Currently there is not clear and sound documentation on how users should set up eremetic so that scheduler failovers are supported. Since currently eremetic does not itself persist its framework_info or master-assigned framework_id, users are required to work around this.
Since Mesos does not have explicit support for self-assigned framework_id values, the workaround mentioned in #181 is not safe. Mesos will start enforcing some validation on used framework_id values with MESOS-9469 which might break workflows where users self-assign certain values.
A sound approach would be
start the scheduler with no framework_id set in the config file
watch logs for the assigned framework_id
update config file to contain the master-assigned framework_id
reregister scheduler
Eremetic should have some clear documentation on how to set up eremetic in a sound way.
Thanks for the heads up @bbannier that does sound like the simplest way of doing this for now. Maybe we could display the frameworkid somewhere to make it easier to find.
Currently there is not clear and sound documentation on how users should set up eremetic so that scheduler failovers are supported. Since currently eremetic does not itself persist its
framework_info
or master-assignedframework_id
, users are required to work around this.Since Mesos does not have explicit support for self-assigned
framework_id
values, the workaround mentioned in #181 is not safe. Mesos will start enforcing some validation on usedframework_id
values with MESOS-9469 which might break workflows where users self-assign certain values.A sound approach would be
framework_id
set in the config fileframework_id
framework_id
Eremetic should have some clear documentation on how to set up eremetic in a sound way.