Checkpoint is not reverted when role aborts with log_error()

[tyll]

log_error() does not only log but also exists the module. In this case, the role does not yet revert the checkpoint. This needs to be changed.

[pcahyna]

does it mean that #119 introduced a regression? Or did it just expose a bug present before?

[pcahyna]

Instead of fixing the module so that it always reverts the checkpoint, wouldn't it be better to let NM itself revert the checkpoint (after a timeout)? There can always be unforeseen circumstances where the module would abort, so letting NM revert the checkpoint would be more robust.

[tyll]

does it mean that #119 introduced a regression? Or did it just expose a bug present before?

Previously the role did not use checkpoints and the new implementation is faulty.

Instead of fixing the module so that it always reverts the checkpoint, wouldn't it be better to let NM itself revert the checkpoint (after a timeout)? There can always be unforeseen circumstances where the module would abort, so letting NM revert the checkpoint would be more robust.

There is also a timeout that will revert the checkpoint eventually. But in the meantime a possibly partial configuration might be active and the role will fail until the checkpoint is taken care of. Reference:
https://github.com/linux-system-roles/network/blob/master/library/network_connections.py#L1895

[pcahyna]

Symptom: when the role fails (for whatever reason), subsequent invocation fails with a message like

msg: 'fatal error: failure to checkpoint_create checkpoint: nm-manager-error-quark: device ''lo'' is already included in checkpoint /org/freedesktop/NetworkManager/Checkpoint/268 (12): {''error'': "nm-manager-error-quark: device ''lo'' is already included in checkpoint /org/freedesktop/NetworkManager/Checkpoint/268 (12)", ''success'': None}'

Workaround: wait 90 seconds for each profile one specified when the role failed originally.
(from #190)

[tyll]

@9minds thank you for closing #190. Could you maybe share how you triggered the failure? It might help to add a test case.

[tyll]

Related bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1816140

[tyll]

Workaround: wait 90 seconds for each profile one specified when the role failed originally.
(from #190)

Another workaround:

1. Get the name/ID of the checkpoint:

gdbus introspect --system --dest org.freedesktop.NetworkManager --object-path /org/freedesktop/NetworkManager/Checkpoint --recurse | awk '/^ *node /{print $2}'

2. Revert it:

gdbus call --system --dest org.freedesktop.NetworkManager --object-path /org/freedesktop/NetworkManager --method org.freedesktop.NetworkManager.CheckpointRollback /org/freedesktop/NetworkManager/Checkpoint/66