2.6.x breaks some Berkeley/Sleepycat DB functionality
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
I have encountered a strange issue in 2.6.0 and 2.6.1
I run a PGP Public key server on this machine and under 2.4.x it's "smooth as silk". But if I boot under 2.6.x, it's gaurenteed failure. If I try to build a database using the build command (this is an sks server, so it's sks build or sks fastbuild) I IMMEDIATELY get Bdb error. But the exact same command with the exact same libraries and input files under 2.4.20 works without a hitch.
Anyone got any ideas? Anything else I can provide to assist in debugging?
- --- Dan
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.2 (MingW32)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org
iD8DBQFAAtZctwT22Jak4/4RAk+nAJ4tclsoZTI/a2LAwxb81KOrPxHLhQCcDxoP Dlbr7aZabky+CeBGD9TnjY4=
=dr0q
-----END PGP SIGNATURE-----
Quoted 5 lines · Dan Egli
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1I have encountered a strange issue in 2.6.0 and 2.6.1
Quoted 8 lines
I run a PGP Public key server on this machine and under 2.4.x it's "smooth as silk". But if I boot under 2.6.x, it's gaurenteed failure. If I try to build a database using the build command (this is an sks server, so it's sks build or sks fastbuild) I IMMEDIATELY get Bdb error. But the exact same command with the exact same libraries and input files under 2.4.20 works without a hitch.
Anyone got any ideas? Anything else I can provide to assist in debugging?
this might be the same issue that hit some rpm versions; some versions of db4 seem to detect the O_DIRECT header presence and starts using O_DIRECT, without honoring the alignment requirements linux puts on O_DIRECT usage...
Quoted 8 lines · Dan Egli
I run a PGP Public key server on this machine and under 2.4.x it's "smooth as silk". But if I boot under 2.6.x, it's gaurenteed failure. If I try to build a database using the build command (this is an sks server, so it's sks build or sks fastbuild) I IMMEDIATELY get Bdb error. But the exact same command with the exact same libraries and input files under 2.4.20 works without a hitch.
Anyone got any ideas? Anything else I can provide to assist in debugging?
Off the top of my head, O_DIRECT horkage? I believe that 2.6 has more stringent buffer alignment requirements. A quick 'strace' would probably show if that's the problem.
Quoted 5 lines · Dan Egli
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1I have encountered a strange issue in 2.6.0 and 2.6.1
Quoted 8 lines
I run a PGP Public key server on this machine and under 2.4.x it's "smooth as silk". But if I boot under 2.6.x, it's gaurenteed failure. If I try to build a database using the build command (this is an sks server, so it's sks build or sks fastbuild) I IMMEDIATELY get Bdb error. But the exact same command with the exact same libraries and input files under 2.4.20 works without a hitch.
Anyone got any ideas? Anything else I can provide to assist in debugging?
On top of what file system berkdb is created? I have a reminiscence that Sleepy Cat used to have a problem with reiserfs, due to large stat->st_blksize value. Oleg do you remember this?
Quoted 2 lines
- --- Dan
Nikita.
Hello!
Quoted 10 lines · Nikita Danilov
Quoted 7 lines
I run a PGP Public key server on this machine and under 2.4.x it's "smooth as silk". But if I boot under 2.6.x, it's gaurenteed failure. If I try to build a database using the build command (this is an sks server, so it's sks build or sks fastbuild) I IMMEDIATELY get Bdb error. But the exact same command with the exact same libraries and input files under 2.4.20 works without a hitch.
Anyone got any ideas? Anything else I can provide to assist in debugging?On top of what file system berkdb is created? I have a reminiscence that Sleepy Cat used to have a problem with reiserfs, due to large
stat->st_blksizevalue. Oleg do you remember this?
No, that problem was different. And it was believed that BErkeley DB might just be performing a bit slower on reiserfs, though I never was able to reproduce and Sleepycat's code contains it own limit on maximal block size (it uses 16k blocksize if on suggested by FS is bigger that 16k).
Definitely there were no crashes.
Bye,
Oleg
Quoted 3 lines · Nikita Danilov
On top of what file system berkdb is created? I have a reminiscence that Sleepy Cat used to have a problem with reiserfs, due to large
stat->st_blksizevalue. Oleg do you remember this?
ext3