Terence Tan
2010-May-12 02:10 UTC
[zfs-discuss] Unexpected rpool mirror behavior found during testing
I was having quite a bit of problems getting the rpool mirroring to work as expected. The issue I have is not about the how to create mirror rpool. I got sufficient information from the various good people in the internet to do this correctly. I am not sure the specific I have here is related to my hardware or if this is a software bug. For those who are in the know, please enlighten me. Here is a quick ran down of the machine that I used for testing before I described my test cases. a. Motherboard AMD: M4A78T-E (16G ECC, Phenom X4) b. 4 SATA drive (WD 500G RE3) c. Drive connected to onboard SATA controller using AHCI (not IDE). c. Opensolaris Build 134 Test cases =========Simple Tests ---------------- 1. Install OS build 134 into the SATA drive (no mirror). Move the SATA drive using the various on board SATA connectors and the system come up fine (with BIOS set correctly for booting). Final status: Good worked as expected 2. Mirror the OS to the another drive, and then move the the mirrors drives to any of the SATA port, and the system come up fine as long as _both_ drive was plugged into SATA connector. Final status: Good worked as expected. Fail over tests ----------------- 3. Cold start the system without one of the mirror drives by pulling it out from the hot-plug drive bay (does not matter which one). The system will fail to boot up completely. The active mirror drive GRUB does works and it was able to get into the Opensolaris splash screen. The system jpwever will remain on this splash screen with the progress status bar continue to move from side to side. Looking at the console log indicated that the boot process did not move beyond the Hostname: os01 (where os01 is the hostname of my box). I have waited for > 20 min for and the boot up will still not complete. I have verified that the boot issue is not due to incorrect BIOS bootup ordering just in case your are wondering (I have checked this numerous time). *If I break the mirror, the active disk will have no problem starting up the OS. I can reattach the mirror and then reset to use the drive *It looks like ZFS rpool is having problem doing fail-over to another drive in a mirror setup. It looks like software bug to me but it equally hard for me to believe that ZFS did not get this basic thing right. Did any one get the rpool mirror drive fail-over (in particular SATA drive) to work correctly. I sure like to know how did you get it all working! I can ditch the build-in SATA controller or the motherboard if hardware/BIOS is the issue but I want to be sure I am doing the right thing. Thanks -- This message posted from opensolaris.org
Ian Collins
2010-May-12 05:35 UTC
[zfs-discuss] Unexpected rpool mirror behavior found during testing
On 05/12/10 02:10 PM, Terence Tan wrote:> I was having quite a bit of problems getting the rpool mirroring to work as expected. > >This appears to be a known issue, see the thread "b134 - Mirrored rpool won''t boot unless both mirrors are present" and http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6923585 -- Ian.