Bug 1245854

Created attachment 1055126 [details]
File: anaconda-tb

Created attachment 1055127 [details]
File: anaconda.log

Created attachment 1055128 [details]
File: dnf.log

Created attachment 1055129 [details]
File: environ

Created attachment 1055130 [details]
File: lsblk_output

Created attachment 1055131 [details]
File: nmcli_dev_list

Created attachment 1055132 [details]
File: os_info

Created attachment 1055133 [details]
File: program.log

Created attachment 1055134 [details]
File: storage.log

Created attachment 1055135 [details]
File: syslog

Created attachment 1055136 [details]
File: ifcfg.log

Created attachment 1055137 [details]
File: packaging.log

This is failing somewhere in this block in PartitionSetFactory.configure():

        log.debug("adding a %s with size %s",
                  self.parent_factory.size_set_class.__name__, total_space)
        size_set = self.parent_factory.size_set_class(members, total_space)
        self.storage.size_sets.append(size_set)
        for member in members[:]:
            member = member.raw_device
            member.req_max_size = size_set.size

custom.py _destroy_device() creates an LVMFactory which gets a child PartitionSetFactory, and it's while that child factory is getting configured that we hit the crash. We hit the "adding a %s with size %s" log.debug, but we don't hit a log.debug added just before the next step - so this seems to be where it explodes. Looking into it further.

Hum, scratch that...must've messed up an updates.img. I found a better traceback in the anaconda-tb now:

21:21:35,652 CRIT anaconda: Traceback (most recent call last):

  File "/tmp/updates/blivet/devicefactory.py", line 794, in configure
    self._configure()

  File "/tmp/updates/blivet/devicefactory.py", line 1344, in _configure
    super(LVMFactory, self)._configure()

  File "/tmp/updates/blivet/devicefactory.py", line 818, in _configure
    self.child_factory.configure()

  File "/tmp/updates/blivet/devicefactory.py", line 1140, in configure
    self._post_create()

  File "/tmp/updates/blivet/devicefactory.py", line 933, in _post_create
    doPartitioning(self.storage)

  File "/tmp/updates/blivet/partitioning.py", line 550, in doPartitioning
    growPartitions(disks, partitions, free, size_sets=storage.size_sets)

  File "/tmp/updates/blivet/partitioning.py", line 1694, in growPartitions
    manageSizeSets(size_sets, chunks)

  File "/tmp/updates/blivet/partitioning.py", line 1598, in manageSizeSets
    chunk.reclaim(request, extra)

  File "/tmp/updates/blivet/partitioning.py", line 1077, in reclaim
    log.debug("reclaim: %s %d (%s)", request, amount, self.lengthToSize(amount))

  File "/tmp/updates/blivet/partitioning.py", line 1358, in lengthToSize
    return sectorsToSize(length, self.sectorSize)

  File "/tmp/updates/blivet/partitioning.py", line 298, in sectorsToSize
    return Size(sectors * sectorSize)

  File "/tmp/updates/blivet/size.py", line 316, in __mul__
    return Size(Decimal.__mul__(self, other))

  File "/tmp/updates/blivet/size.py", line 277, in __new__
    raise ValueError("invalid value %s for size" % value)

ValueError: invalid value NotImplemented for size

OK, so the problem is that when lengthToSize calls 'sectorsToSize(length, self.sectorSize)', length is a float. Here's a simple reproducer:

from blivet.size import Size
fl = float(1)
sz = Size(1)
Size(fl*sz)

That raises an exception, and that's basically what's happening here. sectorsToSize gets hit a whole lot and almost always the type of the first param is int, and that works - but just on this path the type of the first param is float and it explodes.

Now I get to figure out why the hell length is a float! This is fun, if by 'fun' you mean 'no fun at all'.

Aha. OK. I got it. It's a py2 vs. py3 thing. The problem is here, in manageSizeSets:

                        extra = -chunk.sizeToLength(needed) / len(ss.devices)

in context, -chunk.sizeToLength is an int - in my test case it's 2162163712 - and so, obviously, will be len(ss.devices). Now, observe:

[adamw@adam anaconda (master *%)]$ python2
Python 2.7.10 (default, Jun 29 2015, 16:02:04) 
[GCC 5.1.1 20150618 (Red Hat 5.1.1-4)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> type(2162163712/2)
<type 'int'>
>>> 
[adamw@adam anaconda (master *%)]$ python3
Python 3.4.3 (default, Jun 29 2015, 12:16:01) 
[GCC 5.1.1 20150618 (Red Hat 5.1.1-4)] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> type(2162163712/2)
<class 'float'>
>>> 

i.e. in python2, 'int / int' returns an int. In python3, 'int / int' returns a float.

I think the fix for this should be as simple as:

-                        extra = -chunk.sizeToLength(needed) / len(ss.devices)
+                        extra = -chunk.sizeToLength(needed) // len(ss.devices)

*** Bug 1245446 has been marked as a duplicate of this bug. ***

https://github.com/rhinstaller/blivet/pull/177

Merged by dlehman, will be fixed in next blivet.

This is indeed fixed in TC2, openQA and manual testing confirm it.