Description of problem: Anaconda dies (early on) when trying to install to an external USB drive. Version-Release number of selected component (if applicable): 11.4.0.27-1 (2008-01-25) How reproducible: 100% - tested on two very different machines Steps to Reproduce: 1. Attempt NFS install from rawhide with USB drive plugged in 2. 3. Actual results: Expected results: Additional info: The drive has a single Linux partition already on it (with a local copy of rawhide!). Just booting up the installer dies immediately with basically no notice other than "I quit" There are no discernible errors Note: this may be a problem with the bdevid modules stuff as that's the last message in the anaconda.log and no such module seems to be loaded?
Created attachment 293165 [details] Failure logs from two different installs
Are there any helpful error messages on tty1 or tty4?
The attachment contains pertinent logs and other information from the two systems I tried. anaconda.log syslog partitions (copied from /proc/partitions) modules (output of 'lsmod') What else do you need to see?
Also fails on stock Fedora8 (11.3.0.50). Tried on PS3 and i386. At least on the i386, anaconda fails with python throwing a glibc/double free() error. Strangely enough, I was able to install Fedora8 to an i386 system _once_, but I can no longer repeat this.
I've found where this happens. In dmraid.py: def scanForMPath(drives): log.debug("scanning for multipath on drives %s" % (drives,)) mpaths = [] probeDrives = [] for d in drives: probeDrives.append("/dev/%s" % (d,)) import block as _block oldPath = _block.getBdevidPath() if not _bdModulePath in oldPath: _block.setBdevidPath(_bdModulePath + oldPath) del oldPath log.debug("loading bdevid modules from: '%s'" % (_block.getBdevidPath(),)) _block.load("scsi") mpaths = _block.getMPaths(probeDrives) log.debug("mpaths: %s" % (mpaths,)) The call to _block.getMPaths(probeDrives) fails, but the exception doesn't seem to be caught and everything goes downhill from there. If I short-circuit scanForMPath() to just return an empty list, the remainder of the installation carries on. n.b. I've not run this to completion yet, so I don't know what will happen with the bootstrap setup, etc., but at least this is progress toward fixing this bug. Also, I've been testing this against the latest rawhide version 11.4.0.33
Can you attach the /tmp/anacdump.txt file from this failure? That's the most important piece of debugging information we've got for fixing up things like this.
Already attached for two different machines (comment #1) Note that I have now isolated the failure to the call getMPaths(), but I don't know where that code is (it's not part of anaconda proper) or else I'd continue my investigation
You included /tmp/anaconda.log, true. However on tracebacks, you additionally get /tmp/anacdump.txt which includes not only the log, but the traceback itself and the dump of a bunch of internal variables.
Anaconda dies without generating a dump file.
Have you been able to reproduce with recent rawhide? Just want to make sure it's still a problem.
Installing to USB disks here work, I'm dropping this off the blocker list.
Tested against the Beta LiveCD worked for me.
Ok, closing.