[PVE-User] blocked requests

Holger Hampel | RA Consulting h.hampel at rac.de
Tue Mar 14 10:24:07 CET 2017


we don't use a scheduler on OSDs, so priority has no effect. We use this to reduce load during working ours:

        osd scrub begin hour = 18
        osd scrub end hour = 7
        osd scrub load threshold = 3
        osd max scrubs = 1
        osd max backfill = 3
        osd recovery max active = 5

At least " load threshold" has tob e adopted tot he (cpu) power of the host.

Holger Hampel

-----Ursprüngliche Nachricht-----
Von: pve-user [mailto:pve-user-bounces at pve.proxmox.com] Im Auftrag von lists
Gesendet: Montag, 13. März 2017 18:44
An: PVE User List
Betreff: [PVE-User] blocked requests


Another question, if I may... :-)

We are seeing occasional slow requests on our three-node pve ceph 0.94.9 
cluster. I've seen it happening after ceph deep-scrubs start. Googling 
this, refers me to this page on lowering ceph scrub I/O priority:


I'd just like to ask here: Anyone followed that procedure? Seems the 
pve/ceph is recent enough. Any insights to share?

Best regards again,
pve-user mailing list
pve-user at pve.proxmox.com

More information about the pve-user mailing list