Watch an n8n recovery drill: restore, compare and catch a failure

0:00 · From backup to recovery evidence.

A backup is the start. A recovery drill is the next step. This demonstration combines a real isolated n8n restore with the original offline checker in our Recovery Drill Kit. The voiceover is synthetic. The commands, exports, and recorded lab results are real.

0:20 · Start with a bounded test.

The lab used n8n version two point thirty-nine point nine, on Node twenty-four point twenty-one. It contained one two-node workflow and one encrypted dummy credential, with a local HTTP endpoint. The source and restored instances were separate. This was a single-instance SQLite test, with synthetic data and no production system.

0:46 · Inspect the real baseline export.

First, we run the shipped checker against the actual source export. It checks workflow structure and reports the credential reference. It does not decrypt a secret. This n8n export also contains newer metadata that the checker flags for manual review. Exit zero does not erase those warnings. The report still says recovery verified is false.

1:13 · Restore first. Then prove execution.

The lab stopped the source, copied its data folder, and restored that copy into a separate instance. The snapshot and restored-copy hashes matched before startup. The encryption-key configuration was included. Then the real restored workflow executed successfully and authenticated to the local endpoint with the copied dummy credential. That execution is separate evidence. The kit itself did not perform the restore.

1:43 · Compare the restored workflow export.

Now we compare the actual before and after exports. The result is a semantic match, with no missing workflows and no changes in supported content. This is useful evidence that the workflow survived this copy. It is not a certificate for the whole system. The checker deliberately keeps recovery verified false and requires manual restore review.

2:09 · Make a controlled difference visible.

For a negative case, we make a controlled local edit to the real restored export. One HTTP timeout changes from five thousand to four thousand nine hundred ninety-nine. This is not another live n8n export. The comparison now reports a semantic difference in the nodes section and returns exit one. It identifies the changed area without printing the parameter values.

2:37 · The most important boundary: the key.

The strongest boundary test uses another database copy without the original encryption-key configuration. n8n can still export its workflow, and our comparison still reports a match. But real execution fails with a credential decryption error. The local endpoint receives no new request. Matching workflow JSON therefore cannot prove that credentials are usable. That is why the original key, the database, and a controlled execution test belong in the recovery plan.

3:12 · Use the evidence. Repeat your own drill.

This experiment demonstrates one successful same-version SQLite restore and two deliberate negative checks. It does not cover Docker, PostgreSQL, queue mode, external storage, custom nodes, or a version upgrade. Read the full lab report, then repeat a controlled drill for your own topology. Use the offline checker and workbook to collect evidence, record the gaps, and decide which manual checks remain.