Traceback with cylc play and Cylc 7 workflows

[oliver-sanders]

If you try to cylc play a running Cylc 7 workflow (or a crashed one which has left the contact file behind) the following traceback appears:

KeyError: 'CYLC_WORKFLOW_PUBLISH_PORT'

We could grep out the CYLC_VERSION from the contact file.

Should check other commands too as it is likely all client commands fail with this traceback.

[wxtim]

A very crude check suggests that the following scripts are affected:

broadcast
dump
get-workflow-version
kill
pause
ping
play
poll
reload
show
stop
trigger

(Check was)

#!/bin/bash

for script in $(cylc help all | awk '{print $1}'); do
    timeout 4 cylc $script cpbug 2> err 1> /dev/null
    if [[ $(grep CYLC_WORKFLOW_PUBLISH_PORT err) ]]; then
        echo $script
    fi
done

@oliver-sanders I think I have a fix which will cause a single failure if you try to run any of these commands against a Cylc 7 suite. Does that make sense for all these commands. In particular do we want to lose the ability to kill Cylc 7 suites?

[datamel]

This problem seems familiar... there was a similar problem with cylc scan solved by #3509, which was replaced with the new fancy scan.

[oliver-sanders]

In particular do we want to lose the ability to kill Cylc 7 suites?

We can't support Cylc 7 suites from the Cylc 8 command line because the communications layer is completely different. Users can switch to the Cylc 7 command line CYLC_VERSION=7.8.6 cylc stop --kill <suite> to stop their workflows (depending on the presence and configuration of a wrapper script).

which was replaced with the new fancy scan.

Yeah, the new fancy scan allows us to filter out workflows by Cylc version which suppresses this issue.

[wxtim]

No longer a problem·