Work snippetMonitoring + reliability

n8n and Make error monitoring, with retries, a dead-letter queue and alerts

How I would approach 14 automations across n8n and Make for a five-branch group, so each failure is classified, retried only when safe, parked with its full payload, and reported to the right person by severity.

Work snippet with example figures.

Area: Monitoring + reliability
Tools: n8n · Make · Airtable · Slack
Typical scope: US$1,500–3,000
Timeline: About 4 weeks · 4 fixed milestones

Context and systems map

A 60-tech plumbing and electrical group with five branches and 14 automations spread across Make and n8n: job creation, invoice pushes, supplier bill intake, review requests and nightly reports. Failures are found by someone noticing that a number looks wrong.

The systems in play

In the example each automation is wrapped with the same error pattern: classify the failure, retry what is safe to retry, park what is not in a dead-letter queue with the full payload, and tell the right person in the right channel. Silent failures are caught by heartbeat checks.

Systems map: who uses what

System	What it does for the business
Make scenarios	Eight scenarios that connect ServiceM8, Xero and HubSpot. Each gets the shared error handler.
n8n workflows	Six workflows for supplier bills, reports and webhooks. Each gets the same error workflow.
Error workflow	One shared workflow that classifies an error, decides retry or park, and writes the log.
Dead-letter queue	An Airtable table holding each failed item: payload, error, attempts, owner and status.
Alert channels	A Slack channel for operations and email for the finance lead, by severity.
Heartbeat monitor	A scheduled check that expects each automation to have run inside its window.

How the data moves

Automation step fails
Error workflow
Classify
Retry with backoff
Dead-letter queue
Alert by severity
Owner resolves
Replay or close

Problem

What the owner sees

An invoice push that failed three weeks ago and nobody knew
A flood of alerts that people learn to ignore
Failed items that vanish with no payload to replay
Fixes that create duplicates when the item is replayed

Automations rarely stop with a bang. They fail on one record, return a rate-limit error at month end, or quietly stop being triggered after a credential expires. The business needs a pattern that makes each failure small, visible and replayable, without creating duplicates when it is retried.

Where the drift starts

Everyday situation	What goes wrong
Rate limit at month end	A burst of invoices fails and is never retried
Expired credential	Every run fails the same way until someone logs in
Record missing in the target	The error says 404 and the item is lost
Validation error from bad data	The same bad record fails on every retry
Retry without a key	The item is created twice when the first attempt actually succeeded
Scenario switched off by accident	Nothing fails; nothing runs either

Approach

Four steps, in this order. Each one ends with a result that is checked.

LookInventory of every automation: trigger, steps, owner, what a failure costs and how it is noticed today.
Agree the rulesError types, retry schedule, severity levels, owners and the replay procedure are written down.
Build and testThe shared error handler is added to each automation in a copy first, then forced to fail in every class.
Go live in orderHighest-risk automations first. Alert thresholds are tuned for two weeks so noise does not train people to ignore it.

Rules agreed in writing (example settings)

Item	Example setting
Retry types	Rate limit, server error and timeout are retried. Authentication, validation and missing record are not.
Retry schedule	Five attempts with backoff of 1, 2, 4 and 8 minutes
Idempotency	Every write carries a key built from source id and action, so a replay never creates a second record
Dead-letter queue	Any item that fails the last retry or a non-retry class is parked with its full payload
Severity	Critical: money or customers affected. Warning: delayed, will retry. Info: daily digest.
Heartbeat	Each automation has an expected run window and a maximum gap
Replay	Only from the queue, by a named person, with the cause recorded
Owner	Every automation has one named owner and a backup

Flow and data mapping

The first table is the decision made for every error. The second shows what is stored with each failed item so it can be replayed safely.

Error class, decision and alert

Error class	Example	Decision	Alert
Rate limit	HTTP 429 from the accounting API	Retry with backoff	Warning if all retries fail
Server error or timeout	HTTP 5xx, no response in 30 seconds	Retry with backoff	Warning, then Critical if the queue grows
Authentication	HTTP 401 or 403, expired token	No retry; park and pause the automation	Critical, immediately
Validation	HTTP 400, missing tax code	No retry; park with the field named	Warning to the data owner
Missing record	HTTP 404, customer not found	No retry; park	Warning to the data owner
Already exists	HTTP 409 on a replay	Treat as success, log	None
No run in window	Heartbeat missed	Alert	Critical for money flows, Warning otherwise

What is stored for each failed item (example settings)

Field	Content	Use
Item id	Source record id and action	Idempotency key
Automation	Name and version	Find the owner and the steps
Payload	Full input as received, secrets removed	Replay without asking the source again
Error	Class, code and message	Triage
Attempts	Count and times	Retry history
Owner and status	Open, in progress, replayed, closed	Queue discipline
Cause	One of eight causes, chosen on close	Monthly review of the biggest causes

Edge cases handled

Agreed in advance with the finance lead and the operations manager, and tested before go-live.

Edge cases and how each is handled

Situation	How it is handled
Retry succeeds after the original actually worked	Idempotency key makes the second write return the first record
Same bad record fails every day	Parked once; the repeat is counted against the same item, not re-alerted
Credential expires at night	Critical alert at the first failure; the automation pauses so it does not fill the queue
Alert storm after an outage	Alerts grouped by automation and sent as one message every 15 minutes
Webhook sent twice by the source	Duplicate key found, second ignored and counted
Replay of an item that is now out of date	Replay checks the current state of the record first and asks for a decision
Error handler itself fails	Second-level email to Manaz Afzal and the owner, with the raw error
Maintenance window	Alerts muted for the named automations for a fixed time, with a log entry

Worked example

The tables below show one month for an example group with 14 automations. They show the shape of the checks, not a client result.

Retry schedule for a rate-limit error

Attempt	Wait before it	Elapsed since first failure
1	Immediately	0 min after first failure
2	1 min	1 min after first failure
3	2 min	3 min after first failure
4	4 min	7 min after first failure
5	8 min	15 min after first failure

Dead-letter queue by cause (example figures)

Cause	Items	Replayed	Closed without replay
Missing tax code on a bill	11	11	0
Customer not found in Xero	6	4	2
Expired credential	2	2	0
Source sent an invalid date	3	1	2
Total	22	18	4

Alert volume by severity for the month (example figures)

Severity	Alerts sent	Went to
Critical	3	Finance lead by email, operations on Slack
Warning	31	Operations on Slack, grouped
Info	30	One daily digest message

Heartbeat windows (example settings)

Automation	Expected	Maximum gap
Invoice push to Xero	Every 15 minutes, 06:00 to 20:00	45 minutes
Supplier bill intake	Every 30 minutes	2 hours
Nightly branch report	Once, 02:00	26 hours
Review requests	Daily, 16:00	26 hours

Monitoring and alerting

This whole piece is about monitoring, so the alerts are designed to be few. Each one names the automation, the item, the likely cause and the first action.

What is watched, and who hears about it

Signal	Threshold	Alert goes to	First action
Critical failure	Any money or customer flow fails its last retry	Finance lead email and operations Slack	Owner opens the item, fixes the cause, replays
Queue size	More than 10 open items or any older than 24 hours	Operations Slack	Owner reviews the queue the same morning
Heartbeat missed	Gap above the maximum	Owner and operations Slack	Trigger and switch status checked
Credential expiry	14 days before and on the day	Owner email	Credential renewed and tested
Error handler failure	Any	Manaz Afzal and owner by email	Handler fixed, items replayed
Weekly digest	Counts by automation and cause	Owner and finance lead	Review of the top cause

Test plan

Every check has a written pass condition and a written fail condition before it is run. A FAIL stops the next step until it is cleared and re-run.

Test plan with PASS and FAIL conditions

ID	Check	Method	PASS when	FAIL when
T1	Rate limit	Force a 429 in a copy	Five attempts on the schedule, then success or park	Fewer attempts or no park
T2	Authentication	Revoke a test token	No retry, paused, Critical alert at once	Retries or no alert
T3	Validation	Send a bad record	Parked with the field named	Retried or lost
T4	Idempotency	Replay a successful item	No second record	Duplicate created
T5	Heartbeat	Switch off a test automation	Alert inside the maximum gap	No alert
T6	Alert grouping	Fail 20 items at once	One grouped message per window	20 messages
T7	Replay	Fix the cause and replay	Item created once, queue closed with the cause	Item stuck or duplicated
T8	Handler failure	Break the handler in a copy	Second-level email sent	Silence

Go-live and rollback

The handler is added automation by automation. Each one can be unwrapped in minutes, because the original steps are unchanged.

Rollout order and rollback triggers

Step	Owner	Gate to continue	Rollback trigger
Invoice push and bill intake	Manaz Afzal	T1 to T4 pass in a copy	Any duplicate in the target
Review requests and reports	Manaz Afzal	T5 passes	Missed heartbeat not alerted
Remaining automations	Operations manager	Queue worked for a week	Queue above 10 for three days
Alert tuning	Manaz Afzal and owner	Two weeks of review, noise below 5 alerts a day	Owner asks to mute a Critical alert
Rollback	Owner	Handler removed from the automation, original steps untouched	Decision recorded with the reason

Scope and milestones

A fixed scope agreed in writing before anything starts. Each milestone is paid when you sign off its deliverable.

Scope: what a project like this covers

Area	Included
Systems Audit	Read-only review of every Make scenario and n8n workflow, owners and failure history, a 45-minute call, written findings and a fixed quote
Inventory	Every automation with owner, trigger, steps and failure cost
Shared error handler	One pattern for Make and one for n8n, with types and retries
Dead-letter queue	Table, views, owner and status, with replay steps
Alerts	Slack and email by severity, grouped and tuned
Heartbeats	Windows and gaps for each automation
Runbook	What to do for each alert, one page each
Handover	Written SOPs, a recorded handover video and files you own, plus 14 days of fixes after go-live
Not included	Tax filing and statutory returns.
Also outside the scope	Rebuilding the automations themselves. Changing the tools. A 24-hour on-call service. Software subscriptions (billed to you by the vendors).

Fixed milestones

Milestone	Deliverable	Timing	Share
M1 Audit and rules	Inventory, fixed quote, error rules in writing	Week 1	25%
M2 Handler and queue	Handler and queue built and tested in a copy	Week 2	30%
M3 Rollout	Handler live on all automations, heartbeats on	Week 3	30%
M4 Tuning and handover	Alert tuning, runbook, SOPs, handover video, 14 days of fixes	Week 4	15%

Typical scope for this kind of project: US$1,500–3,000

About 4 weeks across 4 fixed milestones. The Systems Audit (US$300, fixed) comes first and confirms the scope.

Describe what is going wrong between your job system and your books, and I will say where I would look first.

Send a short note through the form on the home page. I will reply with where I would start, and whether the US$300 Systems Audit is the right first step.

Hire me on Upwork (opens in a new tab)Hire me on Freelancer (opens in a new tab)Send a message