After setting an execution mode in a TFE workspace to “agent” and configuring a workspace notification for all events in the same workspace to trigger scaling of TFE agents in Fargate via an API GW and a Lambda, I have noticed that Speculative plans don’t trigger workspace notifications, and, as a result, the MR pipeline (VCS-driven workflow) gets stuck as the speculative plan cant finish.
Is this is expected behavior or a bug? TFE version v202111-1.
The only workaround option I see is to have at least one TFE agent running 24/7. However, with dozens of environments and workspaces, each of which has its own TFE agents in separate AWS accounts, this would result in quite a lot of TFE agents running in Fargate, which brings additional costs.
Any suggestions would be appreciated.