N
Hi — if Instance settings already says “Your instance is applying changes” and Restart is missing or does not complete, this is usually a Cloud-side provision that has not finished, not something you can fix from the editor. While you wait for staff: Keep the Cloud console on Instance settings. If Restart becomes available, use it once and wait. Avoid opening the editor in several tabs or repeatedly restarting while the instance is still applying changes. Try the console once in a…
N
Most of the edge cases are in the parsing, so I normalize everything to “milliseconds to wait” in one Code node right after the HTTP Request (with “Include Response Headers and Status” on and “Never Error” enabled, so 429/503 responses arrive as data instead of failing the node): const out = []; for (const item of $input.all()) { const h = item.json.headers || {}; const attempt = item.json.attempt ?? 0; const raw = h['retry-after']; let waitMs = null; if (raw !== undefined && String(raw).trim()…
N
Update: the full setup is now on video (3:46, with chapters), including the pagination fix at 2:06:Never Run Out of Scheduled Posts Again (Free n8n Workflow) Two changes since the first post: Fixed a bug where line breaks in the alert email showed up as a literal \n. The Blotato key is now a Header Auth credential (name blotato-api-key), so you don’t need the community node installed. Updated workflow JSON below.
N
Thanks, hearing it happened in production with a booking write is really useful, especially “the agent announced success anyway”. Here’s the before/after for the “processes now, replies after 10 seconds” case, measured on a real local n8n 2.46.6: n8n Retry shipped twice, Retry On Fail shipped three times (both still marked error), and checking the destination first shipped once. Repository parkseyoung1215-netizen/agent-recovery GitHub - parkseyoung1215-netizen/agent-recovery Contribute to…
N
To handle atomic claims, I push the concurrency control entirely down to the database layer. In my production pipelines, I use Supabase (PostgreSQL) with a strict UNIQUE constraint on a deterministic idempotency key (usually the webhook_id provided by the sender, or a SHA-256 hash of the payload if no ID is provided). Skip the Lookup: When the webhook arrives, I don’t do a “GET” check. I immediately push the payload into a Supabase Postgres node set to INSERT. The Atomic Lock: Because of the…
N
If there is an IF, Switch, or Error Trigger node inside the loop, Everyone often forget to wire every possible outcome back to the Loop node. If an item goes through ‘False’ branch or an error path that isn’t wired back to the start of the loop, the loop silently breaks and stops iterating right there. To make my pipelines fault-tolerant, I always ensure even the empty/failed branches route back to the loop to keep the iterations moving. My second immediate check is the Nested Array Trap. I…
N
In my production pipelines, I parse the header (whether it’s seconds or a date) and route it through a Switch node to check if the wait time exceeds my inline budget. Under Budget (< 60s) : Routes to a standard n8n Wait node to pause and retry inline. Over Budget (> 60s): Instead of holding the execution open, I route the payload straight into a Supabase with a next retry at timestamp. A separate lightweight scheduled workflow polls Supabase every few minutes and replays the deferred payloads…
N
Hello, Thank you for your feedback and for taking the time to review my project! For the scoring logic, I use Google Gemini with predefined guidelines in the prompt. It evaluates several factors, including buying intent, budget, project clarity, and urgency. Based on these criteria, leads are classified into three categories: Cold Lead: 0–39 Warm Lead: 40–69 SQL: 70–100 I’ve tested the workflow locally with three sample leads to verify the different classification scenarios and the complete…
N
Good point, especially about where the approval happens. The current dryRun skips the Zvid render, but the script and voice calls still run, so reviewing the finished draft is already too late to save voice credits. A checkpoint after the script is parsed, with approve/reject before ElevenLabs, makes sense. I like the Sheets queue doubling as the status log too. Saving the voice URL and render job ID alongside the status would help reruns resume without repeating paid steps. And thanks for…
N
Hey @operantsolo, while you wait for a response, here are some things that might help: Suggested resources (click for more details) @Niffzy, @Emmas, @A_A4 - you’ve helped with similar issues before, can you take a look? Automatically suggested by n8n’s community bot. It’s a pilot - please share feedback here.
N
How do you handle Retry-After when it’s a date instead of seconds, or missing entirely? In a policy lab I used the provider value as a floor, added a little jitter after it, and capped inline waits — deferring anything over the budget instead of looping forever. What does your retry policy do when the wait would exceed your inline budget?
N
Mine is the same three, but I run them in a different order. First thing I check is the item count on the Loop node input itself, because the most common cause of ‘only the first item’ is that there is only one item coming in. When the previous node returns a single object with an array inside it (one item with a results field), Loop Over Items sees 1 and behaves fine, which looks like a bug. A Split Out node before the loop fixes that in one step. Second is where the loop-back wire goes. It…
N
That last line is the one that matters, and it is exactly where I am: the workflow is built, the loop is honest, and I still need one agency to put a real inbox through it. If you are up for it, here is a swap that costs us both nothing: I run one real mailbox through Guardian (one workflow, a couple of weeks, real enquiries) and write down what breaks; you point my triage template at a real inbox and tell me where the approval step actually fails. Between us that is one end-to-end case each,…
N
In New York City, every HPD housing code violation comes with a certify-by date: the date by which the owner has to fix the problem and certify the repair with HPD. The violations are published on NYC Open Data, free and without an API key. The catch is that they are written in legal code language, and they sit in that dataset until someone goes looking, building by building. Here is a real one from a Brooklyn building I tested with: § 27-2026, 2027 HMC: PROPERLY REPAIR THE SOURCE AND ABATE THE…
N
Hey n8n Community, I built a reporting dashboard for a client project this week and ran into something I do not have a good answer for. The client named a handful of things they definitely wanted to see, which was easy. But then came the familiar line: everything else is up to you, just give me a good overview. And that is where I always hesitate. Put too little on there and they come back asking for more. Put too much on and it stops being an overview, it becomes a data dump nobody reads. My…
N
@Portalia_Ro This should help docs.n8n.io Use the Cloud console | Deploy | n8n Docs How to open and find your way around the n8n Cloud console, where instance owners manage their n8n Cloud instance, plan, and credits. If your instance is offline or not responding, restart it from the Cloud console: Open the Cloud console and select Instance settings. Select Restart next to the instance status at the top of the page. Select Restart instance to confirm.
N
The “unknown + confirmed should never exist” rule only holds if unknown can’t be edited after the fact. A late success or a read-back by idempotency key can legitimately resolve an unknown, and if you rewrite the original line to ok, your regression query goes blind to the cases where the agent confirmed before it knew. I’d log the resolution as its own append-only event, with the same attempt id and key, outcome ok or failed, plus a resolved_by field (reconcile or human). Then “confirmed” is…
N
Hey @operantsolo, while you wait for a response, here are some things that might help: Suggested resources (click for more details) @elevatedivan, @Mutasem, @Abin_Jilson - you’ve helped with similar issues before, can you take a look? Automatically suggested by n8n’s community bot. It’s a pilot - please share feedback here.
N
What’s your go-to check when Loop Over Items only processes the first item? I usually check input count first, then the return connection, then whether a downstream node has Execute Once on. In a small lab, adding a returned-ID check on Done caught missing and duplicate IDs much faster than watching green nodes. What check do you add first?
N
Thanks all three — this is exactly the gap I was asking about. The UNIQUE constraint + status pattern makes the claim itself atomic, instead of a lookup-then-write — and minimolo’s INSERT … ON CONFLICT DO NOTHING example makes the conflict path concrete: first execution gets a row back, a duplicate gets none. The claimed_at / status timestamp point from williamsj04 is the bit I hadn’t thought through: without that, a crashed run would leave a processing claim blocking its own retry forever. One…
N
Thanks @TaskToAgent — the race-margin framing is exactly what I was missing. I’m on board with both numbers: a 2–3x check-interval grace for the check workflow itself, and two poll intervals as the starting point for the dead-trigger comparison, so I only call the trigger dead when the feed’s newest item is older than the poll interval plus that slack and the main workflow still hasn’t saved it. One follow-up: does that margin change for long intervals, like a trigger that only polls once a day?
N
This is the diagnostics follow-up I promised in the duplicate-bookings post. Same deployment (dental-clinic booking agent, chat and scheduled jobs), same incident - this time about what to log so the next ambiguous failure is findable in minutes instead of being discovered by the patient. THE CORE RULE Every tool call logs exactly one outcome, and the outcome set is ok / failed / unknown - not success / error. A timeout is not an error. It is the one state where the caller cannot tell what…
N
The checkpoint placement is the part we got wrong on the first pass too. We wrote the key inside the booking node itself — Retry On Fail replays the whole node, so every attempt minted a new key and the dedup matched nothing. Moving it out to a sub-workflow that runs before the write, keyed on patient id + slot time, was the actual fix. Worth being loud about, because node-level retry semantics make the wrong version feel like the right one. One thing we learned about the key itself: derive it…
N
Hey @Portalia_Ro, while you wait for a response, here are some things that might help: Suggested resources (click for more details) @vterzic, @Jon_James, @Anshul_Namdev - you’ve helped with similar issues before, can you take a look? Automatically suggested by n8n’s community bot. It’s a pilot - please share feedback here.
N
Hi everyone, My n8n Cloud trial instance (portalia.app.n8n.cloud, version 2.47.3) has been stuck on “In progress - Your instance is starting” since I created the account more than 2 hours ago. Opening the instance URL shows “No instance here”, and Instance settings says “Your instance is applying changes”, so I can’t restart it myself. Could someone from the n8n team please check or restart it? Thank you! Note: the n8n Cloud account that owns this instance is office@portalia.ro (this forum…
N
Self-hosting n8n inside the EU is straightforward once you pass a short checklist. These are the eight things I verify before the instance touches production traffic. Server in an EU region and a DPA with the host Pick a region in the EU and sign a Data Processing Agreement with the host before you go live. That is what you point to when a customer asks where their data lives and who processes it. A host without a DPA is a dead end for GDPR. Set N8N_ENCRYPTION_KEY yourself and back it up…
N
Two things I lean on in client workflows that nobody has mentioned yet. First, batch runs. When a workflow processes lots of items, one bad record shouldn’t kill the whole run. Set the node that hits the API to Continue using error output, branch right after it, and collect the bad ones into a separate failed-items list you log at the end. You end up fixing 3 records instead of re-running 500. Second, when a run dies mid-way, don’t rebuild it. Turn on Save execution progress in the workflow…
N
I appreciate the speed and energy of the responses . But I think the real answer is n8n does not support this. Perhaps there is a logical reason why not, and it would be interesting to hear why not (since it seems like a logical thing to add). In Appgyver (later SAP Build Apps), in the canvas where you created essential a workflow that was triggered by your UI actions, you could do this – drag a flow control onto an existing connecting line. Anyway, thanks for the responses.
N
Hi, you can use JavaScript’s toISOString() method to format the current date and time in the standard ISO 8601 format. For example, new Date().toISOString() produces a timestamp like 2026-11-10T08:55:00.000Z. If you always need T00:00:00Z, you can extract the date portion and append the fixed time string. I also recently came across an interesting gematria date tool that offers a different way to explore dates and their numerical meanings. It’s quite interesting for anyone who enjoys working…
N
Thanks kartike — all three were fair catches, and the human-flag point means a lot coming from someone who’s clearly run these on real leads. Short version: all three are now fixed, and enforced in code rather than just promised in the prompt. The Creator review also came back asking for layout fixes (overlapping notes), so I rebuilt the template properly instead of patching it: Timezone/hours — the clinic’s timezone, open days and open/close hours are now structured settings. A preferred time…