[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: unable to parse package file; aptitude effectively dead.



On Thu, Jan 04, 2007 at 02:52:53PM -0500, Douglas Tutty wrote:
> On Thu, Jan 04, 2007 at 01:09:56PM -0500, hendrik@topoi.pooq.com wrote:
> > On Thu, Jan 04, 2007 at 11:49:39AM -0500, Joey Hess wrote:
> > > hendrik@topoi.pooq.com wrote:
> > > > While I'm waiting for someone to *fix* bug 405506, does anyone have an 
> > > > idea how to recover from it?  Is a reinstall in order?
> > > > 
> > > > Yesterday aptitude reported
> > > >   Unable to parse package file /var/lib/aptitude/pkgstates
> > > > This happened on an etch that was just installed the same day from a 
> > > > previous-day netinstall daily build.
> > > > 
> > > > The file does exist, and appears to be binary gibberish.
> > > 
> > > In the bug report, you say:
> > > 
> > > | tried startx, got the black screen of death -- completely unresponsive
> > > | to mouse or keyboard.
> > > 
> > > Hard rebooting a wedged system can result in corrupted files in rare
> > > cases, and it sounds like this is what you have. The pkgstates file is a
> > > plain text file. The chances of this being an aptitude bug seem very
> > > near to zero. Unless you or someone else can reproduce the bug (either
> > > the hang starting X or the file system corruption), it's certianly not
> > > grave severity.
> > 
> > It is starting to look as if it was a problem with jfs, which, I was 
> > told, was specifically engineered to prevent this kind of trouble.
> > Certainly, if it was an aptitude bug, it would have been grave.  But 
> > since aptitude seems not to blame, I think we should consider it solved 
> > for aptitude.  Perhaps it should be referred to jfs?  I'll leave the 
> > decision up to you.  I don't think there are enough specifics that this 
> > bug would help the jfs maintainers to find the problem, so there may be 
> > no point.
>  
> > I had used jfs as file system because reports on this mailing list 
> > indicated that it was more resistant to this kind of thing than either 
> > ext3 or reiser.  But I've never had problems of this sort with either of 
> > them, and had problems with jfs the first day I used it, so maybe in 
> > the reinstall I should go back to reiser of ext3.
> > 
> > 
> Hi Hendrik,
> 
> I missed the start of this thread but I use JFS (and it may have been my
> comments that prompted you to use it).

It may indeed.

> 
> Was this file open or being used when the system crashed or lost power?

I suppose it's possible that I might have left the interactive aptitude 
running while I did the startx.  But it wasn't doing anything, and it 
would surprise me if aptitude left it open for writing.  Though I 
suppose jfs might haver delayed writing the buffers out after the close 
for reasons of its own.

> If so, I don't think there's any filesystem that can protect an
> individual file in that case.

Journalling is supposed to be able to protect data -- not that it 
necessarily protects everything that was written to the file, but that 
it can guarantee leaving the file in a consistent state.

>  I wish there were.  The problem I had
> with other filesystems is that after such a power loss, the filesystem
> itself would be corrupted and I'd lose data during recovery.
> 
> Its also possible that whatever caused the system crash overwrote the
> files' buffers which then got dutifully committed to disk.  This would
> not be a problem with JFS itself.

No.  But it could possibly be blamed on the monolithic design of the 
Linux kernel, which leaves it impractical to assign responsibility to a 
particular software package.   Not that that's likely to be fixed 
anytime soon.  Maybe I shoudl try hurd one of these years?  Is it 
different?

>  
> I'm also a little leary of X.  Since by its nature it takes over one's
                    leery
> video hardware, if it crashes it can leave you with not console.  I tend
> to leave a getty runing on a serial port for just such emergencies (grab
> a computer or terminal, a serial null-modem cable and away you go).  

One of the things they did right, though, was the client-server 
separation.  You can sometimes isolate bugs by running the client and 
the server on different machines.

> 
> I'l be watching to see what other JFS issues appear from this.

I suspect I'll be wiping the evidence soon, since I see no practical way 
of using it to identify the problem.  And reinstalling seems to be the 
easy solution.


> 
> Good luck,
> 
> Doug.
> 
> 
> -- 
> To UNSUBSCRIBE, email to debian-user-REQUEST@lists.debian.org 
> with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
> 



Reply to: