fix(ops): run backups as an admin login, and stop recording failed dumps as good

The Operaciones panel (backup, restore, sync, re-import) shelled out to
mysqldump as the application user, parsed straight out of DATABASE_URL.
`--single-transaction` issues FLUSH TABLES, which needs the global RELOAD
privilege, and the app user is granted only ALL ON jorgecuadros.* plus
USAGE ON *.*. BACKUP failed outright; SYNC and REIMPORT failed with it,
since both take a safety backup first.

An admin credential is now supplied out of band via OPS_DB_ADMIN_USER /
OPS_DB_ADMIN_PASSWORD, mirroring what deploy/scripts/pre-migrate-backup.mjs
already does, rather than permanently elevating the user the API serves
requests as. Host, port and database still come from DATABASE_URL, so the
override can only change who logs in, never which server. Unset, it falls
back to the DATABASE_URL credentials and warns — local development is
unaffected.

Two defects in the dumps themselves, both shared with the deploy backup
before it was rewritten:

- No --set-gtid-purged=OFF. The production server is the replication source
  with GTID on, so every dump embedded SET @@GLOBAL.GTID_PURGED and was
  unrestorable onto the server it came from — the one thing the restore
  screen is for.

- The pipeline's exit status was gzip's, and gzip succeeded. A mysqldump
  that died on its first statement left a small, perfectly valid archive
  that the job recorded as SUCCESS and the restore screen listed as an
  ordinary restore point. Dumps now run under `set -o pipefail`, assert a
  CREATE TABLE count, and delete their own output on failure. Verified with
  a stubbed mysqldump: a failing dump exits 1, surfaces the real error,
  removes the partial file, and — critically — stops SYNC/REIMPORT before
  the ETL touches anything.

Restores gained pipefail too: a corrupt archive made gunzip fail while
mysql, fed a truncated stream, could still exit 0.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

apps/api/src/ops/ops.service.ts

@@ -31,6 +31,14 @@ export const INGEST_FILES = [
 ] as const;
 export type IngestName = (typeof INGEST_FILES)[number];
 
+/**
+ * Prefix for every command containing a pipe. Without it the exit status of
+ * `mysqldump | gzip` is gzip's, so a dump that failed immediately still looks
+ * like a successful job. Both Alpine's busybox ash (the API image) and macOS
+ * `sh` (dev) support it; POSIX does not require it, so `sh -c` is the contract.
+ */
+const PIPEFAIL = "set -o pipefail; ";
+
 interface MysqlConn {
   host: string;
   port: string;
@@ -175,7 +183,7 @@ export class OpsService implements OnModuleInit {
       );
     }
 
-    const conn = this.parseDbUrl();
+    const conn = this.opsConn();
     const { cmd, resolvedParams } = await this.buildCommand(kind, params, conn);
 
     const job = await this.prisma.opsJob.create({
@@ -207,6 +215,37 @@ export class OpsService implements OnModuleInit {
     };
   }
 
+  /**
+   * The credentials mysqldump/mysql run as — deliberately NOT the application
+   * user. `--single-transaction` issues FLUSH TABLES, which needs the global
+   * RELOAD privilege, and the app user is granted only `ALL ON jorgecuadros.*`
+   * plus `USAGE ON *.*`; `--skip-lock-tables` does not avoid it. A restore of a
+   * dump taken before --set-gtid-purged=OFF likewise needs SUPER to replay its
+   * SET @@GLOBAL.GTID_PURGED. So an admin credential is supplied out of band
+   * rather than elevating the runtime user for the sake of one admin screen —
+   * the same choice deploy/scripts/pre-migrate-backup.mjs makes.
+   *
+   * Host, port and database always come from DATABASE_URL: the ops user is a
+   * different login on the SAME server, never a way to point at another one.
+   *
+   * With the vars unset this falls back to the DATABASE_URL credentials, which
+   * is what local development wants — a dev MySQL grants the app user far more.
+   */
+  private opsConn(): MysqlConn {
+    const conn = this.parseDbUrl();
+    const user = process.env.OPS_DB_ADMIN_USER;
+    const password = process.env.OPS_DB_ADMIN_PASSWORD;
+    if (!user || !password) {
+      this.logger.warn(
+        "OPS_DB_ADMIN_USER/OPS_DB_ADMIN_PASSWORD no configuradas; " +
+          `usando el usuario de la aplicación (${conn.user}) para mysqldump. ` +
+          "En producción esto falla por falta del privilegio RELOAD.",
+      );
+      return conn;
+    }
+    return { ...conn, user, password };
+  }
+
   /** mysql/mysqldump connection flags. The password goes through MYSQL_PWD in
    *  the child env, never on the command line (which would leak via `ps`). */
   private connFlags(c: MysqlConn): string {
@@ -217,6 +256,37 @@ export class OpsService implements OnModuleInit {
     return new Date().toISOString().replace(/[:.]/g, "-").replace("T", "_").slice(0, 19);
   }
 
+  /**
+   * One hardened mysqldump, shared by BACKUP and by the safety backups SYNC and
+   * REIMPORT take first. Kept byte-for-byte in spirit with the dump in
+   * deploy/scripts/pre-migrate-backup.mjs — the two write into the same volume
+   * and both are listed as restore points by this same screen.
+   *
+   * --set-gtid-purged=OFF: the production server is the replication SOURCE with
+   * GTID on, so without it every dump embeds SET @@GLOBAL.GTID_PURGED and is
+   * unrestorable onto the very server it came from.
+   *
+   * The table-count assertion is not belt-and-braces: `gzip -t` passes on the
+   * ~372-byte output of a mysqldump that died on its first statement, so a
+   * failed dump would otherwise be recorded as a successful backup. (`set -o
+   * pipefail` is set by the caller for the same reason — without it the exit
+   * status of the pipeline is gzip's, and gzip succeeded.)
+   *
+   * A failed attempt deletes its own output, so a truncated file never appears
+   * in the restore list looking like an ordinary restore point.
+   */
+  private dumpCommand(flags: string, db: string, out: string): string {
+    return (
+      `( mysqldump ${flags} --single-transaction --routines --triggers ` +
+      `--no-tablespaces --set-gtid-purged=OFF ${db} | gzip -c > ${out} && ` +
+      `gzip -t ${out} && ` +
+      `TABLAS=$(gunzip -c ${out} | grep -c 'CREATE TABLE') && ` +
+      `echo "tablas capturadas: $TABLAS" && ` +
+      `[ "$TABLAS" -ge 1 ] ) || ` +
+      `{ rm -f ${out}; echo 'respaldo incompleto eliminado'; exit 1; }`
+    );
+  }
+
   private async buildCommand(
     kind: OpsJobKind,
     params: Record<string, unknown>,
@@ -229,7 +299,7 @@ export class OpsService implements OnModuleInit {
       const file = `backup-${this.migrationEnv}-${this.timestamp()}.sql.gz`;
       const out = shq(path.join(this.backupDir, file));
       return {
-        cmd: `mysqldump ${flags} --single-transaction --routines --triggers --no-tablespaces ${db} | gzip -c > ${out}`,
+        cmd: `${PIPEFAIL}${this.dumpCommand(flags, db, out)}`,
         resolvedParams: { file },
       };
     }
@@ -241,7 +311,10 @@ export class OpsService implements OnModuleInit {
         throw new NotFoundException(`Respaldo no encontrado: ${name}`);
       });
       return {
-        cmd: `gunzip -c ${shq(full)} | mysql ${flags} ${db}`,
+        // pipefail matters here too: a corrupt archive makes gunzip fail while
+        // mysql, fed a truncated stream, can still exit 0 — a restore that
+        // reported success having replayed only part of the dump.
+        cmd: `${PIPEFAIL}gunzip -c ${shq(full)} | mysql ${flags} ${db}`,
         resolvedParams: { file: name },
       };
     }
@@ -252,8 +325,8 @@ export class OpsService implements OnModuleInit {
       const py = await this.pythonBin();
       const runAll = shq(path.join(this.migrationDir, "run_all.py"));
       const cmd =
-        `echo '== Respaldo de seguridad previo ==' && ` +
-        `mysqldump ${flags} --single-transaction --routines --triggers --no-tablespaces ${db} | gzip -c > ${out} && ` +
+        `${PIPEFAIL}echo '== Respaldo de seguridad previo ==' && ` +
+        `${this.dumpCommand(flags, db, out)} && ` +
         `echo '== Sincronización aditiva desde carpeta de ingesta ==' && ` +
         `${shq(py)} ${runAll} --env ${shq(this.migrationEnv)} --sync`;
       return { cmd, resolvedParams: { safetyBackup: file } };
@@ -266,8 +339,8 @@ export class OpsService implements OnModuleInit {
       const py = await this.pythonBin();
       const runAll = shq(path.join(this.migrationDir, "run_all.py"));
       const cmd =
-        `echo '== Respaldo de seguridad previo ==' && ` +
-        `mysqldump ${flags} --single-transaction --routines --triggers --no-tablespaces ${db} | gzip -c > ${out} && ` +
+        `${PIPEFAIL}echo '== Respaldo de seguridad previo ==' && ` +
+        `${this.dumpCommand(flags, db, out)} && ` +
         `echo '== Reimportación desde carpeta de ingesta ==' && ` +
         `${shq(py)} ${runAll} --env ${shq(this.migrationEnv)} --stage`;
       return { cmd, resolvedParams: { safetyBackup: file } };
@@ -287,6 +360,15 @@ export class OpsService implements OnModuleInit {
     }
   }
 
+  /**
+   * `password` is the ops credential from opsConn(), exported as MYSQL_PWD so it
+   * never reaches argv (which `ps` exposes to every process on the host).
+   *
+   * It does not leak into the Python ETL that SYNC and REIMPORT go on to run:
+   * migration/dbenv.py connects with pymysql using the credentials inside
+   * DATABASE_URL and never consults MYSQL_PWD. The ETL keeps running as the
+   * application user, which is what it should be doing.
+   */
   private run(jobId: string, cmd: string, password: string): void {
     const child = spawn("sh", ["-c", cmd], {
       cwd: this.migrationDir,