> Op 9 februari 2017 om 9:41 schreef Özhan Rüzgar Karaman <oruzgarkaraman@xxxxxxxxx>: > > > Hi Wido; > Thanks for fast response rbd ls -l reads all images header for its sizes > yes it makes sense you are right. > > My main problem is when i refresh a rbd storage pool using virsh over > kvm(Ubuntu 14.04.5) it takes too much time then the old days and i suspect > that virsh makes "rbd ls -l" over Ceph storage so thats why i asked. > > Does virsh use same "rbd ls -l" for pool refresh? > Yes, it does: http://libvirt.org/git/?p=libvirt.git;a=blob;f=src/storage/storage_backend_rbd.c;h=45beb107aa2a5c85b7d65b8687c2b65751871595;hb=HEAD#l425 In short, this C code does (pseudo): images = [] for image in rbd_list(): images.append(rbd_stat(image)) The more images you have, the longer it takes. > So in this case below 22 second is normal for virsh rbd pool refresh? > Yes. One of my goals for libvirt is still to make this refresh a async operation inside libvirt, but that's a bit difficult inside libvirt and have never gotten to actually implementing that. Wido > root@kvmt1:~# time virsh pool-refresh 01b375db-d3f5-33c1-9389-8bf226c887e8 > Pool 01b375db-d3f5-33c1-9389-8bf226c887e8 refreshed > > > real 0m22.504s > user 0m0.012s > sys 0m0.004s > > Thanks > Özhan > > > On Thu, Feb 9, 2017 at 11:30 AM, Wido den Hollander <wido@xxxxxxxx> wrote: > > > > > > Op 9 februari 2017 om 9:13 schreef Özhan Rüzgar Karaman < > > oruzgarkaraman@xxxxxxxxx>: > > > > > > > > > Hi; > > > I am using Hammer 0.49.9 release on my Ceph Storage, today i noticed that > > > listing an rbd pool takes to much time then the old days. If i have more > > > rbd images on pool it takes much more time. > > > > > > > It is the -l flag that you are using in addition. That flag opens each RBD > > image and stats the header of it to get the size. > > > > A regular 'rbd ls' will only read the RADOS object rbd_directory, but it > > is the -l flag which causes the RBD tool to iterate over all the images and > > query their header. > > > > > My clusters health is ok and currently there is no load on the cluster. > > > Only rbd images are used to serve to vm's. > > > > > > I am sending some information below. My level.db size is also 280 mb, i > > > also compacted level.db to 40 mb size but again "rbd ls -l" output is too > > > slow. > > > > > > This timing is important for my vm deploy time to complete because when i > > > refresh a pool/datastore it takes nearly to 20 seconds or more for 350 > > rbd > > > images+snapshots. > > > > > > Thanks for all help > > > > > > Regards > > > Ozhan Ruzgar > > > > > > root@mont3:/var/lib/ceph/mon/ceph-mont3/store.db# ceph -s > > > cluster 6b1cb3f4-85e6-4b70-b057-ba7716f823cc > > > health HEALTH_OK > > > monmap e1: 3 mons at > > > {mont1=172.16.x.x:6789/0,mont2=172.16.x.x:6789/0,mont3= > > 172.16.x.x:6789/0} > > > election epoch 126, quorum 0,1,2 mont1,mont2,mont3 > > > osdmap e20509: 40 osds: 40 up, 40 in > > > pgmap v20333442: 1536 pgs, 3 pools, 235 GB data, 63442 objects > > > 700 GB used, 3297 GB / 3998 GB avail > > > 1536 active+clean > > > client io 0 B/s rd, 3785 kB/s wr, 314 op/s > > > > > > root@mont1:~# time rbd ls -l cst2|wc -l > > > 278 > > > > > > real 0m11.970s > > > user 0m0.572s > > > sys 0m0.316s > > > root@mont1:~# time rbd ls -l cst3|wc -l > > > 15 > > > > > > real 0m0.396s > > > user 0m0.020s > > > sys 0m0.032s > > > root@mont1:~# time rbd ls -l cst4|wc -l > > > 330 > > > > > > real 0m16.630s > > > user 0m0.668s > > > sys 0m0.336s > > > _______________________________________________ > > > ceph-users mailing list > > > ceph-users@xxxxxxxxxxxxxx > > > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com > > _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |