Remove memory section

[javanna]

This seems to be a leftover of #12049, where some sigar specific stats were removed. The available memory metric was left behind but never set. This section can be removed as it is not useful.

Closes #16756

[javanna]

@tlrx can you have a look please?

[tlrx]

I'm afraid it's not a left over but I broke it in #12049. We do have the information in NodeStats so I think we should sum it in ClusterStats rather than simply remove it, no?

[javanna]

yes @tlrx that is an option, I just thought that this info is not useful in cluster stats given that it's broken since 2.0 beta 1 and nobody ever complained about it... I wonder what the purpose is of showing how much memory is available in total throughout all the nodes. You confirm we should make this work instead of removing it?

[javanna]

one more thing: there are still cases where getMem() returns null I believe, then the total is completely useless as it would only take into account only the output from nodes that provided their value.

[tlrx]

one more thing: there are still cases where getMem() returns null I believe,

Do you know which cases? I think we have OsProbeTests that checks the value returned by getMem() and I don't remember it failed on the platform we test on.

I like this info because it gives an estimation of the "size" of the cluster but other than that I'm not sure it is very useful. I'm throwing the ball to @pickypg : did you ever use this field?

[javanna]

Do you know which cases? I think we have OsProbeTests that checks the value returned by getMem() and I don't remember it failed on the platform we test on.

Happened on my machine (mac) while manually testing. (I had made the change you are suggesting at first, before going for removing the field completely).

[pickypg]

did you ever use this field?

@tlrx I don't see any value in having the summed amount of memory across the cluster. It's kind of neat to "see it" but I agree that it does not add value to anyone. I can't really make any useful decision based on seeing "300 GB of memory" for a cluster. It also doesn't tell you if the nodes are heterogenous, which is dangerous.

As long as it's in node stats, then that should be what matters. I don't necessarily think that we should introduce this change for existing version though since it's a non-BWC change (even if the value was useless)?

[javanna]

I don't necessarily think that we should introduce this change for existing version though since it's a non-BWC change (even if the value was useless)?

that's debatable, the value is always set to 0 for any 2.x version. Worse than useless to me... I am all for removing it, just in case I didn't make myself clear yet :) I would be curious to hear of people actually using it...

[pickypg]

I don't think anyone is using it properly, but if they're doing some kind of mass collection of stats, then this could be swooped up. I wouldn't want to break it over a useless value in a .z release :)

[javanna]

That means we need to fix it in 2.x, I can't leave it as-is... but we agreed that this value is useless so we kinda fix it for nothing... I will need to sleep over this :)

[javanna]

ok I updated the title, description and labels of this PR. This one is now targeted at 5.0 only and will remove the fields from the response, which is a breaking change. Another PR will follow-up to fix the bug in 2.x.

[tlrx]

LGTM

[pickypg]

LGTM. Sorry for the extra hassle!

[javanna]

no worries! I also updated the migrate guide, which I had forgotten about at first ;)