ITWorx-Pulse-Public/internal/backup/manager_integration_test.go

package backup

import (
	"context"
	"os"
	"sort"
	"strings"
	"testing"
	"time"

	"github.com/itworx/pulse/internal/database"
	"github.com/jackc/pgx/v5/pgxpool"
)

func TestPostgreSQLBackupRestoreCleanRoom(t *testing.T) {
	sourceDSN := os.Getenv("PULSE_TEST_DATABASE_URL")
	targetDSN := os.Getenv("PULSE_TEST_RESTORE_DATABASE_URL")
	if sourceDSN == "" || targetDSN == "" {
		if os.Getenv("PULSE_REQUIRE_BACKUP_INTEGRATION") == "true" {
			t.Fatal("backup integration is required but both PostgreSQL DSNs are not configured")
		}
		t.Skip("PULSE_TEST_DATABASE_URL and PULSE_TEST_RESTORE_DATABASE_URL are required")
	}
	ctx, cancel := context.WithTimeout(context.Background(), 90*time.Second)
	defer cancel()
	source, err := database.NewPool(ctx, database.Config{URL: sourceDSN})
	if err != nil {
		t.Fatal(err)
	}
	defer source.Close()
	target, err := database.NewPool(ctx, database.Config{URL: targetDSN})
	if err != nil {
		t.Fatal(err)
	}
	defer target.Close()
	if err := database.Migrate(ctx, source); err != nil {
		t.Fatal(err)
	}
	if err := database.Migrate(ctx, target); err != nil {
		t.Fatal(err)
	}
	assertBackupSchemaCoverage(t, ctx, source)
	ids := testIDs()
	now := time.Now().UTC().Truncate(time.Microsecond)
	seed := []struct {
		query string
		args  []any
	}{
		{`INSERT INTO roles (id,name) VALUES ($1,'administrator')`, []any{ids.role}},
		{`INSERT INTO users (id,external_subject,display_name,email) VALUES ($1,$2,'Backup Test','backup@example.invalid')`, []any{ids.user, ids.user}},
		{`INSERT INTO user_roles (user_id,role_id) VALUES ($1,$2)`, []any{ids.user, ids.role}},
		{`INSERT INTO data_sources (id,type,name,configuration_ref) VALUES ($1,'exporter','Backup source','source/ref')`, []any{ids.source}},
		{`INSERT INTO entities (id,entity_type,canonical_name,display_name,first_seen_at) VALUES ($1,'container','backup-test','Backup test',$2)`, []any{ids.entity, now}},
		{`INSERT INTO container_aliases (source_id,runtime_id,entity_id,name,project,service,image_digest,observed_state,observed_health,restart_count,first_seen_at,last_seen_at) VALUES ($1,'runtime-backup-test',$2,'backup-test','pulse','api','sha256:test','running','healthy',2,$3,$3)`, []any{ids.source, ids.entity, now}},
		{`INSERT INTO dashboards (id,slug,name,description,owner_user_id,scope,current_version_id) VALUES ($1,'backup-test','Backup test dashboard','portable',$2,'shared',NULL)`, []any{ids.dashboard, ids.user}},
		{`INSERT INTO dashboard_versions (id,dashboard_id,version_number,schema_version,document,created_by) VALUES ($1,$2,1,1,'{}',$3)`, []any{ids.dashboardVersion, ids.dashboard, ids.user}},
		{`UPDATE dashboards SET current_version_id=$1 WHERE id=$2`, []any{ids.dashboardVersion, ids.dashboard}},
		{`INSERT INTO alert_rules (id,schema_version,name,severity,condition,evaluation_interval_seconds,pending_seconds,resolve_seconds,unknown_behavior,group_by,suppress_when,message,current_version_id,created_by) VALUES ($1,1,'Backup test rule','critical','{}',60,0,0,'become-unknown','[]','[]','{}',NULL,$2)`, []any{ids.rule, ids.user}},
		{`INSERT INTO alert_rule_versions (id,rule_id,version_number,document,created_by) VALUES ($1,$2,1,'{}',$3)`, []any{ids.ruleVersion, ids.rule, ids.user}},
		{`UPDATE alert_rules SET current_version_id=$1 WHERE id=$2`, []any{ids.ruleVersion, ids.rule}},
		{`INSERT INTO alert_instances (id,rule_id,rule_version_id,fingerprint,entity_id,last_evaluated_at) VALUES ($1,$2,$3,'backup-fingerprint',$4,$5)`, []any{ids.instance, ids.rule, ids.ruleVersion, ids.entity, now}},
		{`INSERT INTO incidents (id,correlation_key,title,severity,started_at,owner_user_id,correlation_method,confidence) VALUES ($1,'backup-correlation','Backup test incident','critical',$2,$3,'deterministic',0.900)`, []any{ids.incident, now, ids.user}},
		{`INSERT INTO incident_alerts (incident_id,alert_id,rationale,confidence,correlation_method,added_by) VALUES ($1,$2,'backup test rationale',0.900,'deterministic','test')`, []any{ids.incident, ids.instance}},
		{`INSERT INTO incident_entities (incident_id,entity_id,rationale,confidence) VALUES ($1,$2,'backup test entity',0.900)`, []any{ids.incident, ids.entity}},
		{`INSERT INTO incident_notes (id,incident_id,author,body) VALUES ($1,$2,'test','backup note')`, []any{ids.note, ids.incident}},
		{`INSERT INTO audit_events (id,actor,action,resource_type,resource_id,result,after_diff) VALUES ($1,'backup-test','backup.seed','dashboard',$2,'success','{}')`, []any{ids.audit, ids.dashboard}},
	}
	for _, statement := range seed {
		if _, err := source.Exec(ctx, statement.query, statement.args...); err != nil {
			t.Fatal(err)
		}
	}
	directory := t.TempDir()
	manager := Manager{Pool: source, Directory: directory, Retention: 2, Now: func() time.Time { return now }}
	created, err := manager.Create(ctx)
	if err != nil {
		t.Fatal(err)
	}
	if created.Rows <= 0 || created.SHA256 == "" {
		t.Fatalf("unexpected backup result: %#v", created)
	}
	verified, err := manager.Verify(ctx, created.Path)
	if err != nil {
		t.Fatal(err)
	}
	if verified.FormatVersion != formatVersion {
		t.Fatalf("backup format = %d, want %d", verified.FormatVersion, formatVersion)
	}
	containerAliasManifest := false
	for _, table := range verified.Tables {
		if table.Name == "container_aliases" && table.Rows == 1 && table.SHA256 != "" {
			containerAliasManifest = true
		}
	}
	if !containerAliasManifest {
		t.Fatal("container_aliases is missing from the checksummed manifest")
	}
	if _, err := manager.List(ctx); err != nil {
		t.Fatal(err)
	}
	restored, err := (Manager{Pool: target}).Restore(ctx, created.Path)
	if err != nil {
		t.Fatal(err)
	}
	if restored.BackupID != created.BackupID {
		t.Fatalf("restored manifest = %s, want %s", restored.BackupID, created.BackupID)
	}
	for _, check := range []struct {
		table string
		want  int
	}{
		{"container_aliases", 1}, {"dashboards", 1}, {"dashboard_versions", 1}, {"alert_rules", 1}, {"alert_rule_versions", 1}, {"incidents", 1}, {"incident_alerts", 1}, {"incident_entities", 1}, {"incident_notes", 1}, {"audit_events", 1},
	} {
		var got int
		if err := target.QueryRow(ctx, "SELECT count(*) FROM "+check.table).Scan(&got); err != nil {
			t.Fatal(err)
		}
		if got != check.want {
			t.Fatalf("%s count = %d, want %d", check.table, got, check.want)
		}
	}
	var dashboardVersion, ruleVersion string
	if err := target.QueryRow(ctx, `SELECT current_version_id::text FROM dashboards WHERE id=$1`, ids.dashboard).Scan(&dashboardVersion); err != nil {
		t.Fatal(err)
	}
	if err := target.QueryRow(ctx, `SELECT current_version_id::text FROM alert_rules WHERE id=$1`, ids.rule).Scan(&ruleVersion); err != nil {
		t.Fatal(err)
	}
	if dashboardVersion != ids.dashboardVersion || ruleVersion != ids.ruleVersion {
		t.Fatalf("deferred links = %s/%s", dashboardVersion, ruleVersion)
	}
	var runtimeEntity string
	if err := target.QueryRow(ctx, `SELECT entity_id::text FROM container_aliases WHERE source_id=$1 AND runtime_id='runtime-backup-test'`, ids.source).Scan(&runtimeEntity); err != nil {
		t.Fatal(err)
	}
	if runtimeEntity != ids.entity {
		t.Fatalf("restored container alias entity = %s, want %s", runtimeEntity, ids.entity)
	}
}

func assertBackupSchemaCoverage(t *testing.T, ctx context.Context, pool *pgxpool.Pool) {
	t.Helper()
	rows, err := pool.Query(ctx, `SELECT c.table_name, c.column_name
		FROM information_schema.columns c
		JOIN information_schema.tables t
		  ON t.table_schema = c.table_schema AND t.table_name = c.table_name
		WHERE c.table_schema = 'public' AND t.table_type = 'BASE TABLE'
		ORDER BY c.table_name, c.ordinal_position`)
	if err != nil {
		t.Fatal(err)
	}
	defer rows.Close()
	columnsByTable := map[string]map[string]bool{}
	for rows.Next() {
		var table, column string
		if err := rows.Scan(&table, &column); err != nil {
			t.Fatal(err)
		}
		if columnsByTable[table] == nil {
			columnsByTable[table] = map[string]bool{}
		}
		columnsByTable[table][column] = true
	}
	if err := rows.Err(); err != nil {
		t.Fatal(err)
	}

	classified := map[string]string{"schema_migrations": "migration metadata"}
	for name, reason := range backupExcludedTables {
		classified[name] = reason
	}
	for _, spec := range tableSpecs {
		if previous, duplicate := classified[spec.name]; duplicate {
			t.Fatalf("backup table %q is classified more than once (previous: %s)", spec.name, previous)
		}
		classified[spec.name] = "portable backup"
		tableColumns, exists := columnsByTable[spec.name]
		if !exists {
			t.Fatalf("backup table %q does not exist after migrations", spec.name)
		}
		for _, list := range []string{spec.columns, spec.restoreCols} {
			for _, column := range strings.Split(list, ",") {
				column = strings.TrimSpace(column)
				if column == "" || !tableColumns[column] {
					t.Fatalf("backup table %q references missing column %q", spec.name, column)
				}
			}
		}
	}

	var unclassified, missing []string
	for table := range columnsByTable {
		if _, ok := classified[table]; !ok {
			unclassified = append(unclassified, table)
		}
	}
	for table := range classified {
		if _, ok := columnsByTable[table]; !ok {
			missing = append(missing, table)
		}
	}
	sort.Strings(unclassified)
	sort.Strings(missing)
	if len(unclassified) > 0 || len(missing) > 0 {
		t.Fatalf("backup schema drift: unclassified=%v missing=%v", unclassified, missing)
	}
}

type testIDSet struct {
	role, user, source, entity, dashboard, dashboardVersion, rule, ruleVersion, instance, incident, note, audit string
}

func testIDs() testIDSet {
	id := func() string {
		value, err := newID()
		if err != nil {
			panic(err)
		}
		return value
	}
	return testIDSet{
		role: id(), user: id(), source: id(), entity: id(), dashboard: id(), dashboardVersion: id(),
		rule: id(), ruleVersion: id(), instance: id(), incident: id(), note: id(), audit: id(),
	}
}