Question 1

Explain the concepts of node, connection and execution in n8n.

Accepted Answer

Do not just recite definitions — walk through a workflow. "A node is one step. There are three kinds: triggers that start the workflow, such as Schedule or Webhook; core nodes for logic and generic capability, such as IF or HTTP Request; and app nodes exposing one service's operations, such as Slack. A connection is the line between nodes and defines the order data flows in. One node can branch into several paths — the true and false outputs of an IF node are the obvious example. An execution is one run of the workflow. It can be manual, partial or automatic; each is stored in the history so you can open it and inspect every node's input and output. That history is my main debugging tool."

Question 2

What is a cluster node and how does it differ from a normal node?

Accepted Answer

This question is usually a doorway into the AI topic — make that connection yourself. "A normal node works alone: input in, output out. A cluster node is a group — a root node in the middle with sub-nodes attached underneath that extend what it can do. The most common example is the AI Agent: the agent is the root node, and a Chat Model (which model it uses), Memory (conversation history) and one or more Tools (what the agent can call) attach to it. The practical difference: sub-nodes are not in the main data flow — they are resources the root node reaches for during execution. So it is wrong to read a sub-node as 'the next step'."

Question 3

You published a workflow but it is not running. How do you diagnose it?

Accepted Answer

Give an ordered checklist — that is what shows systematic thinking. 1. Is the workflow actually active? Check the publish state. 2. Look at the Executions list — no executions at all means the problem is at the trigger; executions that fail mean the problem is inside. 3. Check the trigger type — for a Webhook, is the external system calling the production URL or the test URL? The test URL only works while "Listen for test event" is armed. 4. For a Schedule Trigger, check the timezone — the instance timezone may differ from what you assumed. 5. The trigger node itself can fail — in that case the error workflow receives a different payload shape, because the workflow never really executed. "This sequence localises the problem to either the trigger or the body — after that it is ordinary debugging."

Node category	What it does	Example
Trigger node	Starts the workflow; shown with a bolt icon on the canvas	Schedule Trigger, Webhook, Gmail Trigger
Core node	Serves general functionality rather than one service: logic, scheduling, generic API calls	HTTP Request, IF, Code, Merge
App node	Exposes the operations of one specific external service	Google Sheets, Slack, Telegram, Postgres
Cluster node	A group made of one root node and the sub-nodes attached to it	AI Agent (root) + Chat Model, Memory, Tool (sub-nodes)

Workflow, node, connection and execution

📚 Sources and documentation