Broken Pipe errors
I recently added a new machine to my backups, and seem to be having a problem with a couple of file systems with "index tee cannot write [Broken pipe]" errors. Client is Solaris 7 Server is Solaris 7 amanda version 2.4.2p2 gnutar 1.13.19 firewall between the two servers, but no security restrictions (as in all other backups work fine). Server is NAT'ed at firewall I've run some tests just backing one of those filesystems up, and no matter how I fiddle with the dumptype (thinking compression in particular might be part of the problem) I cannot get this error to go away. In test mode I'm trying to dump only to holding disk which has 11.5 GB free. The file system in question is only ~1.5 GBs. Larger filesystems on the same host can be backed up with the same dumptype parameters with no problems. When I was watching the backup being performed with "amstatus" it showed data being transferred, and the correct file was being created and updated on the server. I think when it stopped updating, the size sent reported by amstatus was a few KB (32?) over the estimated backup size (didnt write the data down like a fool) - I realize this may or may not mean a thing, but just throwing it out there. Here are some (hopefully!) appropriate log files (they dont tell me too much): = log.20020225.0 from server = START planner date 20020225 START driver date 20020225 ERROR taper no-tape [no tape online] FINISH planner date 20020225 STATS driver startup time 2.472 FAIL dumper rnbuyer /export/home 0 [mesg read: Connection timed out] sendbackup: start [rnbuyer:/export/home level 0] sendbackup: info BACKUP=/usr/local/bin/tar sendbackup: info RECOVER_CMD=/usr/local/bin/tar -f... - sendbackup: info end FINISH driver date 20020225 time 7744.120 = amdump.1 from server = amdump: start at Mon Feb 25 10:38:00 EST 2002 planner: pid 5123 executable /usr/local/libexec/planner version 2.4.2p2 planner: build: VERSION="Amanda-2.4.2p2" planner:BUILT_DATE="Mon Jul 30 12:22:45 EDT 2001" planner:BUILT_MACH="SunOS utl-atl-05 5.7 Generic_106541-15 sun4u sparc SUNW,Ultra-4" planner:CC="gcc" planner: paths: bindir="/usr/local/bin" sbindir="/usr/local/sbin" planner:libexecdir="/usr/local/libexec" mandir="/usr/local/man" planner:AMANDA_TMPDIR="/tmp/amanda" AMANDA_DBGDIR="/tmp/amanda" planner:CONFIG_DIR="/usr/local/etc/amanda" DEV_PREFIX="/dev/dsk/" planner:RDEV_PREFIX="/dev/rdsk/" DUMP="/usr/sbin/ufsdump" planner:RESTORE="/usr/sbin/ufsrestore" planner:SAMBA_CLIENT="/usr/local/samba/bin/smbclient" planner:GNUTAR="/usr/local/bin/tar" planner:COMPRESS_PATH="/usr/local/bin/gzip" planner:UNCOMPRESS_PATH="/usr/local/bin/gzip" planner:MAILER="/usr/bin/mailx" planner:listed_incr_dir="/usr/local/var/amanda/gnutar-lists" planner: defs: DEFAULT_SERVER="utl-atl-06" DEFAULT_CONFIG="DailySet1" planner:DEFAULT_TAPE_SERVER="utl-atl-06" planner:DEFAULT_TAPE_DEVICE="/dev/null" HAVE_MMAP HAVE_SYSVSHM planner:LOCKING=POSIX_FCNTL SETPGRP_VOID DEBUG_CODE planner:AMANDA_DEBUG_DAYS=4 BSD_SECURITY USE_AMANDAHOSTS planner:CLIENT_LOGIN="siteops" FORCE_USERID HAVE_GZIP planner:COMPRESS_SUFFIX=".gz" COMPRESS_FAST_OPT="--fast" planner:COMPRESS_BEST_OPT="--best" UNCOMPRESS_OPT="-dc" planner: dgram_bind: socket bound to 0.0.0.0.603 READING CONF FILES... startup took 0.010 secs SETTING UP FOR ESTIMATES... setting up estimates for rnbuyer:/export/home rnbuyer:/export/home overdue 11744 days for level 0 setup_estimate: rnbuyer:/export/home: command 4, options: last_level -1 next_level0 -11744 level_days 0 getting estimates 0 (0) -1 (-1) -1 (-1) setting up estimates took 0.000 secs GETTING ESTIMATES... driver: pid 5122 executable /usr/local/libexec/driver version 2.4.2p2 driver: send-cmd time 0.005 to taper: START-TAPER 20020225 taper: pid 5124 executable taper version 2.4.2p2 driver: started dumper0 pid 5126 dumper: dgram_bind: socket bound to 0.0.0.0.606 dumper: pid 5126 executable dumper version 2.4.2p2, using port 606 dumper: dgram_bind: socket bound to 0.0.0.0.607 dumper: pid 5127 executable dumper version 2.4.2p2, using port 607 driver: started dumper1 pid 5127 driver: started dumper2 pid 5128 driver: started dumper3 pid 5129 driver: started dumper4 pid 5130 driver: started dumper5 pid 5131 dumper: dgram_bind: socket bound to 0.0.0.0.608 dumper: pid 5128 executable dumper version 2.4.2p2, using port 608 dumper: dgram_bind: socket bound to 0.0.0.0.610 dumper: pid 5130 exec
Re: Broken pipe errors
>A DDS-3 tape should hold 12GB, compressed, right? It looks like "taper" >dies after 10GB. Does the TAPE-ERROR/short write mean the tape is bad? If >the tape was more full, I wouldn't be suspicious; but it seems like I >should be getting more space out of 125m. Amanda consider only the non-compressed size. Compressed size is a commercial only argument, that has NO reality. Do NOT calculate with compressed size. Olivier
Re: Broken pipe errors
>> You don't, by any chance, have both hardware and software compression >> turned on, do you? > >How do I tell, exactly? I have a dumptype line reading > > compress client fast > >But that just seems like which CPU it uses to compute the compression >algorithm. This line says you are using software compression. It also happens to say it's done on the client, but that doesn't really matter. Since you're using software compression, you **must** turn off hardware compression. How you do that is highly OS dependent (it's based on the tape device name for Solaris and AIX, I think it's an "mt" command for Linux, and so on). If you want to use hardware compression, you should turn off software compression with: compress none ... for all the dumptypes. I wouldn't advise going that route unless you're having trouble (e.g. excessive load or wallclock time) with software compression. Software does a better job (since it can look at longer strings of data) and is more flexible (you could turn it off for just certain disks whose data was already compressed, for instance). >> Which size specification? If you mean the one from taper, it's very >> accurate (to the KByte). > >Good. By "the one from taper" I was referring to the "taper: tape /dev/whatever kb NNN fm NNN ..." line in the "NOTES" section of the E-mail (or in the amdump.NN file). All the other Amanda statistics are based on what was *successfully* done. So if the image being processed at the time of the error was large, they could be way off from where the error happened. >Drew John R. Jackson, Technical Software Specialist, [EMAIL PROTECTED]
Re: Broken pipe errors
* "John R. Jackson" <[EMAIL PROTECTED]>: > > I thought DDS-3 was 12 native and 24 compressed. You're right. I had my terminology mixed up. By "12 compressed" I meant *after* the compression had taken effect. Anyway... > You don't, by any chance, have both hardware and software compression > turned on, do you? How do I tell, exactly? I have a dumptype line reading compress client fast But that just seems like which CPU it uses to compute the compression algorithm. > Double compressing things often makes them expand, > i.e. you don't even get the native capacity, and might explain what > you're seeing. That could explain a lot. I inherited this system from a previous admin who had a few gigabytes less data to back-up. Now, after a few months have passed, I'm the one stuck with the tapes filling up. He very well could have been using both compression types and not have known it. > Which size specification? If you mean the one from taper, it's very > accurate (to the KByte). Good. > If you mean the one from the manufacturer, well, let's just say a lot > of those folks won't ever have an employment problem as long as there > are cars (or snake oil :-). We all can't be engineers... :) -- Drew
Re: Broken pipe errors
>A DDS-3 tape should hold 12GB, compressed, right? ... I thought DDS-3 was 12 native and 24 compressed. >It looks like "taper" dies after 10GB. ... Correct. >Does the TAPE-ERROR/short write mean the tape is bad? ... It means taper got an error. That, in turn, might mean any number of things are bad (tape, drive, cleaning, cables, distance to the nearest candy machine, etc). You don't, by any chance, have both hardware and software compression turned on, do you? Double compressing things often makes them expand, i.e. you don't even get the native capacity, and might explain what you're seeing. >How accurate is the size specification of the tape? Which size specification? If you mean the one from taper, it's very accurate (to the KByte). If you mean the one from the manufacturer, well, let's just say a lot of those folks won't ever have an employment problem as long as there are cars (or snake oil :-). >Drew John R. Jackson, Technical Software Specialist, [EMAIL PROTECTED]
Re: Broken pipe errors
> A DDS-3 tape should hold 12GB, compressed, right? It looks like "taper" > dies after 10GB. Does the TAPE-ERROR/short write mean the tape is bad? If > the tape was more full, I wouldn't be suspicious; but it seems like I > should be getting more space out of 125m. DDS-3 holds 12 GB uncompressed. Period. Don't count on what marketing people say. blah, 24 GB, blah... They have no clue of technical details. It depends on the compressibility of your data how much will fit with _either_ hardware or software (gzip/bzip2) compression onto a tape. _Do not_ use both hard and software compression. _It will_ enlarge your data. You get something around 20 GB of the usual data mix onto DDS-3 using either hardware or software compression. Lots of source code, executables, news or mails extend compressibility, lots of mp3's or gzipped files reduce it. It's the amount of entropy in a file being important for reducing it's size. Do you really expect some algorithm being able to reduce an mp3 in size with no loss of information which in turn already has only 1/10th of the original PCM signal _with_ loss of information? - May this loss be detectable with your ears or not. How do you want to compress _random_ data? What similarities do you expect in random data to be able to write them in a short term? I assume you're using hardware compression with gzipped images. Something between 9.5 to 10 GB is here the usual limit on DDS-3. Using software compression makes things clearer for Amanda. She knows how much an image compresses in full and incrementals. This information will be used in planner stage deciding what to backup in what level. The usable tape size value can be very accurate. Hardware compression saves your CPU times, but Amanda doesn't have a clue about the compressibility of each image. So you use an average value for tape size hoping for the best. > How accurate is the size specification of the tape? When I pushed Amanda to the limit by adding more and more hosts or filesystems using software compression, I noticed tapetype's measure being very accurate. Sometimes it didn't fit onto one tape (DDS-2, 4 GB) because it failed to write 10-50 MB. This is a real good guess! You'll have to pay lot's of money to get a measuring instrument working below 1 % error.
Broken pipe errors
I'm having some seemingly elementary errors that I don't quite understand. Using Solaris 2.7, Amanda 2.4.1p1, Sun StorEdge DDS-3 changer on an E450. >From the amdump log: driver: dumping phg:/export/home directly to tape driver: send-cmd time 11090.642 to taper: PORT-WRITE 00-00043 phg /export/home 0 20010630 driver: result time 11090.656 from taper: PORT 41252 driver: send-cmd time 11090.660 to dumper0: PORT-DUMP 01-00044 41252 phg /export/home 0 1970:1:1:0:0:0 DUMP |;bsd-auth;compress-fast;index; driver: state time 11090.660 free kps: 1357 space: 15843292 taper: writing idle-dumpers: 3 qlen tape q: 0 runq: 0 stoppedq: 0 wakeup: 86400 driver-idle: not-idle driver: interface-state time 11090.660 if : free 1 if LE0: free 1 if LOCAL: free 357 driver: hdisk-state time 11090.660 hdisk 0: free 15843292 dumpers 0 taper: writing end marker. [phg-weekly-015 ERR kb 10020128 fm 22] driver: result time 14749.051 from dumper0: FAILED 01-00044 ["data write: Broken pipe"] dumper: kill index command driver: result time 14749.051 from taper: TAPE-ERROR 00-00043 [writing file: short write] driver: QUITTING time 14749.137 telling children to quit driver: send-cmd time 14749.137 to taper: QUIT Broken Pipe amdump: end at Sat Jun 30 05:05:50 CDT 2001 A DDS-3 tape should hold 12GB, compressed, right? It looks like "taper" dies after 10GB. Does the TAPE-ERROR/short write mean the tape is bad? If the tape was more full, I wouldn't be suspicious; but it seems like I should be getting more space out of 125m. How accurate is the size specification of the tape? -- Drew