Description of problem: ----------------------- On a particular testing with RHHI, it was observed that when VMs are created from template, one after the other before completion of one VM creation, VMs are stuck in 'Image Locked' state When looked at the gluster logs, it was found that the reads are going beyond the shard-size and XFS is not returning proper errno. So clearly we found out 2 issues 1. XFS issue ( BZ 1473549 ) 2. gluster shard problem This bug is to address the second point. Version-Release number of selected component (if applicable): -------------------------------------------------------------- RHGS 3.2.0 async glusterfs-3.8.4-18.6.el7rhgs How reproducible: ----------------- Hit it for the first time, never tried to reproduce Steps to Reproduce: ------------------- This particular issue was seen with the RHHI case There could be a much simpler step, but let me post what I have seen 1. Create RHV setup backed with gluster replica 3 volume ( XFS bricks ) 2. Create RHEL 7.4 template 3. Create multiple VMs from the RHEL 7.4 template ( one after other, before the completion of previous VM ) Actual results: --------------- read on shard has gone beyond the shard size [2017-07-20 12:50:59.982435] E [MSGID: 113040] [posix.c:3119:posix_readv] 0-vmstore-posix: read failed on gfid=0aac1b8b-0ccc-4936-922c-efb524d226e3, fd=0x7fd7c401989c, offset=3584 size=2048, buf=0x7fd7d8af7000 [Unknown error 3072] Expected results: ----------------- read syscall on shard should not have gone beyond shard-size
Wrt the specific logs highlighted in comment #2, the offset is 4193792, which is 512bytes short of 4194304(=4 * 1024 * 1024 = 4MB). So it is a legitimate read; Also in light of the clarification provided at https://bugzilla.redhat.com/show_bug.cgi?id=1473549#c31, I'm closing this bug. -Krutika
On Sat, Jul 22, 2017 at 11:09 PM, Pranith Kumar Karampuri <pkarampu> wrote: I am sorry, I calculated 4MB wrong. Read came inside 4MB size, but I was under the impression it was beyond 4MB size. There is no bug in shard. I am going to update the bz as well.