Re: gvinum gjournal
On Wed, Jan 14, 2009 at 04:23:30PM -0500, Brian McCann wrote: Hi all. I'm cross-posting this since I figure I'll have better luck finding someone who's done this before... I'm building a system that has 4 1.5TB Seagate SATA drives in it. I've setup gvinum and made mirrors for my OS partitions, and a raid5 plex for a big data partition. I'm trying to get gjournal to run on the raid5 volume...but it's doing stuff that isn't expected. First, here's my gvinum config for the array: ---snip--- drive e0 device /dev/ad8s1g drive e1 device /dev/ad10s1g drive e2 device /dev/ad12s1g drive e3 device /dev/ad14s1g volume array1 plex org raid5 128k sd drive e0 sd drive e1 sd drive e2 sd drive e3 ---/snip--- Now...according to the handbook. the volume it creates is essentially a disk drive. So...I run the following gjournal commands to make the journal, and here's what I get: ---snip--- # gjournal label /dev/gvinum/array1 GEOM_JOURNAL: Journal 4267655417: gvinum/plex/array1.p0 contains data. GEOM_JOURNAL: Journal 4267655417: gvinum/plex/array1.p0 contains journal. GEOM_JOURNAL: Journal gvinum/plex/array1.p0 clean. GEOM_JOURNAL: BIO_FLUSH not supported by gvinum/plex/array1.p0. # gjournal list Geom name: gjournal 4267655417 ID: 4267655417 Providers: 1. Name: gvinum/plex/array1.p0.journal Mediasize: 4477282549248 (4.1T) Sectorsize: 512 Mode: r0w0e0 Consumers: 1. Name: gvinum/plex/array1.p0 Mediasize: 4478356291584 (4.1T) Sectorsize: 512 Mode: r1w1e1 Jend: 4478356291072 Jstart: 4477282549248 Role: Data,Journal --/snip--- So...why is it even touching the plex p0? I figured it would, just like on a disk, if I gave it da0, create da0.journal. Moving on, if I try to newfs the journal, which is now gvinum/plex/array1.p0.journal, I get: Hi, It think that it touches it because the .p0 contains the gjournal metadata in the same way that the volume does, so gjournal attaches to that before the volume. One problem is that gjournal attaches to the wrong provider, but it's also silly that the provider is exposed in the first place. A fix for this is in a newer version of gvinum (as the plex is not exposed) if you're willing to try. -- Ulf Lilleengen ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: gvinum gjournal
On Thu, Jan 15, 2009 at 4:33 AM, Ulf Lilleengen ulf.lilleen...@gmail.com wrote: Hi, It think that it touches it because the .p0 contains the gjournal metadata in the same way that the volume does, so gjournal attaches to that before the volume. One problem is that gjournal attaches to the wrong provider, but it's also silly that the provider is exposed in the first place. A fix for this is in a newer version of gvinum (as the plex is not exposed) if you're willing to try. -- Ulf Lilleengen At this point, I'm willing to try anything, but preferably something that's stable since this will be done to at least 15 identical devices and sent out to various places, so I won't have physical access to the machines if something were to go wrong. I looked into graid3 and booting off of a DOM/Flash card, but since I have 4 drives, that won't work since graid3 requires 2N+1 drives. The stuff I've found on graid5 seems to say that it's all still really experimental and has some bugs in it still. :( That said...if you've got it, I'll try it. :) Thanks, --Brian -- _-=-_-=-_-=-_-=-_-=-_-=-_-=-_-=-_-=-_-=-_-=-_ Brian McCann I don't have to take this abuse from you -- I've got hundreds of people waiting to abuse me. -- Bill Murray, Ghostbusters ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: gvinum gjournal
On tor, jan 15, 2009 at 07:14:25am -0500, Brian McCann wrote: On Thu, Jan 15, 2009 at 4:33 AM, Ulf Lilleengen ulf.lilleen...@gmail.com wrote: Hi, It think that it touches it because the .p0 contains the gjournal metadata in the same way that the volume does, so gjournal attaches to that before the volume. One problem is that gjournal attaches to the wrong provider, but it's also silly that the provider is exposed in the first place. A fix for this is in a newer version of gvinum (as the plex is not exposed) if you're willing to try. -- Ulf Lilleengen At this point, I'm willing to try anything, but preferably something that's stable since this will be done to at least 15 identical devices and sent out to various places, so I won't have physical access to the machines if something were to go wrong. I looked into graid3 and booting off of a DOM/Flash card, but since I have 4 drives, that won't work since graid3 requires 2N+1 drives. The stuff I've found on graid5 seems to say that it's all still really experimental and has some bugs in it still. :( Well, if you have the choice between and graid5 and gvinum I'd might go with graid5. It's not included in head(yet) but from what I've read on the mailinglist it seems to work (also included in FreeNAS). But I can't really say anything as I've not tried it much. If you want to try the gvinum version in progress (quite stable, although a few knits for advanced usage) can be found here: http://svn.freebsd.org/base/projects/gvinum Should work on both HEAD and RELENG_7. -- Ulf Lilleengen ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: gvinum gjournal
On Thu, Jan 15, 2009 at 02:22:13PM +0100, Ivan Voras wrote: Ulf Lilleengen wrote: On Wed, Jan 14, 2009 at 04:23:30PM -0500, Brian McCann wrote: Hi all. I'm cross-posting this since I figure I'll have better luck finding someone who's done this before... I'm building a system that has 4 1.5TB Seagate SATA drives in it. I've setup gvinum and made mirrors for my OS partitions, and a raid5 plex for a big data partition. I'm trying to get gjournal to run on the raid5 volume...but it's doing stuff that isn't expected. First, here's my gvinum config for the array: ---snip--- drive e0 device /dev/ad8s1g drive e1 device /dev/ad10s1g drive e2 device /dev/ad12s1g drive e3 device /dev/ad14s1g volume array1 plex org raid5 128k sd drive e0 sd drive e1 sd drive e2 sd drive e3 ---/snip--- Now...according to the handbook. the volume it creates is essentially a disk drive. So...I run the following gjournal commands to make the journal, and here's what I get: ---snip--- # gjournal label /dev/gvinum/array1 GEOM_JOURNAL: Journal 4267655417: gvinum/plex/array1.p0 contains data. GEOM_JOURNAL: Journal 4267655417: gvinum/plex/array1.p0 contains journal. GEOM_JOURNAL: Journal gvinum/plex/array1.p0 clean. GEOM_JOURNAL: BIO_FLUSH not supported by gvinum/plex/array1.p0. # gjournal list Geom name: gjournal 4267655417 ID: 4267655417 Providers: 1. Name: gvinum/plex/array1.p0.journal Mediasize: 4477282549248 (4.1T) Sectorsize: 512 Mode: r0w0e0 Consumers: 1. Name: gvinum/plex/array1.p0 Mediasize: 4478356291584 (4.1T) Sectorsize: 512 Mode: r1w1e1 Jend: 4478356291072 Jstart: 4477282549248 Role: Data,Journal --/snip--- So...why is it even touching the plex p0? I figured it would, just like on a disk, if I gave it da0, create da0.journal. Moving on, if I try to newfs the journal, which is now gvinum/plex/array1.p0.journal, I get: Hi, It think that it touches it because the .p0 contains the gjournal metadata in the same way that the volume does, so gjournal attaches to that before the volume. One problem is that gjournal attaches to the wrong provider, but it's also silly that the provider is exposed in the first place. A fix for this is in a newer version of gvinum (as the plex is not exposed) if you're willing to try. A simpler fix is to use the -h - hardcode provider name switch to the gjournal label command (see the man page). Oh, nice feature. I recommend this then :) -- Ulf Lilleengen ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
gvinum gjournal
Hi all. I'm cross-posting this since I figure I'll have better luck finding someone who's done this before... I'm building a system that has 4 1.5TB Seagate SATA drives in it. I've setup gvinum and made mirrors for my OS partitions, and a raid5 plex for a big data partition. I'm trying to get gjournal to run on the raid5 volume...but it's doing stuff that isn't expected. First, here's my gvinum config for the array: ---snip--- drive e0 device /dev/ad8s1g drive e1 device /dev/ad10s1g drive e2 device /dev/ad12s1g drive e3 device /dev/ad14s1g volume array1 plex org raid5 128k sd drive e0 sd drive e1 sd drive e2 sd drive e3 ---/snip--- Now...according to the handbook. the volume it creates is essentially a disk drive. So...I run the following gjournal commands to make the journal, and here's what I get: ---snip--- # gjournal label /dev/gvinum/array1 GEOM_JOURNAL: Journal 4267655417: gvinum/plex/array1.p0 contains data. GEOM_JOURNAL: Journal 4267655417: gvinum/plex/array1.p0 contains journal. GEOM_JOURNAL: Journal gvinum/plex/array1.p0 clean. GEOM_JOURNAL: BIO_FLUSH not supported by gvinum/plex/array1.p0. # gjournal list Geom name: gjournal 4267655417 ID: 4267655417 Providers: 1. Name: gvinum/plex/array1.p0.journal Mediasize: 4477282549248 (4.1T) Sectorsize: 512 Mode: r0w0e0 Consumers: 1. Name: gvinum/plex/array1.p0 Mediasize: 4478356291584 (4.1T) Sectorsize: 512 Mode: r1w1e1 Jend: 4478356291072 Jstart: 4477282549248 Role: Data,Journal --/snip--- So...why is it even touching the plex p0? I figured it would, just like on a disk, if I gave it da0, create da0.journal. Moving on, if I try to newfs the journal, which is now gvinum/plex/array1.p0.journal, I get: ---snip--- # newfs -J /dev/gvinum/plex/array1.p0.journal /dev/gvinum/plex/array1.p0.journal: 4269869.4MB (8744692476 sectors) block size 16384, fragment size 2048 using 23236 cylinder groups of 183.77MB, 11761 blks, 23552 inodes. newfs: can't read old UFS1 superblock: end of file from block device: No such file or directory ---/snip--- Followed by a panic and reboot: ---snip--- Fatal trap 12: page fault while in kernel mode cpuid = 0; apic id = 00 fault virtual address = 0x0 fault code = supervisor read, page not present instruction pointer = 0x20:0xc0d8d440 stack pointer = 0x28:0xd4e25c44 frame pointer = 0x28:0xd4e25cf4 code segment= base 0x0, limit 0xf, type 0x1b = DPL 0, pres 1, def32 1, gran 1 processor eflags= interrupt enabled, resume, IOPL = 0 current process = 47 (gv_p array1.p0) trap number = 12 panic: page fault cpuid = 0 Uptime: 14m38s Cannot dump. No dump device defined. Automatic reboot in 15 seconds - press a key on the console to abort ---/snip--- Next...I destroyed/cleared/stoped/etc the journal to start fresh, made a new one...it created the same thing (gvinum/plex/array1.p0.journal)...I then rebooted, loaded the gjournal module, and I now see gvinum/array1.journal as the provider, and the provider inside plex is gone. I then run my newfs (newfs -J /dev/gvinum/array1.journal) , and I get ---snip--- Fatal trap 12: page fault while in kernel mode cpuid = 0; apic id = 00 fault virtual address = 0x1c fault code = supervisor read, page not present instruction pointer = 0x20:0xc0d8eec5 stack pointer = 0x28:0xd4e2ecbc frame pointer = 0x28:0xd4e2ecf4 code segment= base 0x0, limit 0xf, type 0x1b = DPL 0, pres 1, def32 1, gran 1 processor eflags= interrupt enabled, resume, IOPL = 0 current process = 50 (gv_v array1) trap number = 12 panic: page fault cpuid = 0 Uptime: 8m18s Cannot dump. No dump device defined. Automatic reboot in 15 seconds - press a key on the console to abort ---/snip--- Does anyone have any ideas here? I assumed gjournal would play nice with any file system. But clearly not. After I clear the journal off of /dev/gvinum/array1, I can do a newfs on it (/dev/gvinum/array1) without the journal fine...so that tests that the RAID5 is ok. Anyone havve any ideas? Thanks! --Brian -- _-=-_-=-_-=-_-=-_-=-_-=-_-=-_-=-_-=-_-=-_-=-_ Brian McCann I don't have to take this abuse from you -- I've got hundreds of people waiting to abuse me. -- Bill Murray, Ghostbusters ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
Re: gvinum gjournal
On Wed, Jan 14, 2009 at 4:23 PM, Brian McCann bjmcc...@gmail.com wrote: Does anyone have any ideas here? I assumed gjournal would play nice with any file system. But clearly not. After I clear the journal off of /dev/gvinum/array1, I can do a newfs on it (/dev/gvinum/array1) without the journal fine...so that tests that the RAID5 is ok. Anyone havve any ideas? Thanks! --Brian I also just got the idea to try turning off write caching on the ata controller...no help. Just thought I'd drop that out there if that clues in on something. --Brian -- _-=-_-=-_-=-_-=-_-=-_-=-_-=-_-=-_-=-_-=-_-=-_ Brian McCann I don't have to take this abuse from you -- I've got hundreds of people waiting to abuse me. -- Bill Murray, Ghostbusters ___ freebsd-questions@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to freebsd-questions-unsubscr...@freebsd.org
