On Wed, 2018-05-16 at 20:13 +0000, Calciano, Jess wrote: > Hello, > > In kernel versions >= 4.16, we're intermittently seeing a ~0.3 sec delay in attaching SCSI devices (iSCSI, iSER, or SRP) to an initiator, measured from the time the iscsiadm command completes. > > For example: > 1. Create 4 SCSI LUNs under a wwn (we use targetcli) > 2. Connect the initiator to the target LUNs: > # iscsiadm -m node -l -T <target-wwn>:ib0 -p <target-ip-addr>:3260 > 3. Result: Devices are attached, but maybe 25% of the time, /dev/sdb is attached 0.2-0.4 seconds after /dev/sdc, /dev/sdd, and /dev/sde. > During the delay, the output of lsscsi looks like this: > [10:0:0:0] disk LIO-ORG iser_ib0_0 4.0 - <-- unavailable /dev/sdb > [10:0:0:1] disk LIO-ORG iser_ib0_1 4.0 /dev/sdc > [10:0:0:2] disk LIO-ORG iser_ib0_2 4.0 /dev/sdd > [10:0:0:3] disk LIO-ORG iser_ib0_3 4.0 /dev/sde > > The system log also reflects the delay. /dev/sdb is the first device to start setup, but it's the last to finish. > > Is this behavior expected? Is there something else we need to configure? > Or is this possibly a bug introduced in 4.16? Does the following patch help: "[PATCH] sd: Increase SCSI disk probing concurrency" (https://www.spinics.net/lists/linux-scsi/msg115657.html)? Bart.��.n��������+%������w��{.n�����{���fk��ܨ}���Ơz�j:+v�����w����ޙ��&�)ߡ�a����z�ޗ���ݢj��w�f
