Worker gets stuck without error and fails to grab its assigned work request

I created an update_something_environments workflow on debusine.debian.net. It creates a pile of mmdebstrap and simplesystemimagebuild tasks most of which are in state pending. debusine-worker-amd64-hades-05 remains idle despite there being more than 200 pending work requests (also autopkgtest ones). It seems like the scheduler fails to schedule any work requests on this worker. The dynamic_metadata of the worker looks ok-ish.

dynamic_metadata of hades-05

{"blhc:version": 1, "noop:version": 1, "sbuild:version": 1, "debdiff:version": 1, "lintian:version": 1, "piuparts:version": 1, "sbuild:available": true, "system:cpu_count": 4, "mmdebstrap:version": 1, "system:worker_type": "external", "autopkgtest:version": 1, "mergeuploads:version": 1, "mmdebstrap:available": true, "system:architectures": ["amd64", "i386"], "autopkgtest:available": true, "executor:qemu:available": true, "system:host_architecture": "amd64", "extractforsigning:version": 1, "executor:unshare:available": true, "executor:incus-vm:available": true, "assemblesignedsource:version": 1, "executor:incus-lxc:available": true, "system:total_physical_memory": 16746045440, "simplesystemimagebuild:version": 1, "makesourcepackageupload:version": 1, "simplesystemimagebuild:available": true}

/cc @stefanor started debugging this and Colin agreed that this is not behaving as expected.