CNFS offset problems

Bettina Fink laura at hydrophil.de
Mon Feb 17 10:08:08 UTC 2003


Russ Allbery <rra at stanford.edu> wrote on 2003-01-31:

> Dave Williams <dave at clues.com> writes:
>> Just to yell out, I'm still seeing this behaviour - a whole bunch of
>> invalid offsets logged for about 1am today and growing CNFS cycbuffs...
> 
>> [I'm not using SEQUENTIAL, the cycbuffs /were/ 2Gb a piece, largefile
>> support is compiled in, code as cvsupd friday morning...]
> 
> Hm.  You've compiled with large file support?  What operating system?  And
> are the cycbuffs *exactly* 2GB in size, or slightly different than that?
> 
> Also, could you send some examples of the log messages?

After some quiet weeks, I'm again seeing problems on my testing
machine:

| Feb 16 17:20:05 nautilus innd: CNFS-sm: CNFSUsedBlock:
| invalid offset 6400200, min = 8000, max = 6400000
|
| Feb 16 17:20:05 nautilus innfeed[27423]: CNFS-sm:
| could not match article size token
| @03004F4E45000000000000031FFF0000004B@ ONE:0x63ffe00:75: 1379

The size of one buffer changed (all buffer used to have the same
size = 102400 / 100 MBytes):

-rw-r--r--    1 news     news     104857600 Feb 17 08:20 five
-rw-r--r--    1 news     news     104857600 Feb 17 06:21 four
-rw-r--r--    1 news     news     104858479 Feb 17 11:00 one  <==
-rw-r--r--    1 news     news     104857600 Feb 17 04:40 three
-rw-r--r--    1 news     news     104857600 Feb 17 01:00 two

This is CURRENT-20030204 on Linux 2.2.23. No large file support.
Buffers are in SEQUENTIAL mode.

If something relating this bug was fixed since CURRENT-20030204,
please drop me a note, then I'll just wipe and upgrade ...

Bye,
Bettina


More information about the inn-workers mailing list