An n8n polling trigger reads a source repeatedly and compares what it sees with an earlier snapshot. In AppStruct's New Row trigger, the first automatic poll establishes that snapshot. A later poll emits IDs it has not seen in the preceding successful snapshot. Updated Row instead emits existing IDs whose returned values changed.
That distinction matters when a service-request table already contains records. A manual preview can show those records while an automatic first poll emits nothing. Neither behavior, by itself, proves that a destination will receive every change exactly once.
Use this exercise to inspect the trigger before attaching a notification, customer message or record-writing step. Download the inactive New Row practice workflow and polling acceptance checklist. They contain no account credentials, private project IDs or customer data.
Install a compatible package and check the account
This walkthrough describes n8n-nodes-appstruct 2.0.1, published on November 15, 2026. It requires self-hosted n8n 2.47.7 or newer and Node.js 24+, with community package installation allowed by the administrator. Review the source and compatibility notes before installing third-party code in your instance.
The package passed the automated n8n registry scanner and has GitHub Actions provenance. Its Creator Portal submission is in progress; this is not a claim of n8n verification, an official partnership or n8n Cloud availability.
Follow the AppStruct n8n setup guide to save your AppStruct account email and password in n8n credentials and perform Project → Get Many first. Compare a returned project with your Apps library. A Docs, Sheets or Slides MCP key is a different connection and cannot replace this credential.
The practice workflow reads a table, but its account credential is not restricted to reads. Use a practice project you are allowed to inspect. Keep executions private because returned row values can appear in n8n execution history.
Import the inactive review workflow
Use n8n's Import from File option and select the downloaded JSON. It contains an AppStruct Trigger and an instruction note, with no downstream action. It is saved with active: false.
Open the trigger and select:
| Setting | Practice value |
|---|---|
| Credential | Your saved AppStruct API credential |
| Trigger On | New Row |
| Project Name or ID | A practice App Builder project selected from your account |
| Table Name or ID | A small practice request table |
| Record Poll Limit | 1000, with fewer than 1000 rows in this exercise |
| Poll schedule | Every minute, supplied in the example; change it to suit your test |
The project and table fields are intentionally empty in the file. Select real values yourself; the labels used in this article do not identify a live project. Do not enable the workflow until those fields and the credential are correct.
Use fictional labels such as POLL-001 and POLL-002 in your table. Keep the actual row IDs returned by AppStruct alongside those labels when checking output. Labels and stable IDs have different jobs: the label makes a fixture recognizable, while the ID distinguishes the stored record.
Separate manual preview from the automatic baseline
Run a manual preview with two existing practice rows. In version 2.0.1, manual mode returns current items and does not overwrite the live polling snapshot. Seeing two items here does not mean two newly created requests were detected.
After checking the output, save and activate or publish the configured workflow using the controls in your n8n version. Wait for an automatic scheduled poll before adding another row. The first successful automatic poll establishes a baseline and emits no changes.
Then create POLL-003 in your practice table, wait for the next scheduled poll and inspect the result. The expected result is the new row's actual ID. A subsequent poll without source changes should emit nothing. Record the observation rather than marking the checklist passed in advance.
| Check | Expected observation in this exercise |
|---|---|
| Manual preview | Current rows may appear; this is a preview |
| First successful automatic poll | Baseline saved; no change output |
| Add POLL-003 after that baseline | Its new ID appears at a later poll |
| Poll again with no source changes | No repeated output for that unchanged ID |
| Change the selected project or table | A separate baseline is established for the new scope |
An execution with no output can therefore be expected. Inspect n8n's available execution evidence and the source record before treating silence as an authentication failure. Do not re-create a workflow just to force output: a new workflow or node can start with new polling state.
Test Updated Row independently
Use a separate practice workflow or deliberately change Trigger On to Updated Row, then establish its own automatic baseline. Changing the trigger type changes the monitored scope.
Update only POLL-001's status and wait for the next poll. Its existing ID should appear with the returned values. Repeat the poll without another change and expect no output. Adding a completely new row is not an update to an existing ID, so it is not emitted by Updated Row during that comparison.
The node compares the fields returned by its table read, rather than using a durable source event log. If a field changes and changes back between polls, the next snapshot can look unchanged. A record created and deleted between polls may also go unseen. Use another delivery design if your process must retain every intermediate event.
Handle limits and failed reads explicitly
The default Record Poll Limit is 1000; supported values are integers from 2 to 10000. The node fails when the number of returned rows reaches the configured limit, including an exactly full result. It cannot establish that this result covers the whole table, so it does not silently advance an incomplete snapshot.
For a small test table, you can deliberately choose a limit equal to its row count to inspect that failure, then restore a suitable limit. Increase the limit only within supported bounds or monitor a smaller table. Do not assume the trigger paginates an arbitrarily large table.
A failed source read does not replace the last successful baseline. A successful later read compares against that retained snapshot. That still does not recover all intermediate changes that happened during the failure.
The saved trigger state contains row IDs and SHA-256 fingerprints instead of copies of row contents. Execution output and the source database have their own storage and access rules; fingerprinted polling state does not make those other copies anonymous.
Design the destination before enabling writes
Only attach a destination after the read exercise behaves as expected. A snapshot comparison suppresses repeated unchanged source rows; it does not provide an end-to-end exactly-once delivery guarantee. The node advances its snapshot after a successful poll, before it knows whether every later node will finish successfully.
For a new-request destination, use a stable key such as project + table + row ID and a unique constraint or idempotent upsert appropriate to that destination. For repeated updates to the same row, include a deliberate version or payload-change policy; deduplicating on row ID alone could discard legitimate later updates.
Test these destination outcomes separately:
- The same source ID is processed twice without creating an unwanted duplicate.
- A failed destination attempt is visible and has a defined retry or replay procedure.
- A retry after a timeout can distinguish an already completed write from one that never completed.
- Changing the monitored scope does not silently route another table's records into the old destination.
The downloadable workflow intentionally has no notification or external write. It cannot send a message or demonstrate a production retry policy. The checklist includes these later checks as Not checked, ready for your implementation and observations.
Decide whether the workflow is ready for a pilot
Keep the first pilot small enough to reconcile: expected source IDs, observed trigger outputs and destination outcomes. Record the package and n8n versions, monitored scope, test time and evidence location in the checklist. Repeat affected checks after an upgrade or a change to the source table.
For the source interface, follow the service-request app exercise. For the connection, return to the n8n setup and manual first read. Start with an inspectable read, then add only the destination behavior you can verify.
The trigger behavior described here is based on the 2.0.1 implementation and its polling regression tests. The CSV's expected results are an acceptance plan, not a report of a customer's deployment.
Compare your result in the n8n Community discussion
The AppStruct polling exercise discussion on n8n Community is open for questions and test observations. It was posted by AppStruct's maker; it is not an independent review or an n8n verification.
If a check differs from the expected result, include your n8n version, package version, trigger type, checklist step and a redacted error. State whether the run was a manual preview or an automatic poll and whether its baseline had already been established. Share the observed behavior, including an empty result when that is what happened.
Do not post account credentials, access tokens, private table contents or an unreviewed workflow export. Use the fictional practice records from this exercise, and keep your evidence location in your own checklist. The discussion complements the setup guide; it does not establish that an untested workflow is ready for production.