As
someone who supported many Fortune 500 companies and Walmart
from 1996 to 2013. 20,000 SLES 11 systems. Many apps dependent
on quirks in libc so they almost never update their boxes - I
can say that many NCR and HP boxes still had the OS that was
installed on day one and only got patches if the application
on that box found a bug. Due to end of life of hardware,
Walmart has slowly converted AIX, HPUX and MP-RAS flavored
UNIX to SLES. My estimation that it would cost over $4 billion
dollars to turn off all the old Unix hardware and go to SLES
virtual machine in VMWare with snapshots. Yes, Walmart has a
few Redhat enterprise boxes because the software Walmart
needed is only supported on Redhat.
All this talk about restoring to a previous state does nothing for the largest Linux/Unix users. The have raw space databases (Oracle, Informix) and many other users spaces (Walmart had 3 outside the normal Linux ones).
Nothing
beats a good Disaster Recovery plan that is tested regularly.
My systems have the ability to go back up to 90 days and I
test it to Bare Iron on a random computer every month. Good
practice. All my daily systems are on a VirtualBox OpenSUSE
15.2 machine. The host is also a 15.2 machine but it only
hosts Multiple Virtual 15.2 systems only one can be seen on
the internet.
Walmart has discovered that snapshots do not recover deleted database rows. Only the OS.
Why
all the need for changing something that worked since 1973
(hint - we had different file systems because we only had 5 mb
disks and wanted to keep older systems able to run Unix) my
first Unix box had to be booted from tape - mounted the
booting disk and then removed it to mount the production disk.
If you were good you got a prompt on the terminal after 10
minutes from the start of the tape boot on a DEC PDP11/30. No
systemd could not speed that up.
At AT&T the 1982 Unix tapes were not 100% accurate and the cameraman fell asleep multiple times during the "exciting" lectures as did most of the audience.
I tried btrfs for one day and went back to ext4 - faster and more reliable. LVM was created to allow multiple separate small disk (when 64 mb was big) to be logically sewn together as a giant disk (I remember a site that had almost a GB in 14 SCSI Disks) it was also added for reliability as you could plex (Veritos term for Mirror) as many copies as you wanted and rebuild as long as you did not go down. That was 35 years ago - journaling file systems had not made it into production and hours long fscks were the norm on a power failure which is why some models had battery backups in their power supplys.
Would
it not be better to teach everyone how to backup and restore
their systems. I spent the late 80's and early 90's flying
from city to city to help recover Unix boxes that the
sysadmins had no idea what went wrong. I think I saw
everything that ever went wrong and saw more than one company
fold because their computers had not been backed up since
installed.
Am Donnerstag, 7. Januar 2021, 10:52:46 CET schrieb Andreas Schwab:On Jan 06 2021, Matěj Cepl wrote:(/var/lib/rpm to /usr/var/lib/rpm?/var/lib/rpm has already been moved to /usr/lib/sysimage/rpm.Given, that /usr might be treated read only and the nature of this content is a database, I don't understand that move at all. Pete