Beginner pitfall: I didn't realize every node runs in the same process. No isolation.

Hey everyone, I just hit a classic beginner snag with LangGraph that I think others should be aware of early on.

I was building a sales qualification workflow, with separate nodes for enriching lead data, checking against our internal blacklist, and then doing a sentiment analysis on the lead's last email. I built each node as its own tidy function, assuming they’d run in some isolated way—maybe like separate microservices or at least separate threads.

Nope! 😅 I learned the hard way that by default, **all nodes execute in the same single process**. This caused two issues for me:
* A memory leak in my enrichment logic (from a third-party library) started affecting the sentiment analysis node.
* A long-running, blocking call in the blacklist check held up the *entire* graph execution.

It seems obvious in retrospect, but coming from a background where tasks are often distributed, I didn't even think to check. There’s no automatic isolation unless you explicitly build for it.

So, a heads-up for fellow builders: if you have a node that’s resource-heavy, prone to crashes, or might block, you probably need to look into:
* Using LangGraph's ability to call external actors or subgraphs.
* Wrapping risky operations in separate processes yourself.
* Designing nodes to be as fault-tolerant as possible.

Has anyone else run into this? What patterns are you using to handle node isolation or faults in production?

— Aiden