule_key) id,rule_key,version,display_name,event_kind,minimum_severity,enabled,effective_from
FROM bell.rule_versions WHERE tenant_id=$1 AND effective_from <= $3 AND (site_id IS NULL OR site_id=$2)
ORDER BY rule_key,version DESC,effective_from DESC) latest ORDER BY rule_key`, tenantID, siteID, occurredAt)
if err != nil {
return false, errors.New("read effective Bell rules")
}
defer rows.Close()
type effectiveRule struct {
id, key, name, kind, minimum string
version int
enabled bool
}
rules := make([]effectiveRule, 0)
for rows.Next() {
var rule effectiveRule
if err := rows.Scan(&rule.id, &rule.key, &rule.version, &rule.name, &rule.kind, &rule.minimum, &rule.enabled); err != nil {
return false, errors.New("scan effective Bell rule")
}
rules = append(rules, rule)
}
if err := rows.Err(); err != nil {
return false, errors.New("iterate effective Bell rules")
}
if err := rows.Close(); err != nil {
return false, errors.New("close effective Bell rules")
}
eventRank, _ := alert.SeverityRank(severity)
for _, rule := range rules {
matched, reason := true, "matched"
minimumRank, _ := alert.SeverityRank(rule.minimum)
switch {
case !rule.enabled:
matched, reason = false, "disabled"
case rule.kind != kind:
matched, reason = false, "event_kind"
case eventRank < minimumRank:
matched, reason = false, "severity"
}
value := "no_match"
if matched {
value = "matched"
}
evaluationID := newPrefixedID("eva_")
if _, err := tx.ExecContext(ctx, `INSERT INTO bell.rule_evaluations(id,event_id,rule_version_id,result,reason) VALUES($1,$2,$3,$4,$5)`, evaluationID, eventID, rule.id, value, reason); err != nil {
return false, fmt.Errorf("append Bell rule evaluation: %w", err)
}
if !matched {
continue
}
alertID := newPrefixedID("alt_")
if _, err := tx.ExecContext(ctx, `INSERT INTO bell.alerts(id,tenant_id,site_id,rule_evaluation_id,rule_version_id,severity,title) VALUES($1,$2,$3,$4,$5,$6,$7)`, alertID, tenantID, siteID, evaluationID, rule.id, severity, rule.name); err != nil {
return false, errors.New("create Bell alert")
}
if _, err := tx.ExecContext(ctx, `INSERT INTO bell.alert_events(alert_id,event_id) VALUES($1,$2)`, alertID, eventID); err != nil {
return false, errors.New("link Bell alert event")
}
if _, err := tx.ExecContext(ctx, `INSERT INTO bell.alert_transitions(id,alert_id,sequence,from_state,to_state,actor_ref) VALUES($1,$2,1,NULL,'open','system:rule-worker')`, newPrefixedID("trn_"), alertID); err != nil {
return false, errors.New("open Bell alert")
}
_ = deviceID
}
if err := tx.Commit(); err != nil {
return false, errors.New("commit Bell rule evaluation")
}
return true, nil
}
func (p *Postgres) ListAlerts(ctx context.Context, tenantID, siteID int64, state string, limit int, cursor string) (alert.Page, error) {
if limit < 1 || limit > alert.MaxPageSize {
return alert.Page{}, errors.New("invalid alert page size")
}
if state != "" && state != "open" && state != "acknowledged" && state != "closed" {
return alert.Page{}, errors.New("invalid alert state")
}
rows, err := p.db.QueryContext(ctx, `SELECT a.id,a.severity,a.title,t.to_state,r.rule_key,r.version,a.created_at
FROM bell.alerts a JOIN bell.rule_versions r ON r.id=a.rule_version_id
JOIN LATERAL(SELECT to_state FROM bell.alert_transitions WHERE alert_id=a.id ORDER BY sequence DESC LIMIT 1)t ON true
WHERE a.tenant_id=$1 AND a.site_id=$2 AND ($3='' OR t.to_state=$3)
AND ($4='' OR (a.created_at,a.id) < (SELECT c.created_at,c.id FROM bell.alerts c WHERE c.id=$4 AND c.tenant_id=$1 AND c.site_id=$2))
ORDER BY a.created_at DESC,a.id DESC LIMIT $5`, tenantID, siteID, state, cursor, limit+1)
if err != nil {
return alert.Page{}, errors.New("list Bell alerts")
}
defer rows.Close()
page := alert.Page{Items: make([]alert.Summary, 0, limit)}
for rows.Next() {
var item alert.Summary
if err := rows.Scan(&item.ID, &item.Severity, &item.Title, &item.State, &item.RuleKey, &item.RuleVersion, &item.CreatedAt); err != nil {
return alert.Page{}, errors.New("scan Bell alert list")
}
page.Items = append(page.Items, item)
}
if err := rows.Err(); err != nil {
return alert.Page{}, errors.New("iterate Bell alert list")
}
if len(page.Items) > limit {
cursor := page.Items[limit-1].ID
page.NextCursor = &cursor
page.Items = page.Items[:limit]
}
return page, nil
}
func (p *Postgres) GetAlert(ctx context.Context, tenantID, siteID int64, alertID string) (alert.Detail, error) {
var detail alert.Detail
err := p.db.QueryRowContext(ctx, `SELECT a.id,a.severity,a.title,t.to_state,r.rule_key,r.version,a.created_at
FROM bell.alerts a JOIN bell.rule_versions r ON r.id=a.rule_version_id
JOIN LATERAL(SELECT to_state FROM bell.alert_transitions WHERE alert_id=a.id ORDER BY sequence DESC LIMIT 1)t ON true
WHERE a.tenant_id=$1 AND a.site_id=$2 AND a.id=$3`, tenantID, siteID, alertID).Scan(
&detail.ID, &detail.Severity, &detail.Title, &detail.State, &detail.RuleKey, &detail.RuleVersion, &detail.CreatedAt)
if errors.Is(err, sql.ErrNoRows) {
return alert.Detail{}, alert.ErrNotFound
}
if err != nil {
return alert.Detail{}, errors.New("read Bell alert")
}
detail.EvidenceStatus = "not_enabled"
detail.DeliveryStatus = "not_enabled"
detail.Events = []alert.EventRef{}
detail.Transitions = []alert.Transition{}
rows, err := p.db.QueryContext(ctx, `SELECT e.id,e.device_id,e.kind,e.severity,e.occurred_at FROM bell.alert_events ae JOIN bell.events e ON e.id=ae.event_id WHERE ae.alert_id=$1 ORDER BY e.occurred_at,e.id`, alertID)
if err != nil {
return alert.Detail{}, errors.New("read Bell alert events")
}
for rows.Next() {
var value alert.EventRef
if err := rows.Scan(&value.ID, &value.DeviceID, &value.Kind, &value.Severity, &value.OccurredAt); err != nil {
rows.Close()
return alert.Detail{}, errors.New("scan Bell alert event")
}
detail.Events = append(detail.Events, value)
}
if err := rows.Close(); err != nil {
return alert.Detail{}, errors.New("close Bell alert events")
}
rows, err = p.db.QueryContext(ctx, `SELECT sequence,from_state,to_state,actor_ref,note,occurred_at FROM bell.alert_transitions WHERE alert_id=$1 ORDER BY sequence`, alertID)
if err != nil {
return alert.Detail{}, errors.New("read Bell alert transitions")
}
defer rows.Close()
for rows.Next() {
var value alert.Transition
if err := rows.Scan(&value.Sequence, &value.FromState, &value.ToState, &value.ActorRef, &value.Note, &value.OccurredAt); err != nil {
return alert.Detail{}, errors.New("scan Bell alert transition")
}
detail.Transitions = append(detail.Transitions, value)
}
return detail, rows.Err()
}
func (p *Postgres) Command(ctx context.Context, tenantID, siteID int64, alertID, command, key, actorRef string, note *string) (int, alert.CommandResponse, error) {
if (command != "ack" && command != "close") || tenantID < 1 || siteID < 1 || !alert.ValidIdempotencyKey(key) || !alert.ValidActorRef(actorRef) {
return 0, alert.CommandResponse{}, errors.New("invalid Bell alert command")
}
if note != nil && len([]rune(*note)) > 500 {
return 0, alert.CommandResponse{}, errors.New("Bell alert command note is too long")
}
digest := sha256.Sum256([]byte(strings.Join([]string{alertID, command, actorRef, valueOrEmpty(note)}, "\x00")))
tx, err := p.db.BeginTx(ctx, nil)
if err != nil {
return 0, alert.CommandResponse{}, errors.New("begin Bell alert command")
}
defer tx.Rollback()
if _, err := tx.ExecContext(ctx, `SELECT pg_advisory_xact_lock(hashtextextended($1,0))`, fmt.Sprintf("alert-key:%d:%s", tenantID, key)); err != nil {
return 0, alert.CommandResponse{}, errors.New("lock Bell alert command key")
}
var storedHash, storedBody []byte
var storedStatus int
err = tx.QueryRowContext(ctx, `SELECT command_hash,response_status,response_body::text FROM bell.alert_command_receipts WHERE tenant_id=$1 AND idempotency_key=$2`, tenantID, key).Scan(&storedHash, &storedStatus, &storedBody)
if err == nil {
if !bytes.Equal(storedHash, digest[:]) {
return 0, alert.CommandResponse{}, alert.ErrIdempotencyConflict
}
var response alert.CommandResponse
if err := json.Unmarshal(storedBody, &response); err != nil {
return 0, alert.CommandResponse{}, errors.New("decode Bell alert command receipt")
}
return storedStatus, response, tx.Commit()
}
if !errors.Is(err, sql.ErrNoRows) {
return 0, alert.CommandResponse{}, errors.New("read Bell alert command receipt")
}
if _, err := tx.ExecContext(ctx, `SELECT pg_advisory_xact_lock(hashtextextended($1,0))`, "alert:"+alertID); err != nil {
return 0, alert.CommandResponse{}, errors.New("lock Bell alert")
}
var current, currentActor string
var currentTime time.Time
var sequence int
err = tx.QueryRowContext(ctx, `SELECT t.to_state,t.actor_ref,t.occurred_at,t.sequence FROM bell.alerts a
JOIN LATERAL(SELECT to_state,actor_ref,occurred_at,sequence FROM bell.alert_transitions WHERE alert_id=a.id ORDER BY sequence DESC LIMIT 1)t ON true
WHERE a.id=$1 AND a.tenant_id=$2 AND a.site_id=$3`, alertID, tenantID, siteID).Scan(&current, &currentActor, &currentTime, &sequence)
if errors.Is(err, sql.ErrNoRows) {
return 0, alert.CommandResponse{}, alert.ErrNotFound
}
if err != nil {
return 0, alert.CommandResponse{}, errors.New("read Bell alert state")
}
status := 200
response := alert.CommandResponse{AlertID: alertID, State: current, ActorRef: currentActor, OccurredAt: currentTime}
allowed := (command == "ack" && current == "open") || (command == "close" && current == "acknowledged")
if !allowed {
status = 409
switch {
case command == "ack" && (current == "acknowledged" || current == "closed"):
response.Code = "already_acknowledged"
_ = tx.QueryRowContext(ctx, `SELECT actor_ref,occurred_at FROM bell.alert_transitions WHERE alert_id=$1 AND to_state='acknowledged' ORDER BY sequence LIMIT 1`, alertID).Scan(&response.ActorRef, &response.OccurredAt)
case command == "close" && current == "open":
response.Code = "acknowledgement_required"
case command == "close" && current == "closed":
response.Code = "already_closed"
default:
response.Code = "invalid_state"
}
} else {
next := "acknowledged"
if command == "close" {
next = "closed"
}
response.State = next
response.ActorRef = ""
response.OccurredAt = time.Time{}
if err := tx.QueryRowContext(ctx, `INSERT INTO bell.alert_transitions(id,alert_id,sequence,from_state,to_state,actor_ref,note)
VALUES($1,$2,$3,$4,$5,$6,$7) RETURNING actor_ref,occurred_at`,
newPrefixedID("trn_"), alertID, sequence+1, current, next, actorRef, note).Scan(&response.ActorRef, &response.OccurredAt); err != nil {
return 0, alert.CommandResponse{}, errors.New("append Bell alert transition")
}
}
body, err := json.Marshal(response)
if err != nil {
return 0, alert.CommandResponse{}, errors.New("encode Bell alert command response")
}
if _, err := tx.ExecContext(ctx, `INSERT INTO bell.alert_command_receipts(tenant_id,idempotency_key,command_hash,response_status,response_body) VALUES($1,$2,$3,$4,$5::jsonb)`, tenantID, key, digest[:], status, body); err != nil {
return 0, alert.CommandResponse{}, errors.New("append Bell alert command receipt")
}
if err := tx.Commit(); err != nil {
return 0, alert.CommandResponse{}, errors.New("commit Bell alert command")
}
return status, response, nil
}
func newPrefixedID(prefix string) string { return prefix + ulid.Make().String() }
func valueOrEmpty(value *string) string {
if value == nil {
return ""
}
return *value
}