In our set up, we have a WUI host that runs all of the Cylc UI Servers and scheduler hosts that run all the workflows.
All hosts share the same $HOME, and have common global.cylc files.
The scheduler hosts are allowed to talk to the job host, but the WUI host is not.
The “available” hosts as defined in the global.flow file do not include the WUI host. Yet if a stopped workflow is “played” via the WUI it launches on the WUI host, which causes submission failures because the WUI host can’t talk to the job host.
Is this expected behaviour? Do you have any suggestions on workarounds?
To help diagnose the problem, can you take the web UI out of the equation? If you launch a stopped workflow via cylc play on the command line, whilst logged on to the web UI host, does the same thing happen?
To help diagnose the problem, can you take the web UI out of the equation
Good idea, the play button in the UI causes the cylc play button to be run on the host that runs the UI Server.
Also worth checking the output of cylc config and possibly inspecting any CYLC_* environment variables that could mess with the location of the Cylc config file (e.g. CYLC_CONF_PATH).