I give up
Greetings Guys & Dolls;
I give up, I'm whupped.
When I installed FC6, I let it use LVM2 to handle the disks because I was tired of fighting with the most broken, nasty, evil disk partitioner ever dreamed up, disk druid. I personally think that abomination was written by an M$ plant, specifically to poison users against linux.
Now the 64k$ question: While running with LVM2 managed disks, is it possible to run without dm_mod, the device-mapper? If so, please tell me how to achieve this.
I rebooted last night, from 2.6.21-rc6 to 2.6.21-rc6 and for some reason device-mapper went from #252 (cat /proc/devices) to #254, and has also been seen at #253 in addition to the temporary address of #238 where it was before that patch was reverted for 2.6.21-rc6
Anytime this changes, the device numbers in a stat report change, and tar thinks its all new & restarts everything on a 45GB system from a level 0. My vtape setup only has room for 8.5GB per vtape, so the catchup time is many days. I can speed that up by temporarily making the vtape bigger in my amanda.conf, but now the disk itself that the vtapes are on is at 96% and will not tolerate another such catchup run.
So I'm pleading for either a stable, known address for device-mapper, or a workaround that will immunize tar against its apparently uncontrolled wanderings.
Please, pretty please, with sugar and cream in it even, please, give some stability to this situation....
Signature · 7 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) George Washington was first in war, first in peace -- and the first to have his birthday juggled to make a long weekend. -- Ashley Cooper
Quoted 1 line · Gene Heskett
Now the 64k$ question: While running with LVM2 managed disks, is it possible to run without
dm_mod, the device-mapper? If so, please tell me how to achieve this.
No; device mapper is the kernel portion of LVM2.
Jeff, who actively avoids LVM on home computers
Quoted 8 lines · Jeff Garzik
Quoted 3 lines · Gene Heskett
Now the 64k$ question: While running with LVM2 managed disks, is it possible to run without
dm_mod, the device-mapper? If so, please tell me how to achieve this.No; device mapper is the kernel portion of LVM2.
Jeff, who actively avoids LVM on home computers
Ya shoulda warned me. :-)
It should have lots bigger warning labels, in bright red self-illuminating signs, than it does. It in fact seems to work well, but it is a major breakage for some common apps, like doing backups...
Signature · 7 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) Gates' Law: Every 18 months, the speed of software halves. -- From a Slashdot.org post
Quoted 11 lines · Gene Heskett
Quoted 5 lines · Jeff Garzik
Gene Heskett wrote:
No; device mapper is the kernel portion of LVM2.
Jeff, who actively avoids LVM on home computers
Ya shoulda warned me. :-)
It should have lots bigger warning labels, in bright red self-illuminating signs, than it does.
Actually, you have to blame some FC developer for making an LVM-based setup the _default_ for home installs (aka "FC", enterprise is RHEL/¢OS).
Quoted 2 lines
It in fact seems to work well, but it is a major breakage for some common apps, like doing backups...
dm is on 254 for me.. in opensuse with a 2.6.20 that is. I wonder why it even moves around. However, even then, those who use udev and device names rather than (major,minor) tuples should not have any problem.
Jan
Quoted 5 lines · Jan Engelhardt
dm is on 254 for me.. in opensuse with a 2.6.20 that is. I wonder why it even moves around. However, even then, those who use udev and device names rather than (major,minor) tuples should not have any problem.
It moves around because someone at some point thought it was a great idea to assign dynamic majors to core functionality.
It stays put at 254 for you since that's the highest (and thus first-used) dynamic major.
-hpa
Quoted 6 lines · H. Peter Anvin
Quoted 4 lines · Jan Engelhardt
dm is on 254 for me.. in opensuse with a 2.6.20 that is. I wonder why it even moves around. However, even then, those who use udev and device names rather than (major,minor) tuples should not have any problem.
Quoted 7 lines
It moves around because someone at some point thought it was a great idea to assign dynamic majors to core functionality.
It stays put at 254 for you since that's the highest (and thus first-used) dynamic major.
-hpa
Until I also build pkdcdvd or md, each of which shove it down the list.
Now, as an update here at the coyote.den, fixing the typu in my /etc/modprobe.conf (Daves original message used dm_mod in the sample and it should have been dm-mod) and another rebuild/install cycle, and it is now at the LANANA assigned address that the patch which was reverted put it at, 238=EE and EE00h=60928. Now all we have to do is write a script to fix the reference files tar is given for incremental detection so that the device number is corrected to the current device as reported by a 'stat /' command.
However, there seems to be a wholesale conviction that the files are encrypted or otherwise smunched, they are not, but absolutely plain text:
[amanda@coyote gnutar-lists]$ cat coyote_var_5
1175495769
60928 19169380 ./run/cups
60928 19300353 ./cache/gstreamer-0.8
60928 19169448 ./lib/scrollkeeper/az
60928 19595604 ./tmp/RealPlayer/share/locale/ko
60928 19300370 ./cache/yum/updates-source
60928 19169358 ./cache/man/local
60928 19235048 ./tmp/kdecache-root/background
60928 19988737 ./tmp/kdecache-root/kio_help/root/.kde/share/doc 60928 19791895 ./spool/cups-pdf
So writing a script that compares that first number to one of the FE, FD, FC variations and corrects it to the EE value as converted above should be relatively trivial and I'll try to do that yet today, but there are a couple of honeydo's to take care of first.
Signature · 7 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) Don't talk to me about naval tradition. It's nothing but rum, sodomy and the lash. -- Winston Churchill
Quoted 9 lines · H. Peter Anvin
Quoted 4 lines · Jan Engelhardt
dm is on 254 for me.. in opensuse with a 2.6.20 that is. I wonder why it even moves around. However, even then, those who use udev and device names rather than (major,minor) tuples should not have any problem.
It moves around because someone at some point thought it was a great idea to assign dynamic majors to core functionality.
What were they smoking, I want some of that!
Quoted 4 lines
It stays put at 254 for you since that's the highest (and thus first-used) dynamic major.
-hpa
Signature · 7 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) <Culus> OH MY GOD NOT A RANDOM QUOTE GENERATOR <netgod> surely you didnt think that was static? how lame would that be? :-)
Quoted 11 lines · Gene Heskett
Quoted 9 lines · H. Peter Anvin
Quoted 4 lines · Jan Engelhardt
dm is on 254 for me.. in opensuse with a 2.6.20 that is. I wonder why it even moves around. However, even then, those who use udev and device names rather than (major,minor) tuples should not have any problem.
It moves around because someone at some point thought it was a great idea to assign dynamic majors to core functionality.
What were they smoking, I want some of that!
Do you actually use udev?
Jan
Quoted 8 lines · Jan Engelhardt
Quoted 7 lines · Gene Heskett
Quoted 6 lines · H. Peter Anvin
Quoted 4 lines · Jan Engelhardt
dm is on 254 for me.. in opensuse with a 2.6.20 that is. I wonder why it even moves around. However, even then, those who use udev and device names rather than (major,minor) tuples should not have any problem.
Quoted 6 lines
Quoted 4 lines
Quoted 3 lines
It moves around because someone at some point thought it was a great idea to assign dynamic majors to core functionality.
What were they smoking, I want some of that!
Do you actually use udev?
udev doesn't help the problem he is having (and he is using it, since he is using Fedora).
His problem is that GNU tar stores the device number of the file in a database that it uses to perform incremental updates, and considers the file new (and re-dumps it) when the device number changes.
So, when he changes the kernel config, adding any block device that uses a dynamic major (and is loaded before device-mapper), or goes between kernels that have the reserve-LANANA numbers patch, his incremental dumps become full ones because device-mapper gets a different dynamic major.
Signature · 1 line
Dave Dillow <[email protected]>
Quoted 12 lines · Dave Dillow
Quoted 11 lines · Jan Engelhardt
Quoted 10 lines · Gene Heskett
Quoted 9 lines · H. Peter Anvin
Quoted 4 lines · Jan Engelhardt
dm is on 254 for me.. in opensuse with a 2.6.20 that is. I wonder why it even moves around. However, even then, those who use udev and device names rather than (major,minor) tuples should not have any problem.
It moves around because someone at some point thought it was a great idea to assign dynamic majors to core functionality.
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
What were they smoking, I want some of that!
Do you actually use udev?
udev doesn't help the problem he is having (and he is using it, since he is using Fedora).
However, it also doesn't explain what the point is of backing up /dev when it's dynamically created.
Dave
Signature · 1 line
http://www.codemonkey.org.uk
Quoted 15 lines · Dave Jones
Quoted 14 lines · Dave Dillow
Quoted 13 lines · Jan Engelhardt
Quoted 11 lines · Gene Heskett
Quoted 9 lines · H. Peter Anvin
Quoted 4 lines · Jan Engelhardt
dm is on 254 for me.. in opensuse with a 2.6.20 that is. I wonder why it even moves around. However, even then, those who use udev and device names rather than (major,minor) tuples should not have any problem.
It moves around because someone at some point thought it was a great idea to assign dynamic majors to core functionality.
What were they smoking, I want some of that!
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
Do you actually use udev?
udev doesn't help the problem he is having (and he is using it, since he is using Fedora).
However, it also doesn't explain what the point is of backing up /dev when it's dynamically created.
It's not /dev he's backing up -- its /home, /usr, and others. GNU tar saves the device and inode numbers from the {,l}stat() call on each file and decides it is a new file if either number changes from run to run.
You are correct about /dev -- who'd back it up if it is dynamically created? But that is also not the problem Gene is having.
Quoted 6 lines · Dave Dillow
Quoted 2 lines
However, it also doesn't explain what the point is of backing up /dev when it's dynamically created.
It's not /dev he's backing up -- its /home, /usr, and others. GNU tar saves the device and inode numbers from the {,l}stat() call on each file and decides it is a new file if either number changes from run to run.
Ah apologies, I jumped into the thread halfway, and misunderstood the problem.
Dave
Signature · 1 line
http://www.codemonkey.org.uk
Quoted 3 lines · Dave Dillow
It's not /dev he's backing up -- its /home, /usr, and others. GNU tar saves the device and inode numbers from the {,l}stat() call on each file and decides it is a new file if either number changes from run to run.
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.
Quoted 8 lines · Olaf Hering
Quoted 4 lines · Dave Dillow
It's not /dev he's backing up -- its /home, /usr, and others. GNU tar saves the device and inode numbers from the {,l}stat() call on each file and decides it is a new file if either number changes from run to run.
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.
YOU Tell that to the tar/star people, they are flabbergasted that its not stable. It apparently is for every other OS tar can be run on.
Signature · 6 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) All your people must learn before you can reach for the stars. -- Kirk, "The Gamesters of Triskelion", stardate 3259.2
Quoted 10 lines · Gene Heskett
Quoted 7 lines · Olaf Hering
Quoted 4 lines · Dave Dillow
It's not /dev he's backing up -- its /home, /usr, and others. GNU tar saves the device and inode numbers from the {,l}stat() call on each file and decides it is a new file if either number changes from run to run.
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.YOU Tell that to the tar/star people, they are flabbergasted that its not stable. It apparently is for every other OS tar can be run on.
It's just the new world that we live in.
I don't see us ever going back to 100% static major/minors.
Though as a point of history, Linux has /always/ supported dynamic major/minor numbers, even back in 0.99 days. We avoided the problem then because the dynamic major/minors were only used by drivers that had not yet had static numbers assigned, or their author chose to avoid LANANA for some other reason.
Jeff
Quoted 9 lines · Gene Heskett
Quoted 8 lines · Olaf Hering
Quoted 4 lines · Dave Dillow
It's not /dev he's backing up -- its /home, /usr, and others. GNU tar saves the device and inode numbers from the {,l}stat() call on each file and decides it is a new file if either number changes from run to run.
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.
I forgot to add: '.. not stable across reboots.'
Quoted 2 lines
YOU Tell that to the tar/star people, they are flabbergasted that its not stable. It apparently is for every other OS tar can be run on.
They probably have a point with the st_dev usage.
You better find out why your major:minor pairs keep jumping around. Simply because the 'not stable across reboot' statement holds only for added/removed disks and maybe if the detection order changes. If your setup relies on a certain order, make sure the drivers get loaded in a fixed order. Its not clear from your other mails what exactly caused it. If its only due to a temporary change in testkernels, noone can do anything about it.
Quoted 17 lines · Olaf Hering
Quoted 9 lines · Gene Heskett
Quoted 8 lines · Olaf Hering
Quoted 4 lines · Dave Dillow
It's not /dev he's backing up -- its /home, /usr, and others. GNU tar saves the device and inode numbers from the {,l}stat() call on each file and decides it is a new file if either number changes from run to run.
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.I forgot to add: '.. not stable across reboots.'
Quoted 2 lines
YOU Tell that to the tar/star people, they are flabbergasted that its not stable. It apparently is for every other OS tar can be run on.
Quoted 8 lines
They probably have a point with the
st_devusage.
You better find out why your major:minor pairs keep jumping around. Simply because the 'not stable across reboot' statement holds only for added/removed disks and maybe if the detection order changes. If your setup relies on a certain order, make sure the drivers get loaded in a fixed order. Its not clear from your other mails what exactly caused it. If its only due to a temporary change in testkernels, noone can do anything about it.
he's running into problems when he adds additional drivers to the kernel (I don't remember which drivers, he's mentioned them earlier in this thread)
David Lang
Quoted 7 lines · Gene Heskett
Quoted 6 lines · Olaf Hering
Quoted 4 lines · Dave Dillow
It's not /dev he's backing up -- its /home, /usr, and others. GNU tar saves the device and inode numbers from the {,l}stat() call on each file and decides it is a new file if either number changes from run to run.
Quoted 5 lines
Quoted 2 lines
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.YOU Tell that to the tar/star people, they are flabbergasted that its not stable. It apparently is for every other OS tar can be run on.
FreeBSD also seems to be quite "dynamic". /dev/da0 is (0,92) for the 'fixit shell' -- how about you?
Jan
Quoted 8 lines · Jan Engelhardt
Quoted 7 lines · Gene Heskett
Quoted 6 lines · Olaf Hering
Quoted 4 lines · Dave Dillow
It's not /dev he's backing up -- its /home, /usr, and others. GNU tar saves the device and inode numbers from the {,l}stat() call on each file and decides it is a new file if either number changes from run to run.
Quoted 9 lines
Quoted 5 lines
Quoted 2 lines
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.YOU Tell that to the tar/star people, they are flabbergasted that its not stable. It apparently is for every other OS tar can be run on.
FreeBSD also seems to be quite "dynamic".
/dev/da0is (0,92) for the 'fixit shell' -- how about you?
I don't have such a beast here. Whats that supposed to do?
Quoted 2 lines
Jan
Signature · 6 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) Life sucks, but death doesn't put out at all. -- Thomas J. Kopp
Quoted 11 lines · Gene Heskett
Quoted 10 lines
Quoted 6 lines
Quoted 3 lines
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.YOU Tell that to the tar/star people, they are flabbergasted that its not stable. It apparently is for every other OS tar can be run on.
FreeBSD also seems to be quite "dynamic".
/dev/da0is (0,92) for the 'fixit shell' -- how about you?I don't have such a beast here. Whats that supposed to do?
Well I was just pointing out that Linux is not the only one to have device numbers (for promiment block-backed storage) that can move across reboots.
Jan
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Jan" == Jan Engelhardt <[email protected]> writes:
Quoted 11 lines · Gene Heskett
Quoted 11 lines
Quoted 10 lines
Quoted 6 lines
Quoted 3 lines
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.YOU Tell that to the tar/star people, they are flabbergasted that its not stable. It apparently is for every other OS tar can be run on.
FreeBSD also seems to be quite "dynamic".
/dev/da0is (0,92) for the 'fixit shell' -- how about you?I don't have such a beast here. Whats that supposed to do?
Jan> Well I was just pointing out that Linux is not the only one to Jan> have device numbers (for promiment block-backed storage) that can Jan> move across reboots.
I guess one work around would be to export your local filesystems via NFS and then have amanda back them up that way, which means gnutar should then ignore the devmajor and devminor numbers.
Or you could just use a hack of the perl script 'tarcust' to hack the minor/major numbers stored in the index files, and make them match the current major/minor pair for each filesystem just before you do a backup.
A total hack, but probably what I would do, since gtar should (IMNSHO) just ignore the devmajor/devminor and just go by the name. It's not like the sysadmin can't shoot himself in the foot anyway.
Or, again, get away from Amanda and tar and instead move to Bacula. Smarter software.
John
Quoted 28 lines · John Stoffel
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Jan" == Jan Engelhardt <[email protected]> writes:
Quoted 11 lines · Gene Heskett
Quoted 11 lines
Quoted 9 lines
Quoted 6 lines
Quoted 2 lines
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.YOU Tell that to the tar/star people, they are flabbergasted that its not stable. It apparently is for every other OS tar can be run on.
FreeBSD also seems to be quite "dynamic".
/dev/da0is (0,92) for the 'fixit shell' -- how about you?I don't have such a beast here. Whats that supposed to do?
Jan> Well I was just pointing out that Linux is not the only one to Jan> have device numbers (for promiment block-backed storage) that can Jan> move across reboots.
I guess one work around would be to export your local filesystems via NFS and then have amanda back them up that way, which means gnutar should then ignore the devmajor and devminor numbers.
Or you could just use a hack of the perl script 'tarcust' to hack the minor/major numbers stored in the index files, and make them match the current major/minor pair for each filesystem just before you do a backup.
Quoted 8 lines
A total hack, but probably what I would do, since gtar should (IMNSHO) just ignore the devmajor/devminor and just go by the name. It's not like the sysadmin can't shoot himself in the foot anyway.
Or, again, get away from Amanda and tar and instead move to Bacula. Smarter software.
John
Does Bacula not use tar for its dirty work?
Signature · 5 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) Nobody ever died from oven crude poisoning.
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Gene" == Gene Heskett <[email protected]> writes:
Gene> Does Bacula not use tar for its dirty work?
Nope, it uses it's own filesystem walking code.
John
Quoted 7 lines · John Stoffel
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Gene" == Gene Heskett <[email protected]> writes:
Gene> Does Bacula not use tar for its dirty work?
Nope, it uses it's own filesystem walking code.
John
Humm, interesting John. Neither yumex, or smart, is offering it to me as an installable rpm for FC6.
Signature · 5 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) Everything should be built top-down, except this time.
Gene> Humm, interesting John. Neither yumex, or smart, is offering it Gene> to me as an installable rpm for FC6.
Go and grab it from http://www.bacula.org, I'm pretty sure there are RPMs available on there. You might also need to install seperate RPMs for the follwoing packages:
bacula-common
bacula-director
bacula-sd
bacula-fd
bacula-console
bacula-mysql
Oh yeah, bacula really wants to use an SQL DB for it's file index. So I'd strongly suggest you go and read the manual,which is quite complete, before you do anything else.
Feel free to contact me off-list for help, or join the bacula-users list. We're a bit off topic for LKML now...
John
Quoted 7 lines · Jan Engelhardt
Quoted 6 lines · Gene Heskett
Quoted 6 lines
Quoted 5 lines
Quoted 2 lines
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.YOU Tell that to the tar/star people, they are flabbergasted that its not stable. It apparently is for every other OS tar can be run on.
Quoted 9 lines
Quoted 4 lines
Quoted 2 lines
FreeBSD also seems to be quite "dynamic".
/dev/da0is (0,92) for the 'fixit shell' -- how about you?I don't have such a beast here. Whats that supposed to do?
Well I was just pointing out that Linux is not the only one to have device numbers (for promiment block-backed storage) that can move across reboots.
Which is one more argument in favor of fixing tar, Jorg not withstanding. Yeah, that Jorg... AFAIC, time & technology move on, and I believe I see a future where the actual device number might not be the same across a reboot, for reasons a hell of a lot less clear than what triggered this latest 3 week long battle with something so simple as making a backup.
But, its a battle we'll never win with Jorg stirring the pot, he is the single, most vociferous opponent of anything labeled progress that I've ever had the displeasure to trying to follow in his rants, and that IS what they are, on this and on the scsi list when I was subscribed there.
I do think that a reasoned argument could be made to Paul Eggert, for at least a command line option to be made available that would cause tar to ignore it, process it yes, but ignore it for --listed-incremental in particular. Maybe even make it part of the --listed-incremental as a permanent feature. For us, that would be the ideal. It is not an issue for level 0's of course.
But my single voice is not going to be able to effect a campaign, its going to take a general consensus of all the movers and shakers of linux, TPTB if you will, all ganging up on Paul. Or someone doing a patch to that effect, then making distro packages out of it, such as debs, rpm's, etc, etc. A fork I think its called.
There is another also listed in the tar ChangeLog, but I haven't rx'd a response from that person although he was Cc:'d at the time.
Dave Dillow, you mentioned another routine in tar that might be an educational file to walk through, besides the compare.c I tried to play with and apparently broke, can you repeat that phrase again? I might take a stick and poke around in that one too. Maybe I can un-break it? :)
Signature · 5 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) Earn cash in your spare time -- blackmail your friends.
Quoted 8 lines · Olaf Hering
Quoted 4 lines · Dave Dillow
It's not /dev he's backing up -- its /home, /usr, and others. GNU tar saves the device and inode numbers from the {,l}stat() call on each file and decides it is a new file if either number changes from run to run.
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.
YOU Tell the tar people, they are flabbergasted that linux is apparently the only unstable OS that tar can be run on.
Signature · 5 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) Stay the curse.
Quoted 5 lines
Quoted 2 lines
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.YOU Tell the tar people, they are flabbergasted that linux is apparently the only unstable OS that tar can be run on.
Linux is not unstable. It's developmental! Linux is like one giant international research project marching toward the future of computing. MacOS (oh, how I miss System 7.5.5), Windows (shoot me, please), BSD (wannabe GNU/Linux), and everything else are there just for companies to spend money on and idiots to drool over until Linux is ready to stand up and say "Yo. It's just bugfixes and improvements from here. You can put the Legos down now."
Quoted 6 lines · John Anthony Kazos Jr.
Quoted 5 lines
Quoted 2 lines
So fix tar to not do silly things.
Kernel major:minor numbers are not stable.YOU Tell the tar people, they are flabbergasted that linux is apparently the only unstable OS that tar can be run on.
Quoted 7 lines
Linux is not unstable. It's developmental! Linux is like one giant international research project marching toward the future of computing. MacOS (oh, how I miss System 7.5.5), Windows (shoot me, please), BSD (wannabe GNU/Linux), and everything else are there just for companies to spend money on and idiots to drool over until Linux is ready to stand up and say "Yo. It's just bugfixes and improvements from here. You can put the Legos down now."
I don't know who you are, but that's not germane to this discussion, what is germane is that once the experimental label has been removed from something as capable of screwing up the system as the LMV stuff is, then there is no excuse not to give it a LANANA compatible address and quit screwing up peoples backup systems just because they turned on pktcdvd or md or some other similar option in a make xconfig. FWIW, md, for raid users, should get the same treatment since raids often can have terrabytes of data to be backed up.
The patch to do that caught tar users totally off-guard, but once we(I) understood why it was done, it was obviously a good patch. But rather than explaining WTF it was doing so we/I understood the reason for the pain, the *&%^ patch was reverted, starting the backup related screwups all over again from scratch.
Now this thread has a life of its own, and the noise it continues to make will prevent the re-application of that patch which WILL stabilize things, probably until hell freezes over and pigs can use it for a takeoff runway.
I have a fix (thanks Dave D.) worked out for MY system, and you guys can do as you damned well please, until you break my fix anyway. In the meantime, I have backup.sh, my amdump wrapper, running every 3 hours around the clock without a verify stage until it has managed to update all 118 (about 20 megs worth) of its reference files. I expect that to take 3 or 4 days. Hopefully my drives are up to the torture, if not, well they aren't that precious & I CAN drop the card for more & might even do that yet tonight unless Circuit City or Staples think that a 500Gb is still worth its weight in gold bar. Yes, I spend my money locally where possible.
Signature · 6 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) "Don't discount flying pigs before you have good air defense." -- [email protected]
Quoted 1 line · Gene Heskett
YOU Tell the tar people, they are flabbergasted that linux is apparently the only unstable OS that tar can be run on.
How about cygwin/windows? It has no concept of static device numbers. And what about external usb disks on any other unix os? Surely they don't have static minor device numbers in the face of hot plugging?
The bug is in tar and that is what needs fixed. It should not attach any meaning to the device number, but rather should assume that the admin knows what he is doing when he said to backup a given path using a given incremental backup log, and that he isn't yanking tar's chain and pointing it to a different path between incremental backups.
Quoted 8 lines · Phillip Susi
Quoted 2 lines · Gene Heskett
YOU Tell the tar people, they are flabbergasted that linux is apparently the only unstable OS that tar can be run on.
How about cygwin/windows? It has no concept of static device numbers. And what about external usb disks on any other unix os? Surely they don't have static minor device numbers in the face of hot plugging?
Quoted 5 lines
The bug is in tar and that is what needs fixed. It should not attach any meaning to the device number, but rather should assume that the admin knows what he is doing when he said to backup a given path using a given incremental backup log, and that he isn't yanking tar's chain and pointing it to a different path between incremental backups.
I violently agree on that point, but the limited conversations I've had with the tar maintainer so far indicates that AFAHIC, its linux that's broken. A new device number is a new disk and must be treated as such.
Signature · 6 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) QOTD: "I'll listen to reason when it comes out on CD."
Quoted 1 line · Gene Heskett
I violently agree on that point, but the limited conversations I've had with the tar maintainer so far indicates that AFAHIC, its linux that's broken. A new device number is a new disk and must be treated as such.
Sumbit a patch adding an option to ignore the device number, and slap him around a bit with a large trout ;)
Quoted 8 lines · Phillip Susi
Quoted 4 lines · Gene Heskett
I violently agree on that point, but the limited conversations I've had with the tar maintainer so far indicates that AFAHIC, its linux that's broken. A new device number is a new disk and must be treated as such.
Sumbit a patch adding an option to ignore the device number, and slap him around a bit with a large trout ;)
Well, I made one pass at that, and blew it up with a 2 pound block of C4. I know just enough about C to be dangerous, and my real C knowledge is pretty much K&R.
As for the large trout, them are definitely on the hard to find list here in WV. Now if I still lived in Rapid City S.D., this time of the year is a great time to pick up the odd old hook nosed Brown Trout that might go as high as 8 or 9 pounds and take up 30+ inches laying on your yardstick. I had 2 such on my stringer about 40 years ago one late April day but they got to fighting it, broke it and got away. I thought my wife, in the other end of the boat, was going to throw me overboard to retrieve them, but I don't swim well, don't even float, and the nearest solid earth was 170 feet away, straight down... Fortunately for me she didn't. Unfortunately, she had a stroke and died a year later, at age 34. I still miss her, she was what you would call a good woman.
Signature · 7 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) Politics is not the art of the possible. It consists in choosing between the disastrous and the unpalatable. -- John Kenneth Galbraith
Quoted 6 lines
Quoted 3 lines
I violently agree on that point, but the limited conversations I've had with the tar maintainer so far indicates that AFAHIC, its linux that's broken. A new device number is a new disk and must be treated as such.
Sumbit a patch adding an option to ignore the device number, and slap him around a bit with a large trout ;)
I think that's especially true. If a user begins with a single full disk for their entire filesystem, uses tar to backup, and then later adds a second disk, copies everything from /usr and /home onto partitions there (making sure to preserve all interesting bits like ctime/mtime), and mounts them over the original directories, tar should not decide that every file in /usr and /home was deleted and recreated.
Quoted 6 lines · John Anthony Kazos Jr.
I think that's especially true. If a user begins with a single full disk for their entire filesystem, uses tar to backup, and then later adds a second disk, copies everything from /usr and /home onto partitions there (making sure to preserve all interesting bits like ctime/mtime), and mounts them over the original directories, tar should not decide that every file in /usr and /home was deleted and recreated.
That wouldn't work, even if you managed to copy ctime, the target inode number will be different than the original and backup has to assume it's a different file.
For filesystems with no stable inode numbers my backup scheme assumes that the same file = the same name and times but it's really fragile.
An option to ignore block device number does make sense, of course. Especially when backing up only one filesystem (-xdev etc).
Signature · 1 line
Krzysztof Halasa
Quoted 9 lines · Krzysztof Halasa
Quoted 6 lines · John Anthony Kazos Jr.
I think that's especially true. If a user begins with a single full disk for their entire filesystem, uses tar to backup, and then later adds a second disk, copies everything from /usr and /home onto partitions there (making sure to preserve all interesting bits like ctime/mtime), and mounts them over the original directories, tar should not decide that every file in /usr and /home was deleted and recreated.
Quoted 3 lines
That wouldn't work, even if you managed to copy ctime, the target inode number will be different than the original and backup has to assume it's a different file.
Time for rsync and/or rsync -c. And some target filesystem that is fused to a [can-be-]non-solid archive, like 7z.
Jan
Quoted 17 lines · Dave Jones
Quoted 16 lines · Dave Dillow
Quoted 14 lines · Jan Engelhardt
Quoted 11 lines · Gene Heskett
Quoted 8 lines · H. Peter Anvin
Quoted 4 lines · Jan Engelhardt
dm is on 254 for me.. in opensuse with a 2.6.20 that is. I wonder why it even moves around. However, even then, those who use udev and device names rather than (major,minor) tuples should not have any problem.
It moves around because someone at some point thought it was a great idea to assign dynamic majors to core functionality.
What were they smoking, I want some of that!
Do you actually use udev?
Quoted 7 lines
Quoted 2 lines
udev doesn't help the problem he is having (and he is using it, since he is using Fedora).
However, it also doesn't explain what the point is of backing up /dev when it's dynamically created.
Dave
I'm glad you mentioned that, I was, leftovers from RH7.3 probably. :(
Signature · 6 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) Windows Tip of the Day: Add DEVICE=FNGRCROS.SYS to your CONFIG.SYS file.
Quoted 8 lines · Jan Engelhardt
Quoted 7 lines · Gene Heskett
Quoted 6 lines · H. Peter Anvin
Quoted 4 lines · Jan Engelhardt
dm is on 254 for me.. in opensuse with a 2.6.20 that is. I wonder why it even moves around. However, even then, those who use udev and device names rather than (major,minor) tuples should not have any problem.
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
It moves around because someone at some point thought it was a great idea to assign dynamic majors to core functionality.
What were they smoking, I want some of that!
Do you actually use udev?
Yes, and it works just fine, for non-LVM filesystems.
Quoted 3 lines
Jan
Signature · 6 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) "The greatest warriors are the ones who fight for peace." -- Holly Near
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Gene" == Gene Heskett <[email protected]> writes:
Quoted 8 lines · Jeff Garzik
Quoted 8 lines
Quoted 3 lines · Gene Heskett
Now the 64k$ question: While running with LVM2 managed disks, is it possible to run without
dm_mod, the device-mapper? If so, please tell me how to achieve this.No; device mapper is the kernel portion of LVM2.
Jeff, who actively avoids LVM on home computers
Gene> Ya shoulda warned me. :-)
Gene> It should have lots bigger warning labels, in bright red Gene> self-illuminating signs, than it does. It in fact seems to work Gene> well, but it is a major breakage for some common apps, like Gene> doing backups...
Quoted 1 line
From the GNU info doc on the tar program:
Metadata stored in snapshot files include device numbers, which, obviously is supposed to be a non-volatile value. However, it turns out that NFS devices have undependable values when an automounter gets in the picture. This can lead to a great deal of spurious redumping in incremental dumps, so it is somewhat useless to compare two NFS devices numbers over time. The solution implemented currently is to considers all NFS devices as being equal when it comes to comparing directories; this is fairly gross, but there does not seem to be a better way to go.
This to me, seems to be another borkage of the tar program. Can you tell Amanda to use 'dump/restore' instead of tar instead? Or heck, just move to Bacula, it's smarter than amanda/tar and it supports writing to DVDs, etc. http://www.bacula.org
This tar borkness is pretty annoying, since Amanda (from the docs page) only keeps indexes in terms of host/filesystem/path/date, nothing about the major/minor (device) numbers.
It just re-inforces my desire to never use Amanda again.
Quoted 31 lines · John Stoffel
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Gene" == Gene Heskett <[email protected]> writes:
Quoted 8 lines · Jeff Garzik
Quoted 8 lines
Quoted 3 lines · Gene Heskett
Now the 64k$ question: While running with LVM2 managed disks, is it possible to run without
dm_mod, the device-mapper? If so, please tell me how to achieve this.No; device mapper is the kernel portion of LVM2.
Jeff, who actively avoids LVM on home computers
Gene> Ya shoulda warned me. :-)
Gene> It should have lots bigger warning labels, in bright red Gene> self-illuminating signs, than it does. It in fact seems to work Gene> well, but it is a major breakage for some common apps, like Gene> doing backups...
Quoted 1 line
From the GNU info doc on the tar program:
Metadata stored in snapshot files include device numbers, which, obviously is supposed to be a non-volatile value. However, it turns out that NFS devices have undependable values when an automounter
gets in the picture. This can lead to a great deal of spurious redumping in incremental dumps, so it is somewhat useless to compare two NFS devices numbers over time. The solution implemented currently is to considers all NFS devices as being equal when it comes to comparing directories; this is fairly gross, but there does not seem to be a better way to go.
Quoted 5 lines
This to me, seems to be another borkage of the tar program. Can you tell Amanda to use 'dump/restore' instead of tar instead? Or heck, just move to Bacula, it's smarter than amanda/tar and it supports writing to DVDs, etc. http://www.bacula.org
I wouldn't touch dump/restore with a 50 foot pole, particularly since I have serious doubts about it viability in the LVM environment.
For those of you with big tapes that can hold a complete dump of every partition (and partitions is the only way dump works in case some have forgotten), go ahead and use dump/restore. Tar quite simply, allows one to break his backup files down into small enough pieces that a tape drive that's only 20% of the system drives size is totally usable. I ran dds2 tapes for a long time, and it wasn't at all unusual to have amanda fill those to the 95% or better mark every night for a week running, without ever hitting EOT.
Quoted 5 lines
This tar borkness is pretty annoying, since Amanda (from the docs page) only keeps indexes in terms of host/filesystem/path/date, nothing about the major/minor (device) numbers.
It just re-inforces my desire to never use Amanda again.
Your call, that's what this linux thing is all about. But do I not recall your name from the amanda lists at some point in the last 8+ years?
Quoted 3 lines
Signature · 7 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) <Culus> OH MY GOD NOT A RANDOM QUOTE GENERATOR <netgod> surely you didnt think that was static? how lame would that be? :-)
Quoted 1 line · Gene Heskett
For those of you with big tapes that can hold a complete dump of every partition (and partitions is the only way dump works in case some have forgotten), go ahead and use dump/restore. Tar quite simply, allows one to break his backup files down into small enough pieces that a tape drive that's only 20% of the system drives size is totally usable. I ran dds2 tapes for a long time, and it wasn't at all unusual to have amanda fill those to the 95% or better mark every night for a week running, without ever hitting EOT.
Wow, people still use tapes for backup?
With current hard drive prices (200GB @ US$55, 500GB @ US$120) you can just keep buying hard drives :)
Surely tape price/GB is higher than hard drive price/GB...
Jeff
Quoted 6 lines · Jeff Garzik
Wow, people still use tapes for backup?
With current hard drive prices (200GB @ US$55, 500GB @ US$120) you can just keep buying hard drives :)
Surely tape price/GB is higher than hard drive price/GB...
Tapes can be archived for long time, drives can't. Drives need to be mounted into fairly bulky subsystems, tapes can be easily handled by robots.
Ralf
Quoted 1 line · Ralf Baechle
Tapes can be archived for long time, drives can't. Drives need to be
True.
Quoted 2 lines
mounted into fairly bulky subsystems, tapes can be easily handled by robots.
Actually there are HD robots now :)
Jeff
Quoted 6 lines · Jeff Garzik
Quoted 3 lines
mounted into fairly bulky subsystems, tapes can be easily handled by robots.
Actually there are HD robots now :)
The best part is you can just electrify each slot, no need for mechanical arms.
-hpa
Quoted 4 lines · Jeff Garzik
With current hard drive prices (200GB @ US$55, 500GB @ US$120) you can just keep buying hard drives :)
Surely tape price/GB is higher than hard drive price/GB...
Erm. No. We're in the middle of installing an StorageTek SL8500 for backups, simply because an LTO3 tape holds 300G at a price point of $100 or so. And there's no sane way to build a half-petabyte disk farm at that price. (And yes, our tape backup service is well into the half-petabyte range, and will probably shoot over that as soon as the SL8500 is in production and we deal with all the backlogged requests for backup service).
Don't forget to factor in the cost of disk shelves, power, cooling, and all that when you're building something to hold 2,500 200G drives. Oh. and controllers. And machines to put the controllers in... and all the rest of it.
And remember - you *dont* want to be backing up critical data on the sort of hard drives that cost $55 for 200G - you'll want multiple copies, probably some RAID, etc etc.
Quoted 12 lines · [email protected] wrote:
Quoted 3 lines · Jeff Garzik
With current hard drive prices (200GB @ US$55, 500GB @ US$120) you can just keep buying hard drives :)
Surely tape price/GB is higher than hard drive price/GB...
Erm. No. We're in the middle of installing an StorageTek SL8500 for backups, simply because an LTO3 tape holds 300G at a price point of $100 or so. And there's no sane way to build a half-petabyte disk farm at that price. (And yes, our tape backup service is well into the half-petabyte range, and will probably shoot over that as soon as the SL8500 is in production and we deal with all the backlogged requests for backup service).
Quoted 3 lines
Don't forget to factor in the cost of disk shelves, power, cooling, and all that when you're building something to hold 2,500 200G drives. Oh. and controllers. And machines to put the controllers in... and all the rest of it.
Who says you have to keep the hard drives plugged into a machine at all times?
Quoted 3 lines
And remember - you *dont* want to be backing up critical data on the sort of hard drives that cost $55 for 200G - you'll want multiple copies, probably some RAID, etc etc.
The same can be said for tape. Anyone who only has a single tape backup of critical data is an idiot. Plenty of war stories where such scenarios go awry...
Jeff
Quoted 10 lines · Jeff Garzik
Quoted 8 lines · Gene Heskett
For those of you with big tapes that can hold a complete dump of every partition (and partitions is the only way dump works in case some have forgotten), go ahead and use dump/restore. Tar quite simply, allows one to break his backup files down into small enough pieces that a tape drive that's only 20% of the system drives size is totally usable. I ran dds2 tapes for a long time, and it wasn't at all unusual to have amanda fill those to the 95% or better mark every night for a week running, without ever hitting EOT.
Quoted 8 lines
Wow, people still use tapes for backup?
With current hard drive prices (200GB @ US$55, 500GB @ US$120) you can just keep buying hard drives :)
Surely tape price/GB is higher than hard drive price/GB...
Jeff
I haven't seen any 200GB for $55 yet, more like $129 & maybe a rebate at Circuit City. We don't have a Fry's around here.
Signature · 5 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) You will be called upon to help a friend in trouble.
Gene> I haven't seen any 200GB for $55 yet, more like $129 & maybe a Gene> rebate at Circuit City. We don't have a Fry's around here.
Newegg.com, 320Gb for $85 ea, plus shipping, plus a SATA controller board, just under $200. I'm happy. And thanks for the SATA controller work Jeff!
John
Quoted 1 line · Gene Heskett
I haven't seen any 200GB for $55 yet, more like $129 & maybe a rebate at Circuit City. We don't have a Fry's around here.
pricewatch.com is your friend :)
Jeff
Quoted 2 lines · Gene Heskett
I haven't seen any 200GB for $55 yet, more like $129 & maybe a rebate at Circuit City. We don't have a Fry's around here.
Wow. 200GB HDs can be had for AUD91 here. I think you need to shop around. The internet can be your friend. :)
Signature · 3 lines
"To the extent that we overreact, we proffer the terrorists the
greatest tribute."
- High Court Judge Michael KirbyQuoted 1 line · Gene Heskett
I haven't seen any 200GB for $55 yet, more like $129 & maybe a rebate at Circuit City. We don't have a Fry's around here.
Fry's:
http:/http://shop1.outpost.com/product/4697788?site=sr:SEARCH:MAIN_RSLT_PG
500 GB SATA for $119. Pricewatch shows several other vendors having the same pricing.
-hpa
Quoted 4 lines · H. Peter Anvin
Quoted 2 lines · Gene Heskett
I haven't seen any 200GB for $55 yet, more like $129 & maybe a rebate at Circuit City. We don't have a Fry's around here.
Quoted 8 lines
Fry's:
http:/http://shop1.outpost.com/product/4697788?site=sr:SEARCH:MAIN_RSLT_ PG500 GB SATA for $119. Pricewatch shows several other vendors having the same pricing.
-hpa
I got a 320GB Maxtor, 3 year warranty, at Staples tonight for $90 & change. It could have been worse locally.
Signature · 6 lines
Cheers, Gene "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) QOTD: "I'll listen to reason when it comes out on CD."
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
"Gene" == Gene Heskett <[email protected]> writes:
Gene> I wouldn't touch dump/restore with a 50 foot pole, particularly Gene> since I have serious doubts about it viability in the LVM Gene> environment.
Why? It's can't be any worse than Tar with it's silly assumption about the static device numbers, which are shown in their own docs to be an invalid assumption over NFS, so why enforce them locally?
On Linux, dump/restore are open source, so it's not like you can't get your data back, you know the format it's written in. Any backup software has problems with filesystems which change underneath them as they run, esp for incremental backups.
The only sure strategy is to do your dumps in single user mode, with only one process touching the filesystem during the backup window. This obviously won't fly in a 24x7 environment, so people take their calculated chances and run backups with the system live.
Some people are working to be able to take snapshots of live filesystems, then backups the snapshots, but that's still a work in progress right now. I'd personally love to do that, but not quite yet.
Gene> For those of you with big tapes that can hold a complete dump of Gene> every partition (and partitions is the only way dump works in Gene> case some have forgotten), go ahead and use dump/restore. Tar Gene> quite simply, allows one to break his backup files down into Gene> small enough pieces that a tape drive that's only 20% of the Gene> system drives size is totally usable.
So does Bacula. *grin* It will happily span backups across multiple volumes (read tapes).
The problem with Amanda is that you have to *manually* break down your filesystem into chunks small enough to fit on a single volume. I don't have to think about it with Bacula.
Gene> I ran dds2 tapes for a long time, and it wasn't at all unusual Gene> to have amanda fill those to the 95% or better mark every night Gene> for a week running, without ever hitting EOT.
The problem with Amanda for the longest time was that if you had a full backup which spanned more than one tape, you were screwed.
Quoted 5 lines
Quoted 5 lines
This tar borkness is pretty annoying, since Amanda (from the docs page) only keeps indexes in terms of host/filesystem/path/date, nothing about the major/minor (device) numbers.
It just re-inforces my desire to never use Amanda again.
Gene> Your call, that's what this linux thing is all about. But do I Gene> not recall your name from the amanda lists at some point in the Gene> last 8+ years?
Probably, but it's been a long time since I was active in Amanda, probably back in my WPI days ([email protected]) or my Fluent days ([email protected]), but I honestly don't remember.
As I said, the big problem with Amanda for the longest time (and which has now been worked around I understand) was a single backup which had to span multiple volumes.
Bacula does this all for you, and works better than Amanda IMHO. It also allows you to stage to disk and then to DVD or other types of volumes.
And it doesn't have amanda's inistence of writing to a new piece of media each day either. Which I admit is a tradeoff, but for a home system, not having to swap tapes daily is a big help, and saves me money. DLTIV tapes aren't cheap, even on ebay...
Heck, if you don't trust Dump, find tar to be broken, what about cpio? *grin*
John
Quoted 4 lines · John Stoffel
The only sure strategy is to do your dumps in single user mode, with only one process touching the filesystem during the backup window. This obviously won't fly in a 24x7 environment, so people take their calculated chances and run backups with the system live.
That's what snapshots are for.
No need for single user mode or taking chances.
A snapshot gives you a consistent filesystem that you can dump or tar.
Jeff