Description of problem: Currently it is not possible to use LVM on top of iSCSI devices. Basically, LVM initialization happens during disk initialization, which is right at the start of init. Where as the network and iSCSI (also multipath) services start way later. Hence, without iSCSI being started, LVM is not able to initialize the PVs because iscsid isn't started and no LUNs are mapped to the host. There is a work around to this issue by manually re‑scanning the PVs after the OS has completely booted. But this will be a problem if there are applications which are configured to use the LV as their data path because the applications will start automatically on system startup and will fail by not finding the Logical Volume/Mount Point. Version-Release number of selected component (if applicable): All How reproducible: Always Steps to Reproduce: 1. Configure iSCSI and map some LUNs to the host 2. Create a VG/LV on top of those LUNs 3. Set them to auto-mount at boot. 4. Upon boot, the devices aren't mounted because the PV/VG aren't seen to the OS Actual results: Logical Volumes are not found because the LUN was not mapped at the time of LVM initialization which further leads to failure of mount points which are part of LVM Expected results: Device-Mapper should have the ability to do an LVM re-scan when it senses that a new block device has been plugged in. Additional info: This isn't specific to iSCSI. Can be reproducible with any block device providing application which starts-up in init after LVM initialization is done.
Did you try to add the _netdev option to fstab? There is some check in the netfs script where if it is set, then it kicks off /sbin/lvm.static vgscan.
Per NetApp, more testing will be done on this, but outlook is good on testing the GA RHEL 5 version.
The _netdev option might help. Except with GFS2 where the _netdev option leads to a mount error. Workaround here: Change gfs2 init.d script to include vgchange -ay before mount.
Please let us know whether the solution in commet 2 or 4 resolve this problem.
Thanks. The issue is resolved with the option mentioned in Comment #2. Closing the bug.