verify the number of osds deployed

[Joe Talerico]

TripleO should verify the number of OSDs that are actually deployed before reporting the deployment succeeded. I have seen where TripleO reports Deployment Complete, however the # of OSDs was much lower than what I was expecting.

It should be quite simply to sum the number of ceph-storage nodes + the osd map, then look at ceph-s on the controller to determine if the number of osds match. <-- possibly a gross over simplification.

[Ben England (bengland)]

So one extreme is to declare the deployment complete when 1/3 of the OSDs are there that we expected, and the other extreme is to require *all* osds - in a large-scale deployment, it is possible that some drive will have died somewhere in the system, isn't seated properly, etc. etc. In any sufficiently large cluster, probabilities dictate that some hardware somewhere is always offline. So IMHO we should declare the deployment complete if some percentage of the OSDs show up, and make this percentage an adjustable parameter with 95% as the default value. In other words, we declare it complete if enough of the hardware becomes available that we can safely start to use the system and fix the missing OSDs later, as a day-2 activity. For example, if we have a 3-OSD-node cluster with 24 drives per OSD, this would be 72 OSDs, and 95% of that would be 68 OSDs. Make sense?

[Joe Talerico (jtaleric)]

That would assume we allow partial deployment, I think we default it to 100%, but allow the user to tune this value, thoughts?

[Juan Antonio Osorio Robles (juan-osorio-robles)]

Is this still being worked on?

[John Fulton (jfulton-org)]

A variation of https://review.opendev.org/#/c/657175 but for OSD percentage could be used to do this.

[John Fulton (jfulton-org)]

We addressed a variation of this in the following:

 https://review.opendev.org/#/c/657175/
 https://review.opendev.org/#/c/657173/