Adeyemi Adesanya
2010-May-14 16:41 UTC
[Lustre-discuss] Is using resize2fs on the MDT a supported option?
We are are about to install a new Lustre 1.8.2 installation with ~1PB of filesystem space. We have to make a decision regarding the MDT storage and someone suggested that in the event we run out of inodes on the MDT, using resize2fs would be an option (assuming the filesystem was created on an LVM partition). Most of the documentation I have read stresses the importance of MDT planning. There is a small mention of resize2fs in the manual but it is not clear if this is a robust, reliable technique or if this would actually give us more inodes. Should we really be setting up a new Lustre installation with the intention of using resize2fs in the future? ------- Yemi
Andreas Dilger
2010-May-18 16:53 UTC
[Lustre-discuss] Is using resize2fs on the MDT a supported option?
On 2010-05-14, at 10:41, Adeyemi Adesanya wrote:> We are are about to install a new Lustre 1.8.2 installation with ~1PB > of filesystem space. We have to make a decision regarding the MDT > storage and someone suggested that in the event we run out of inodes > on the MDT, using resize2fs would be an option (assuming the > filesystem was created on an LVM partition). > > Most of the documentation I have read stresses the importance of MDT > planning. There is a small mention of resize2fs in the manual but it > is not clear if this is a robust, reliable technique or if this would > actually give us more inodes. Should we really be setting up a new > Lustre installation with the intention of using resize2fs in the future?I''ve done this on my personal system w/o trouble, but it isn''t something we test internally. I would recommend making a "dd" backup of the MDT filesystem first, and then test the resize on the backup first, running "e2fsck -f" both before and after to verify it. Then make another "dd" backup before doing the resize on the real filesystem. It is always good to have a backup of the MDT in any case. Cheers, Andreas
Hello, Would lustre 1.8 be able to work with an LFS created by 1.4? The 1.8 rolling upgrade strategy discussed on the Lustre wiki is wonderful ( http://wiki.lustre.org/index.php/Upgrading_to_a_New_Version_of_Lustre#Performing_a_Rolling_Upgrade), but it only mentions compatibility with 1.6; I''m curious if it would/should also work with an older version of Lustre like 1.4. Thank you, Adam
On Thu, 2010-06-10 at 16:04 -0400, Adam wrote:> Hello,Hi,> Would lustre 1.8 be able to work with an LFS created by 1.4?As 1.4 has been EOL for nearly (if not more) 2 years now, we have done no interoperability testing between 1.4 and 1.8. I would be very surprised if they do inter-operate.> The 1.8 > rolling upgrade strategy discussed on the Lustre wiki is wonderful ( > http://wiki.lustre.org/index.php/Upgrading_to_a_New_Version_of_Lustre#Performing_a_Rolling_Upgrade),Yes, it is. :-)> but it only mentions compatibility with 1.6;That''s right, because that''s the extent of the testing we have done with it.> I''m curious if it > would/should also work with an older version of Lustre like 1.4.Yes, a rolling upgrade from (perhaps only later versions of) 1.4 to 1.6 should work and then a rolling upgrade from 1.6 to 1.8. I would consider trying to upgrade from 1.4 to 1.8 a very risky, and likely-to-fail, proposition. b. -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 198 bytes Desc: This is a digitally signed message part Url : http://lists.lustre.org/pipermail/lustre-discuss/attachments/20100610/330aad7c/attachment.bin
Brian J. Murrell wrote:>> I''m curious if it >> would/should also work with an older version of Lustre like 1.4. >> > > Yes, a rolling upgrade from (perhaps only later versions of) 1.4 to 1.6 > should work and then a rolling upgrade from 1.6 to 1.8. > > I would consider trying to upgrade from 1.4 to 1.8 a very risky, and > likely-to-fail, proposition. >That''s what seems so great about the idea of a rolling-upgrade, if you have enough spare hardware to do it, then a 1.4 and 1.8 environment can be made available for testing at the same time. If the 1.8 set fails then no worries, flip back to the 1.4 set, then try again by downgrading the 1.8 set to 1.6. Is there risk in terms of something destructive that 1.8 might try to do to a component of the LFS? Thanks, Adam
Brian J. Murrell wrote:> On Thu, 2010-06-10 at 16:04 -0400, Adam wrote: > >> Hello, >> > > Hi, > > >> Would lustre 1.8 be able to work with an LFS created by 1.4? >> > > As 1.4 has been EOL for nearly (if not more) 2 years now, we have done > no interoperability testing between 1.4 and 1.8. I would be very > surprised if they do inter-operate. >Actually it was June 2009 -- barely one year ago. It is Lustre 1.6 that is being driven to EOL very quickly -- scheduled for this month. I believe the older Lustre 1.6 manual covered the necessary migration steps from 1.4 -- it was NOT as painless as moving from 1.6 to 1.8. The process might also be in the mail archives.>> The 1.8 >> rolling upgrade strategy discussed on the Lustre wiki is wonderful ( >> http://wiki.lustre.org/index.php/Upgrading_to_a_New_Version_of_Lustre#Performing_a_Rolling_Upgrade), >> > > Yes, it is. :-) > > >> but it only mentions compatibility with 1.6; >> > > That''s right, because that''s the extent of the testing we have done with > it. > > >> I''m curious if it >> would/should also work with an older version of Lustre like 1.4. >> > > Yes, a rolling upgrade from (perhaps only later versions of) 1.4 to 1.6 > should work and then a rolling upgrade from 1.6 to 1.8. > > I would consider trying to upgrade from 1.4 to 1.8 a very risky, and > likely-to-fail, proposition. > > b. > >
On Thu, 2010-06-10 at 14:29 -0600, Kevin Van Maren wrote:> > Actually it was June 2009 -- barely one year ago.Ahhh. Yes, you are right. Maybe consider it wishful thinking that it (and it''s lconf configuration tools) was 2 years EOL. :-)> It is Lustre 1.6 that > is being driven to EOL > very quickly -- scheduled for this month.Indeed.> I believe the older Lustre 1.6 manual covered the necessary migration > steps from 1.4 -- it was NOT as > painless as moving from 1.6 to 1.8.Right, which only goes to underscore my warnings about trying to jump from 1.4 to 1.8. Thanx for keeping me honest Kevin. b. -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 198 bytes Desc: This is a digitally signed message part Url : http://lists.lustre.org/pipermail/lustre-discuss/attachments/20100610/28bb178c/attachment.bin
On Thu, 2010-06-10 at 16:21 -0400, Adam wrote:> > That''s what seems so great about the idea of a rolling-upgrade, if you > have enough spare hardware to do it, then a 1.4 and 1.8 environment can > be made available for testing at the same time. If the 1.8 set fails > then no worries, flip back to the 1.4 set, then try again by downgrading > the 1.8 set to 1.6.Well, if you think you have enough hardware and want to do some testing, and think that you can fairly represent your production environment, by all means go ahead and enough testing that you are satisfied with putting your production data through such an upgrade process. I stand by my warnings however.> Is there risk in terms of something destructive that 1.8 might try to do > to a component of the LFS?There is always risk when doing something untested. b. -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 198 bytes Desc: This is a digitally signed message part Url : http://lists.lustre.org/pipermail/lustre-discuss/attachments/20100610/677e91b3/attachment.bin
On 2010-06-10, at 14:21, Adam wrote:>>> I''m curious if it >>> would/should also work with an older version of Lustre like 1.4. >> >> Yes, a rolling upgrade from (perhaps only later versions of) 1.4 to 1.6 >> should work and then a rolling upgrade from 1.6 to 1.8. >> >> I would consider trying to upgrade from 1.4 to 1.8 a very risky, and >> likely-to-fail, proposition. > > That''s what seems so great about the idea of a rolling-upgrade, if you > have enough spare hardware to do it, then a 1.4 and 1.8 environment can > be made available for testing at the same time. If the 1.8 set fails > then no worries, flip back to the 1.4 set, then try again by downgrading > the 1.8 set to 1.6.Note that the network protocols between 1.4 and 1.8 are not compatible, so you cannot do a rolling upgrade. The on-disk data format (excluding the configuration files) is compatible, so in theory it should be possible to run the 1.4->1.6 upgrade process, mount the OST/MDT once and do a quick client test, and then go immediately to 1.8.> Is there risk in terms of something destructive that 1.8 might try to do > to a component of the LFS?If you go through 1.6 (even if only briefly) it should update the on-disk configuration appropriately, and then the upgrade to 1.8 should work cleanly, and (in theory) it should still be possible to downgrade to 1.4 if needed. That said, I''d recommend trying this first and/or making a backup (which are always good to have). Cheers, Andreas -- Andreas Dilger Lustre Technical Lead Oracle Corporation Canada Inc.
Thank you Andreas, Brian, and Kevin. I think we''ll go with a 1.4->1.6 then 1.6->1.8 rolling upgrade plan if we decide to upgrade the older systems. Cheers, Adam Andreas Dilger wrote:> On 2010-06-10, at 14:21, Adam wrote: > >>>> I''m curious if it >>>> would/should also work with an older version of Lustre like 1.4. >>>> >>> Yes, a rolling upgrade from (perhaps only later versions of) 1.4 to 1.6 >>> should work and then a rolling upgrade from 1.6 to 1.8. >>> >>> I would consider trying to upgrade from 1.4 to 1.8 a very risky, and >>> likely-to-fail, proposition. >>> >> That''s what seems so great about the idea of a rolling-upgrade, if you >> have enough spare hardware to do it, then a 1.4 and 1.8 environment can >> be made available for testing at the same time. If the 1.8 set fails >> then no worries, flip back to the 1.4 set, then try again by downgrading >> the 1.8 set to 1.6. >> > > Note that the network protocols between 1.4 and 1.8 are not compatible, so you cannot do a rolling upgrade. The on-disk data format (excluding the configuration files) is compatible, so in theory it should be possible to run the 1.4->1.6 upgrade process, mount the OST/MDT once and do a quick client test, and then go immediately to 1.8. > > >> Is there risk in terms of something destructive that 1.8 might try to do >> to a component of the LFS? >> > > If you go through 1.6 (even if only briefly) it should update the on-disk configuration appropriately, and then the upgrade to 1.8 should work cleanly, and (in theory) it should still be possible to downgrade to 1.4 if needed. That said, I''d recommend trying this first and/or making a backup (which are always good to have). > > Cheers, Andreas > -- > Andreas Dilger > Lustre Technical Lead > Oracle Corporation Canada Inc. > >-- Adam Munro System Administrator | SHARCNET | http://www.sharcnet.ca Compute Canada | http://www.computecanada.org 519-888-4567 x36453