>> I've been running into problems with Linux totally >> locking up during extremely large updates. This leaves >> the state of the update indeterminate and the state of >> RPM's database a complete mess. > >Although it would be good to be able to recover from >this situation we >should also be figuring out what is causing the lock >ups in the first >place. > >Can you provide more details about these lockups... I think I've pinned the problem down to a memory leak. If I run a moderately large Yum update, memory doesn't get freed up properly in the kernel. Subsequent Yum updates - even if small - will likely trigger the OOM killer. For very large updates (the magic number for my system is 650), memory is completely exhausted. The OOM killer is ineffective, as most of the memory consumed has leaked out the ears of the kernel and isn't associated with anything. >How large are the updates you're talking about? 650+ items (so normally about 1300+ events once the updates themselves begin) >What distro, RPM version, Yum version? Red Hat Fedora Core 4 (plus stuff from Fedora Extras, no other RPM repository) with pretty much everything from the development section, so it's essentially Core 5 pre 3. RPM is 4.4.2-15.1 YUM is 2.6.0-1 Kernel is 2.6.15-1.2032 >How often do the lockups occur? Any time the updates exceed 650 items. >Is it just the Yum process that freezes up or the >whole system? >Is the system running out of RAM/trashing during the >"lockup"? It's a total system lockup. The keyboard dies, so can't magic sysrq in. The system leaks memory, so when it runs out, there's nothing it can do. I do not believe that the memory leak is actually Yum-related - it seems to be more associated with recent kernels. On rebooting, the RPM database is usually left with spurious entries. Once upon a time, I'd clean those out with the --justdb flag, but I think Yum now does do cleanup. At least, I'm seeing lots of extra messages when it reaches such entries indicating that it is, so I'm assuming that's now taken care of. >Dependency breakages shouldn't happen. That's the >whole point of Yum >and >RPM. I'm guessing you're seeing packages that have >incorrect or >incomplete dependencies meaning that Yum can't be >expected to magically >fix things. In this case the packages need to be >fixed. Can you be more >specific and give an example? In some cases, you're absolutely correct - the dependency header in the file has been mangled. I can give an excellent example of that right now: Error: Missing Dependency: lib765=0.3.3-5.fc5 is needed by package lib765-devel There seem to be a number of fat-fingered dependencies of this kind, where - has been mistyped as =. Yum can't be expected to handle such cases. (Well, I suppose Yum could use the Apache fuzzy spelling module...) Jonathan __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com