Standardization of additional meta information of instances

[berendt]

It looks as if certain dynamic meta information (ps_scheduled_downtime, ps_restart_after_maint ) has been introduced on the PCO which provides the instance with additional information to which it can react:

plusserver/docs@4113402

As in this specific case, this makes it possible to notify the instance of an upcoming reboot and to influence the restart behavior after a host fail.

The meta information introduced on the PCO seems to be independent of SCS, it is marked with a prefix ps. From my point of view, it would be useful for the implementation of standard operations, especially with regard to this specific case with local storage, to define this meta information in an SCS standard.

This would also open up possibilities to create tooling for within the instances. Other useful meta information is obvious. You could also inform about live migrations, a flavor or image that will soon no longer be available. However, this meta information is only useful if it is uniform and well-defined. If we now have meta information of this type on individual CSPs and not on others, then we have functional differences from the user's perspective when using the instances. This also has an impact on operations as this meta information needs to be generated and distributed at the right time.

[garloff]

The behavior and the instance metadata would be useful to standardize.

[mbuechse]

today's meeting of Team IaaS concluded (tentatively):

• either lay down that SCS will never standardize metadata (it will never be portable)
• or lay down that SCS will use a specific prefix