This is a PackageKit bug, from the above:

 File : /usr/share/PackageKit/helpers/yum/yumBackend.py, line 1434, in
update_packages
    txmbr = self.yumbase.update(po=pkg)

...the lines leading upto this are:

                pkg, inst = self._findPackage(package)
                if pkg:
                    txmbr = self.yumbase.update(po=pkg)

...here inst means "this package is installed" ... and you aren't checking it so you can be passing available packages to the update() method and then, yeh, update() kind of assumes the po is installed if you pass a po.

the original code i did was:

        pkg,inst = self._findPackage(package)
        if pkg:
            txmbr = self.yumbase.update(name=pkg.name)

Then YumBase.update will do the right thing

it was broken in 
9e2e8562f85f8234567dc6c6a1226e2e4e41a5ac

diff --git a/backends/yum/helpers/yumBackend.py b/backends/yum/helpers/yumBackend.py
index d70d8dc..bdd70e5 100644
--- a/backends/yum/helpers/yumBackend.py
+++ b/backends/yum/helpers/yumBackend.py
@@ -675,7 +675,7 @@ class PackageKitYumBackend(PackageKitBaseBackend):
         # FIXME: This is a hack, it simulates a removal of the
         # package and return the transaction
         if inst and pkg:
-            txmbrs = self.yumbase.remove(name=pkg.name)
+            txmbrs = self.yumbase.remove(po=pkg)
             if txmbrs:
                 rc,msgs =  self.yumbase.buildTransaction()
                 if rc !=2:
@@ -903,7 +903,7 @@ class PackageKitYumBackend(PackageKitBaseBackend):
                 if not already_warned and not repo.gpgcheck:
                     self.message(MESSAGE_WARNING,"The untrusted package %s will be installed from %s." % (pkg.name, repo))
                     already_warned = True
-                txmbr = self.yumbase.install(name=pkg.name)
+                txmbr = self.yumbase.install(po=pkg)
                 txmbrs.extend(txmbr)
             if inst:
                 self.error(ERROR_PACKAGE_ALREADY_INSTALLED,"The package %s is already installed", pkg.name)
@@ -1038,7 +1038,7 @@ class PackageKitYumBackend(PackageKitBaseBackend):
             for package in packages:
                 pkg,inst = self._findPackage(package)
                 if pkg:
-                    txmbr = self.yumbase.update(name=pkg.name)
+                    txmbr = self.yumbase.update(po=pkg)
                     txmbrs.extend(txmbr)
         except yum.Errors.RepoError,e:
             self.error(ERROR_REPO_NOT_AVAILABLE,str(e))
@@ -1135,7 +1135,7 @@ class PackageKitYumBackend(PackageKitBaseBackend):
         pkg,inst = self._findPackage(package)
         if pkg and inst:
             try:
-                txmbr = self.yumbase.remove(name=pkg.name)
+                txmbr = self.yumbase.remove(po=pkg)
             except yum.Errors.RepoError,e:
                 self.error(ERROR_REPO_NOT_AVAILABLE,str(e))

this was ok, in the YumBase.install and YumBase.remove, but not in YumBase.update

Good find Tim. The reason I changed it from:

-                    txmbr = self.yumbase.update(name=pkg.name)
+                    txmbr = self.yumbase.update(po=pkg)

Is that we need to match on name and version, as we could have multiple versions of updates in a repo, and we need to choose the correct one, not just the latest.

What about something like this:

-                    txmbr = self.yumbase.update(po=pkg)
+                    txmbr = self.yumbase.update(name=pkg.name, epoch=pkg.epoch, version=pkg.version, release=pkg.release, arch=pkg.arch)

That would give us the exact matching and not the exact po match.

This might do it, but i'm not sure

But after looking at the YumBase.update code, i can see it work, with both YumBase.update(po=avail_po) & YumBase.update(po=inst_po).

(YumBase.update)

        instpkgs = []
        availpkgs = []
        if po: # just a po
            if po.repoid == 'installed':
                instpkgs.append(po)
            else:
                availpkgs.append(po)

The part of the code there is failing, is the one there handles available packages.

        for available_pkg in availpkgs:
            for updated in self.up.updating_dict.get(available_pkg.pkgtup, []):
                if self.tsInfo.isObsoleted(updated):
                ..

                ..

                else:
                    updated_pkg =  self.rpmdb.searchPkgTuple(updated)[0]
  
The problem here is that self.rpmdb.searchPkgTuple(updated) returns []
i looks very weird, 
updated is the pkgtup for the package to be updated and the error indicates that that package dont exits in the rpmdb.
There must be some inconsistency somewhere, something is not in sync.

Matej, can you reproduce this or was it a one-off?

Yeah, that was one-off and it probably relates to the general brokeness of rpm (I have to rm -f /var/lib/rpm/__db* all the time to make rpm work). Also, the problem is that PK when rpm is broken just sits there doing nothing (or what).

you have to clean up your rpmdb all the time? Are you running a special kernel, bdb or glibc?

(In reply to comment #8)
> you have to clean up your rpmdb all the time? Are you running a special kernel,
> bdb or glibc?

On all my computers (all Rawhide, two i386 notebooks -- one mine, other my wife's, other is x86_64 destop at the office) quite often I get "Thread <uid> broken" or something like that, and the only medicine I found for it was

rm -f /var/lib/rpm/__db ; rpm --rebuilddb

And no, all of them have standard glibc, rpm, kernel from distribution and fully upgraded (i.e., I always run yum upgrade without any exclusion except for occassional exclusion of openoffice.org for saving the bandwidth).

of course, it should be

rm -f /var/lib/rpm/__db* ; rpm --rebuilddb

instead

This sounds to me like bug 468437... I suspect PackageKit might be going wild with threads, but librpm is not thread-safe, all access to rpmdb should happen in the same thread that opened the db.

(In reply to comment #11)
> This sounds to me like bug 468437... I suspect PackageKit might be going wild
> with threads, but librpm is not thread-safe, all access to rpmdb should happen
> in the same thread that opened the db.

yeah, that's it. Will provide backtrace, when it happens again.

PackageKit spawns a new process to talk to yum using stdin and stdout as IPC. It does not use threads at all when using a spawned backend like yum.

This bug appears to have been reported against 'rawhide' during the Fedora 10 development cycle.
Changing version to '10'.

More information and reason for this action is here:
http://fedoraproject.org/wiki/BugZappers/HouseKeeping

Matej, do you still get this problem?

I don't have PackageKit on F10 anymore (doesn't work and shouldn't work with staff_u SELinux user). Closing as INSUFFICIENT_DATA.