All disks are not showed at the time of partman/early-command execution

Bug #1413687 reported by bugproxy
8
This bug affects 1 person
Affects Status Importance Assigned to Milestone
debian-installer (Ubuntu)
Invalid
Undecided
Taco Screen team

Bug Description

== Comment: #0 - Haren Myneni <email address hidden> - 2015-01-07 21:34:09 ==
---Problem Description---
Using preseed partman/early-command to select the disk based on user selection with location ID. But the 'iprconfig show-config' command could not see all expected disks. Looking at the dmesg at the time of this command execution, all disks are not probed. On our system, only HDD disks are probed and SDD disks are not available.

After some time, the dmesg shows the kernel found other disks. we are seeing the same behavior even with kernel parameter 'scsi_mod.scan=sync'

So with this issue, user can not select some disks with automated installation.

Contact Information = <email address hidden>

---uname output---
14.10 release

Machine Type = Any power system with more disks

---Debugger---
A debugger is not configured

---Steps to Reproduce---
 Select system which has more disks. Select last disk in automated installation

Install method: NFS

Install disk info: SDD

Install ISO Information: Ubuntu Utopia

== Comment: #3 - Haren Myneni <email address hidden> - 2015-01-07 21:42:54 ==
preseed file:

d-i partman/early_command string \
 mkdir /tmp/updates; cd /tmp/updates; \
 wget -o /tmp/wget_log http://9.18.10.27/install/ppc/Ubuntu14_1410/preload_updates.tar.gz; \
 tar zxvf preload_updates.tar.gz; \
 /bin/sh /tmp/updates/preload/hooks/preseed/pre.sh; \
 DISKA=; \
 debconf-set partman-auto/disk "$DISKA";

We execute ipcronfig comamnd as part of pre.sh. As we see in the log file, only disks sda-sdf are listed.

(In reply to comment #6)
> (In reply to comment #5)
...
> But it is installer issue - means all disks should be probed first, then the
> installer should start downloading preseed file and execute
> partman/early-command. As I mentioned before, 'scsi_mod.scan=sync' dis not
> solve the issue. Please note that even now we can install the default disk
> sda since it is already probed, but we should be able to select any disk for
> install which is requirement for preload in manufacturing.

== Comment: #8 - Haren Myneni <email address hidden> - 2015-01-16 11:40:32 ==
It is not the issue of not recognizing some disks. As you see in the final dmesg output, all disks are probed and recognized. But the issue here is only few are probed before the installer started. So the problem is why the installer started before probing all storage devices.

== Comment: #10 - Haren Myneni <email address hidden> - 2015-01-21 15:24:51 ==
As mentioned in the bug description, first dmesg was collected at the time of partman/early-command execution. It shows only sda-sdf disks are probed. We were attempting to install on sdh and set this disk as install disk with partman/early-command. Since the iprconfig command could not list sdh (since it is not probed at this time), the installer could not set this disk and then failed.

We collected the final dmesg later when we noticed the the failure. The final dmesg shows all expected disks sda- sdl are probed and we verified with iprconfig command.

So this issue could be timing of starting the installer before all devices are probed. We should expect that the kernel boot should be completed and then the installer should start.

Revision history for this message
bugproxy (bugproxy) wrote : dmesg at the time of partman/early-command execution

Default Comment by Bridge

tags: added: architecture-ppc64le bugnameltc-120044 severity-high targetmilestone-inin---
Revision history for this message
bugproxy (bugproxy) wrote : final dmesg output

Default Comment by Bridge

Revision history for this message
bugproxy (bugproxy) wrote : select disk based on location ID log

Default Comment by Bridge

Dave Heller (hellerda)
affects: ubuntu → debian-installer (Ubuntu)
Revision history for this message
bugproxy (bugproxy) wrote : Comment bridged from LTC Bugzilla

------- Comment From <email address hidden> 2015-01-31 02:00 EDT-------
I believe the install worked with long delay, but asked Lucas (manufacturing) to verify this test. He is still waiting for the system to test.

Revision history for this message
Mauricio Faria de Oliveira (mfo) wrote :

Canonical,

It seems the problem here is the delay some disks in the IPR controller encounter to show up.

From dmesg there's 2 IPR controllers, and the kernel module initializes the controllers sequentially, so the disks from the 2nd controller take even more time to show up.

I'm not sure 'udevadm settle' would wait for a SCSI scan to finish, but maybe only for the processing of the udev events of the disks/partitions being added?

From some testing, it seems that 'modprobe ipr; udevadm settle' ends after all disks show up, but that seems to be mostly because modprobe takes a long time to finish (thus udevadm settle would be more of a guarantee).

Is it possible to implement something like that (ensure all disks finished detection) in the hw-detect stage, before finishing and allowing partman to start -- so the partman/early-command executes only when all disks are detected/SCSI scans finished?

Luciano Chavez (lnx1138)
Changed in debian-installer (Ubuntu):
assignee: nobody → Taco Screen team (taco-screen-team)
bugproxy (bugproxy)
tags: added: targetmilestone-inin1604
removed: targetmilestone-inin---
Revision history for this message
bugproxy (bugproxy) wrote : dmesg at the time of partman/early-command execution

Default Comment by Bridge

Revision history for this message
Luciano Chavez (lnx1138) wrote :

Closing due to lack of interest.

Changed in debian-installer (Ubuntu):
status: New → Invalid
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.