Does format command 1. read from memory from previous scan 2. reach the device via other functional ports
I dont think so because from your output, there doesnt seem to be two paths per port, c4 is port 1 and c5 is port 2. And for example c4t6d111 device is a unique LUN path through port 1 passing by storage processor 2, this is unique and it cannot pass trough another port in my opinion.
3. reach point to point using the physical path of the native device
If the hba port is not connected, I dont think there is another OS device that can give you access through those paths ..
I also though that running format checked all the paths status, but if you have a large number of devices, and that seem to be the case from the device numbering .. you might have fallen on a special behavior of the format command
Solaris rebuilds the kernel on system "reconfiguration" boots. Part 0of this rebuild is to also rebuild the device tree, what devices Solaris sees and what devices could be on the system. I say could be, becasue if the OS and the drivers are configured properly, you can also do a devfsadm or (drvconfig -v and a disks) command to rebuild this device tree.
Format reads off of this device tree, so it will not care about what is currently happening. It will read off the last good information it was given. It does not read real-time unless you refresh the device tree.
If you were to have run a devfsadm command when a link was down, the devices associated with the link would disappear, and format would record that disappearance as you expected - the devices would be in error.
The best way to test for a healthy HBA is to use the vendor utilities or, if you have Sun's Leadville drivers installed, using the SAN Tools (cfgadm I believe). Also, you can tail the /var/adm/messages file and watch the errors from any activity to a down path.
The Powerpath driver sits lower in the OS I/O stack, so it intercepts any I/O to a dead device such as format doing a label read and redirects it to an alive path.
This is to prevent application failure if there is hard reference to a /dev entry
Even if you do a reconfig reboot the dead device should still be visable and usable as long as you have other active paths to it and have done a powermt save at some pont.
You would only see a problem in the format command if you lost ALL paths to the device, so PP had nowhere to redirect the I/O. Then it would be unable to read the label and say or something like that.
YanM1
1 Message
0
December 15th, 2007 19:00
1. read from memory from previous scan
2. reach the device via other functional ports
3. reach point to point using the physical path of the native device
I also though that running format checked all the paths status, but if you have a large number of devices, and that seem to be the case from the device numbering .. you might have fallen on a special behavior of the format command
Yan
TreyNix
7 Posts
0
June 25th, 2008 05:00
Solaris rebuilds the kernel on system "reconfiguration" boots. Part 0of this rebuild is to also rebuild the device tree, what devices Solaris sees and what devices could be on the system. I say could be, becasue if the OS and the drivers are configured properly, you can also do a devfsadm or (drvconfig -v and a disks) command to rebuild this device tree.
Format reads off of this device tree, so it will not care about what is currently happening. It will read off the last good information it was given. It does not read real-time unless you refresh the device tree.
If you were to have run a devfsadm command when a link was down, the devices associated with the link would disappear, and format would record that disappearance as you expected - the devices would be in error.
The best way to test for a healthy HBA is to use the vendor utilities or, if you have Sun's Leadville drivers installed, using the SAN Tools (cfgadm I believe). Also, you can tail the /var/adm/messages file and watch the errors from any activity to a down path.
Hope that helps.
JasonBailey
147 Posts
0
June 27th, 2008 17:00
This is to prevent application failure if there is hard reference to a /dev entry
Even if you do a reconfig reboot the dead device should still be visable and usable as long as you have other active paths to it and have done a powermt save at some pont.
You would only see a problem in the format command if you lost ALL paths to the device, so PP had nowhere to redirect the I/O. Then it would be unable to read the label and say or something like that.