Hi Thomas, for pressure, I don't really need rrd or history, as I can already broadcast last 10s,60s, 300s. so could simply broadcast them in kvstore.
This was the original patch in 2020 https://lists.proxmox.com/pipermail/pve-devel/2020-October/045493.html but Dietmar asked about to put them in rrd ;) So, as you want, just tell me :) I don't care about rrd for my need. Anyway, For vm balancing, I was planning to compute average metric instead parsing at lot of rrd. for ksm, I really need current value for qemu hostcpu/hostmem, last (1-5) minutes average could be enough, so I could compute from pvestatd too. Alexandre Le mercredi 25 mai 2022 à 17:03 +0200, Thomas Lamprecht a écrit : > On 25/05/2022 08:52, Alexandre Derumier wrote: > > Hi, > > > > I'm still working on vm balancing/scheduling, and need some new > > metrics. > > > > but you don't actually need the metrics themselves but only the > computed > moving avg result and such of them? Because if that's the case we > could just > keep the last X values we require for the moving window cached > locally in > pvestatd and use those to broadcast the calculated ones use for DRS > once or > twice every minute or so via the per-node kv status in the pvestatd? > > Could mean less data an no new RRD format/paths (iow. no touching of > pve-cluster at all). What do you think? > > _______________________________________________ pve-devel mailing list pve-devel@lists.proxmox.com https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
