Add a hwdb format option

[kov]

This new subcommand generates an entry in the iocost hardware database format for the given merged result. It is used by the CI pipeline to generate the hwdb file that will be used by systemd.

Signed-off-by: Gustavo Noronha Silva [email protected]

[kov]

Not 100% happy with having to look at the props to make opts decisions, but I can't think of any way other than that. Should we add a second format subcommand that is undecorated, for instance?

Another thing I'd like your input on is on the default solutions. The reason I did that is as we've seen on some of the results some solutions are missing, so I thought we should have a list of preferred solutions and get the best one we find. I expect this will not be a big issue once we have enough results, as it'll be rare that a merged result is missing a solution, but still good to have it handled properly.

[htejun]

re. undecoreated, I think what you did is fine. Eventually, what we need to do is separating out the format option parse phase into a separate callback so that the benchmark has a say in whether a given format command should be decorated or not in a non-hacky way but this doesn't concern the user, so this can and should be handled internally.

So, I'm not sure whether having default makes sense at hwdb generation step. It really should contain all the solutions generated for the device. In terms of work-conersvancy (if the aim is not losing any work that the ssd can do), the order would be no control -> naive -> bandwidth -> isolated-bandwidth -> isolation. If the aim is isolation, the order is reversed. It makes sense to present the ordering but I'm not sure whether whether we can pick one solution as "the default" at hwdb generation time.

[kov]

haha I was just writing a message to suggest we land this version =D thank you @htejun!