home *** CD-ROM | disk | FTP | other *** search
-
- Procedure for doing a dump (revised $Date: 92/06/05 13:11:17 $).
-
- Right now the dumping procedure is kind of labor intensive.
-
- Finding a tape:
-
- A level 0 (weekly) dump will take about 2 tapes (4 tapes using the old
- Exabyte 8200's). A single tape can hold about a week's worth of level
- 1 (daily) dumps. Tapes are stored in 608-2 Evans. Some are in the
- tape holder on the wall, more are in the file cabinet. Here is the
- method for reusing tapes. Daily dumps can be reused if there have
- been two weekly dumps done since they were created, although you'll
- want to save them as long as possible. That way we don't lose anything
- should a single weekly dump be unreadable. Weekly dumps are saved for
- a month, and every sixth weekly dump set is saved forever. After that
- weekly dump tapes can be reused as well. A database of what tapes
- were used for and when is kept in /sprite/admin/dump/dumpdates.
-
- The tapes in the tape holder are stored in chronological order
- left-to-right, like a circular buffer. Weekly dump tapes have
- rubberbands around them. The most current weekly dump has a paperclip
- in the rubberband. Older tapes are kept in the filing cabinet. The
- boxes marked "reuse" are ok for using. Boxes marked "weekly" are for
- weekly dump tapes. The box marked "bad" contains defective tapes that
- can be given away. When you clean out a slot in the tape holder put
- the daily dump tapes in a reuse box. If you are cleaning out the
- upper left slot the weekly tapes go into a weekly box. For any other
- slot the weekly tapes go into a reuse box. We should probably put
- tapes into one box and get them to use from another so that tapes
- don't get overwritten right away.
-
- Finding a drive:
-
- There are currently two drives being used for backups. To avoid
- problems with server crashes, they are generally hooked up to a client
- machine, often the home machine of the person doing the dumps. If you
- move the drives, you must edit the crontab of the new and old hosts so
- that the right machine does the daily dumps (see "Daily dumps" below).
- The drives are named /dev/rst04.nr and /dev/rst05.nr.
-
- Putting the tape in the drive:
-
- Push the button on the exabyte. After about 30 seconds the door will
- open. Make sure the red write-protect slider on the tape is not visible.
- Put in the new tape and close the door. Wait about 30 seconds
- and the green light will go on. Now you're ready to go. While the tape
- is being accessed the red light will flicker.
-
- Initializing a tape:
-
- Once you have found a tape you need to initialize it.
- Notes: you must wait until the green light goes on before initializing.
- You must be root to do the initialization.
-
- a) Initializing a blank tape
-
- If it is a blank tape use the command "dump -i <tape#> -f <device>".
- (See "Finding a drive" above for the list of device names.) The next
- tape number can be determined from the dumpdates file. In the long
- run, you can just look for the last "Initializing" line. For the time
- being, though, you should verify that there aren't any "Initializing"
- lines with a higher tape number than the one you plan to use.
-
- b) Initializing an old tape
-
- If you're reusing a tape use "dump -r -f <device>". Make sure the
- device you use is a "norewind" tape device.
-
- Daily dumps:
-
- Daily dumps are done via the script /sprite/admin.$MACHINE/dailydump.
- This script is run every night by cron. It will send mail to "dumper"
- when it completes or if there are errors. If the machine doing the
- dumps crashes while the dumps are in progress, "dumper" will get mail
- with an error message when it reboots. If this happens, you will need
- to start a new tape. This is because the dump program gets confused
- by the partial file at the end of the tape, at least when used with an
- 8500 tape drive.
-
- You can also check /sprite/admin/dump/dumplog for error messages. You
- can use /sprite/admin.$MACHINE/dumpsize to get an idea when the daily
- tape is about to fill up.
-
- If you take over as dumper during the middle of the week, you should
- edit /sprite/admin/dump/dumpalias by hand (so that it names you
- instead of the old dumper). If your first job as dumper is to run the
- weekly dumps, the weekly dump script will change the dumpalias file
- for you.
-
- Weekly dumps:
-
- 1. Log in as yourself to the machine with the drives, su to root, and
- run "/sprite/admin.$MACHINE/weeklydump -i". This will initialize the
- tape in the drive and start the dumps. This will disable the daily
- dumps by creating a lockfile "/sprite/admin/dump/doingweekly". This
- will also change the dumper alias to your userid. You should either
- run it from the console or run it with stdout and stderr redirected to
- a file or /dev/null. If you run it from, e.g., an rlogin shell
- without redirecting the output, the dump will abort when you log out.
- If something goes wrong with the weekly dump and you want to start over
- again, you will need to run "weeklydump startFs" where startFs is the first
- file system that you wish to start dumping. Otherwise, it will start dumping
- from the file system after the last successfully dumped file system.
-
- 2. When the tape on one drive fills up, the dumps will automatically
- resume using the next drive in the tape list (see
- /sprite/admin/dump/tapelist). If the dumps exhaust the list of tapes,
- the weeklydump script will send mail to "dumper", saying that the
- dumps need attention and giving the name of the next filesystem that
- needs to be dumped. (Alternatively, you can look in the dumplog or
- dumpdates files to see which filesystem couldn't be dumped.) With the
- old 8200's it took 10-12 hours to write the first two tapes. With the
- 8500's it takes about 15 hours to do the complete dumps (on two
- tapes). If the drives do run out, replace the tapes with new ones.
- Rerun "weeklydump -i". It will look in the lock file to determine the
- next directory to dump and resume dumping.
-
- As with the daily dumps, "dumper" will get mail in case of errors,
- crashes, or completion of the dumps.
-
- 3. One the dumps are complete, label the tapes (with the date and what
- filesystems they contain). Put another tape in the first drive (the
- one listed first in /sprite/admin/dump/tapelist) and reenable the
- daily dumps with "dailydump -i".
-
- Cleaning a tape drive:
-
- The tape drives should be cleaned once a quarter, using the drive
- cleaning kits in the filing cabinet in 608-2. Each kit is good for
- three cleanings. When you start a new kit, mark both the cartridge
- and the bottle with the drive that you are cleaning them with. Also
- write down the date of the cleaning on the cartridge. When reusing a
- kit, use the cartridge and bottle that are labeled for the drive you
- want to clean. (And of course write down the date on the cartridge.)
- Specific instructions on how to use the kit are in with the cartridge.
-
- Tape errors:
-
- Occasionally the dumps will fail due to some problem with the tape
- drive. There should be a syslog message from the tape driver which
- contains some magic numbers detailing the failure. To look up the
- meaning of the numbers, check out Appendices C and D of the Exabyte
- 8500 User's Manual. The "sense code" and "qualifier" in the syslog
- message correspond to the ASC and ASCQ numbers in Appendix C.
- Unfortunately, you have to guess what the Sense Key was. The "symptom
- code" in the syslog message corresponds to the FSC numbers in Appendix
- D.
-
- Other documentation:
-
- You may also wish to read the man pages for the following programs and
- scripts: dump, dailydump, weeklydump.
-