Sqlite hangs on fsync on Hammer partition
I've a little program that uses the Sqlite database. After it calls
seven times fsync (sometimes only twice), the whole process blocks. The
process cannot be killed (-9) anymore, and when I change into the
directory where I executed the program and do a "ls", I'm blocked as
When I execute the same program on a UFS partition, it doesn't hang. So
it seems to be a Hammer-related issue.
According to ktrace/kdump, it hangs on fsync.
I'm now creating a core-dump and uploading it.
#5 Updated by dillon about 9 years ago
: Michael, that vmcore.4 on leaf is not a patched kernel. I disassembled
: one of the routines that was patched and it didn't have the extra flag
: I had added.
: In anycase, I am MFCing all the HAMMER work to 2.0 today.
Wait, I take it back. I was looking at the wrong bit of code.
It does have the patch.
Ok, I'm looking at the core!
#6 Updated by dillon about 9 years ago
Ok, please try this patch folks:
I think that catches all the cases. If not, Michael, please upload
another core to leaf.
I will commit this and MFC the whole mess to 2.0 a bit later today.
#9 Updated by dillon about 9 years ago
Ok, try this patch Gergo (relative to HEAD):
This is the patch relative to an unpatched 2.0:
#10 Updated by bastyaelvtars about 9 years ago
On Sat, 2 Aug 2008 10:27:04 -0700 (PDT)
Matthew Dillon <email@example.com> wrote:
OK, it seems to work now, at least I could not make it hang with the
above command, even though I am creating hundreds of databases
sequentially. I have even tried in two separate volumes.
Thanks for the fix!