Re: Helping with Softraid testing
On 19/11/2007, Stuart Henderson [EMAIL PROTECTED] wrote: On 2007/11/19 05:04, Siju George wrote: One all the features in your mind has been implemented to softraid will it make RAIDFRAME redundant? This is all future stuff, I think I'm right in saying that what's needed first and foremost is test reports for the less popular machine architectures. I can test on sparc64 if you like, but I am unsure what the driver is capable of at this stage. -- Best Regards Edd --- http://students.dec.bournemouth.ac.uk/ebarrett
Re: Helping with Softraid testing
On 2007/11/19 10:27, Edd Barrett wrote: On 19/11/2007, Stuart Henderson [EMAIL PROTECTED] wrote: On 2007/11/19 05:04, Siju George wrote: One all the features in your mind has been implemented to softraid will it make RAIDFRAME redundant? This is all future stuff, I think I'm right in saying that what's needed first and foremost is test reports for the less popular machine architectures. I can test on sparc64 if you like, but I am unsure what the driver is capable of at this stage. RAID1, autoconfigured at boot (before root is mounted). Currently, if a drive fails, you need to dd to clone the working drive, a bit like ccd(4) mirrors in that respect. Do give it a try if you haven't, if you don't have free disk space you could always borrow a bit from /tmp or swap (or run under qemu if need be). It's pretty slick.
Re: Helping with Softraid testing
On Mon, Nov 19, 2007 at 05:04:53AM +0530, Siju George wrote: On Nov 18, 2007 7:46 PM, Marco Peereboom [EMAIL PROTECTED] wrote: On Sun, Nov 18, 2007 at 04:32:58AM +0530, Siju George wrote: Thank you so much Most of your questions are around rebuild or derivatives. This does not exist yet. My current push is to get softraid working on all arches so that it could get re-enabled. The rebuild stuff comes after that. Thank you so much Marco for the detailed reply. Just one more quick question please :-) One all the features in your mind has been implemented to softraid will it make RAIDFRAME redundant? RAIDFRAME for all its quirks and uglies still does way more than softraid. I'll be one happy man if we get to a stage that we can replace it. Besides this we'll have to make sure we are not screwing users of RAIDFRAME out there. Though I love and use RAIDFRAME and have overcome the initial qualms of rebuilding the kernel from source after doing it a number of times now I think it will still be great if Software RAID can be implemented without having to re compile the kernel. I know I cannot escape recompiling the kernel because it is necessary for updates. But as far as possible I would like to stay away from it on production machines :-) As far as I know I have one bug remaining that needs fixing. Sometimes at shutdown time one of the metadata updates does not make it onto a disk and on the subsequent reboot the disk is not auto-assembled like it should. The other thing I really need is to make dead sure that there are no booting issues on ANY platform. Once I am passed these two issues I can move forward into the realm that is rebuilds and derivatives. Thanks a million for all the detailed answers once again :-))) Kind Regards Siju
Re: Helping with Softraid testing
On Mon, Nov 19, 2007 at 12:54:04AM +, Stuart Henderson wrote: Marco, what arch are you missing reports for now? That is the best question to ask :-) The arches that I want more testing on are: alpha armish hp300 hppa landisk luna88k ma68k mvme68k mvme88k sgi sparc vax zaurus Preferable I'd like to see the testing using real disks. Yes USB is important but it is also super slow and has different issues than actual disk. I have done a lot of testing using USB at this time and need more love on actual disk. Spare me the qemu/i386/amd64 ones I have tons and tons of those. Thanks!
Re: Helping with Softraid testing
On 19/11/2007, Marco Peereboom [EMAIL PROTECTED] wrote: sparc Preferable I'd like to see the testing using real disks. Do the slices need to be on different disks to make useful testing? I ask because my sparc(64) box has a single FCAL (Fibre Channel) disk, and these things are not easy to come by. I have been meaning to buy one, but OpenCON has cleaned my wallet out for now :P -- Best Regards Edd --- http://students.dec.bournemouth.ac.uk/ebarrett
Re: Helping with Softraid testing
Nah, single disk is fine. On Mon, Nov 19, 2007 at 01:12:29PM +, Edd Barrett wrote: On 19/11/2007, Marco Peereboom [EMAIL PROTECTED] wrote: sparc Preferable I'd like to see the testing using real disks. Do the slices need to be on different disks to make useful testing? I ask because my sparc(64) box has a single FCAL (Fibre Channel) disk, and these things are not easy to come by. I have been meaning to buy one, but OpenCON has cleaned my wallet out for now :P -- Best Regards Edd --- http://students.dec.bournemouth.ac.uk/ebarrett
Re: Helping with Softraid testing
On Nov 19, 2007 5:12 AM, Ray Percival [EMAIL PROTECTED] wrote: On Nov 18, 2007, at 3:34 PM, Siju George wrote: snip I know I cannot escape recompiling the kernel because it is necessary for updates. But as far as possible I would like to stay away from it on production machines :-) That's what releases are for. I haven't tried releases yet, thanks the next days I will do that :-) Kind regards Siju
Re: Helping with Softraid testing
On Sun, Nov 18, 2007 at 04:32:58AM +0530, Siju George wrote: On Nov 17, 2007 12:04 AM, Marco Peereboom [EMAIL PROTECTED] wrote: I'll take this as the documentation isn't good enough. Can you point me to the area that isn't clear? Thanks for the offer Marco. I currently use RAIDFRAME for Raid 1 A few doubts that remain after I have read the man page is. 1) Is it possible to run / from a Softraid volume? Yes you can. The real issue is where do you load your kernel from. Also all the recovery pieces are not in place and I would therefore not use it for this just yet. 2) Currently the RAID 1 support does not have the ability to recover a failed chunk. Does this mean If I have two chunks on two different hard disks and if one disk fails I need to a) copy the data from the working disk to some where b) get a new disk and re create RAID 1 as I did in the begining and copy back the data to the RAID 1 volume? The procedure, albeit ugly, at this point is to boot into a single user environment and dd the correct disk on top of the failed disk. Then use bioctl to assemble the disks with force (-C force). This obviously sucks and I will make this much more transparent in the future. 3) The RAID 1 discipline does not initialize the mirror upon creation. So what should be done manually to initialize the mirror? Nothing. Initializing a RAID set is mostly silly. The only reason to do it is to read all sectors and when one of them is bad it can be remapped either by the drive or by some higher layer. This is nice of course but surely not critical. See the data you want to read has been written before. If it hasn't you don't know or care what is in it. I will eventually do some sort of initialize the RAID set but that'll be an extension of to rebuild. They are virtually the same function and should be treated as such. Also if the system suffers a power loss will the system boot up only after parity is checked? Right now it won't do that yet. Again that is a function of the rebuild that does not exist yet. Thank you so much Most of your questions are around rebuild or derivatives. This does not exist yet. My current push is to get softraid working on all arches so that it could get re-enabled. The rebuild stuff comes after that.
Re: Helping with Softraid testing
On Nov 18, 2007 7:46 PM, Marco Peereboom [EMAIL PROTECTED] wrote: On Sun, Nov 18, 2007 at 04:32:58AM +0530, Siju George wrote: Thank you so much Most of your questions are around rebuild or derivatives. This does not exist yet. My current push is to get softraid working on all arches so that it could get re-enabled. The rebuild stuff comes after that. Thank you so much Marco for the detailed reply. Just one more quick question please :-) One all the features in your mind has been implemented to softraid will it make RAIDFRAME redundant? Though I love and use RAIDFRAME and have overcome the initial qualms of rebuilding the kernel from source after doing it a number of times now I think it will still be great if Software RAID can be implemented without having to re compile the kernel. I know I cannot escape recompiling the kernel because it is necessary for updates. But as far as possible I would like to stay away from it on production machines :-) Thanks a million for all the detailed answers once again :-))) Kind Regards Siju
Re: Helping with Softraid testing
On Nov 18, 2007, at 3:34 PM, Siju George wrote: snip I know I cannot escape recompiling the kernel because it is necessary for updates. But as far as possible I would like to stay away from it on production machines :-) That's what releases are for. Thanks a million for all the detailed answers once again :-))) Kind Regards Siju
Re: Helping with Softraid testing
On 2007/11/19 05:04, Siju George wrote: One all the features in your mind has been implemented to softraid will it make RAIDFRAME redundant? This is all future stuff, I think I'm right in saying that what's needed first and foremost is test reports for the less popular machine architectures. The time for other disciplines and rebuild/scrubbing, is when the basics are well-tested, any problems are fixed, it's back in GENERIC... then it's much easier for people to test anything which gets added. Marco, what arch are you missing reports for now?
Re: Helping with Softraid testing
On Nov 17, 2007 12:04 AM, Marco Peereboom [EMAIL PROTECTED] wrote: I'll take this as the documentation isn't good enough. Can you point me to the area that isn't clear? Thanks for the offer Marco. I currently use RAIDFRAME for Raid 1 A few doubts that remain after I have read the man page is. 1) Is it possible to run / from a Softraid volume? 2) Currently the RAID 1 support does not have the ability to recover a failed chunk. Does this mean If I have two chunks on two different hard disks and if one disk fails I need to a) copy the data from the working disk to some where b) get a new disk and re create RAID 1 as I did in the begining and copy back the data to the RAID 1 volume? 3) The RAID 1 discipline does not initialize the mirror upon creation. So what should be done manually to initialize the mirror? Also if the system suffers a power loss will the system boot up only after parity is checked? Thank you so much Kind Regards Siju
Re: Helping with Softraid testing
I'll take this as the documentation isn't good enough. Can you point me to the area that isn't clear? On Fri, Nov 16, 2007 at 11:29:20AM -0700, Chris Cameron wrote: I'm in a good position to test Softraid on an AMD and an UltraSPARC, however I've realized I don't know a lot about it (what -exactly- it's working to accomplish, and commands to use). Is there an overview of Softraid to get me started so I can be of some use? Chris
Re: Helping with Softraid testing
On 11/16/07, Chris Cameron [EMAIL PROTECTED] wrote: I'm in a good position to test Softraid on an AMD and an UltraSPARC, however I've realized I don't know a lot about it (what -exactly- it's working to accomplish, and commands to use). Is there an overview of Softraid to get me started so I can be of some use? RAID lets you cat disks together in a variety of ways: for redundancy, for extending the sizes, c. softraid is a new feature just released in 4.2 that supports on of these ways: 'mirroring'. Mirroring writes every piece of data to multiple disks, so that if any of them fail the data is not lost, and the disk can be replaced more or less transparently. You'll need to get a bunch of harddrives of different and the same sizes, and plug them all in. Then follow the instructions here: http://www.openbsd.org/cgi-bin/man.cgi?query=softraidsektion=4 It would help to read this too: http://www.openbsd.org/cgi-bin/man.cgi?query=bioctlsektion=8 -Nick
Helping with Softraid testing
I'm in a good position to test Softraid on an AMD and an UltraSPARC, however I've realized I don't know a lot about it (what -exactly- it's working to accomplish, and commands to use). Is there an overview of Softraid to get me started so I can be of some use? Chris
Re: Helping with Softraid testing
I thought the manpage was just covering things that worked well, and in the code itself were things waiting to be tested better. It shows a 3 chunk raid 1 setup, but doesn't mention anything about hot standby. I'm not aware of 3 disk RAID 1 otherwise. Also, for some reason (I think past misc@ posts) I was under the impression that this would be similar to Vinum. From what I'm hearing back it's actually a RAIDFrame replacement. The manpage doesn't really go over it's final goal. Testing related: I saw your message on the 15th asking for help. I plan to run that through a SPARC machine, but I'm not sure if there are different ways to poke at the new code. Will different underlying hardware (besides architecture) make a difference, or is this a layer above that? Chris Marco Peereboom wrote: I'll take this as the documentation isn't good enough. Can you point me to the area that isn't clear? On Fri, Nov 16, 2007 at 11:29:20AM -0700, Chris Cameron wrote: I'm in a good position to test Softraid on an AMD and an UltraSPARC, however I've realized I don't know a lot about it (what -exactly- it's working to accomplish, and commands to use). Is there an overview of Softraid to get me started so I can be of some use? Chris
Re: Helping with Softraid testing
On Fri, Nov 16, 2007 at 01:26:32PM -0700, Chris Cameron wrote: I thought the manpage was just covering things that worked well, and in the code itself were things waiting to be tested better. It shows a 3 chunk raid 1 setup, but doesn't mention anything about hot standby. I'm not aware of 3 disk RAID 1 otherwise. Also, for some reason (I think past misc@ posts) I was under the impression that this would be similar to Vinum. From what I'm hearing back it's actually a RAIDFrame replacement. The manpage doesn't really go over it's final goal. Current goal is to get to a functional raid 1 and then we'll move on to the rest. Testing related: I saw your message on the 15th asking for help. I plan to run that through a SPARC machine, but I'm not sure if there are different ways to poke at the new code. Will different underlying hardware (besides architecture) make a difference, or is this a layer above that? It does a little. Some drivers are more reliable than other under failure conditions. The crappy part is that there is only so much we can do under certain circumstances. I do care to see reports though. Chris Marco Peereboom wrote: I'll take this as the documentation isn't good enough. Can you point me to the area that isn't clear? On Fri, Nov 16, 2007 at 11:29:20AM -0700, Chris Cameron wrote: I'm in a good position to test Softraid on an AMD and an UltraSPARC, however I've realized I don't know a lot about it (what -exactly- it's working to accomplish, and commands to use). Is there an overview of Softraid to get me started so I can be of some use? Chris