rubenf
rubenf3mo ago

Agent workers

Hi @Chad23 , 1. yes 2. only self hosted and cloud enterprise 3. yes windmill handle remote agents. Some customers have 600 of them: https://www.windmill.dev/docs/core_concepts/agent_workers
Agent Workers | Windmill
Agent Workers are a Cloud plans and Self-Hosted Enterprise Feature.
4 Replies
Alper
Alper3mo ago
Even with restricted permissions agent workers are able to corrupt data via update on queries. That makes it impossible to crowd source workers for very large projects (think seti@home or similar). This could be solved if workers have an alternate mode that connects to the API instead of the DB and then having granular tokens. Or maybe there could be one dedicated table per worker. Those could then be combined by materialised views.
rubenf
rubenfOP3mo ago
That's not true @Alper , you can use specific postgresql user per worker agents the customer we are working with is in a very secure setup it does require our expertise on setting up RLS permissions which we have tried to distill in the article above It doesn't mean we can't improve the way it currently works but http apis do not have the same capabilities as postgres transactions so if we were to do a pure http api mode, it would come at the expense of either performance, correctness, observability or reliability of the workflow engine (executing jobs itself is trivial enough without pg transactions)
Alper
Alper3mo ago
thanks for the clarification. i wasn't aware that the tradeoffs of using an http API would be that big also, i need to understand better how the RLS separation would work, will investigate
rubenf
rubenfOP3mo ago
RLS is as secure as you can get since it's the database enforcing the rules at the row-level you can't cheat RLS (but rules need to be well set) RLS is likely much more auditable than code logic to ensure security