"G. Anthony Reina" <reina@nsi.edu> writes:
> ERROR: vacuum: can't destroy lock file!
> NOTICE: AbortTransaction and not in in-progress state
> NOTICE: AbortTransaction and not in in-progress state
> pqReadData() -- backend closed the channel unexpectedly.
Ah-hah! Oliver is right, then, at least in part --- that error message
from vacuum suggests that the vc_abort bug *is* biting you. However,
there may be more going on, because what Oliver and others observed did
not include a coredump (at least not that I heard about).
You can probably suppress the problem by installing the patch I posted
to pgsql-patches a few days ago. However, I'd appreciate it if you'd
first try to reproduce the problem again with debug/assert turned on,
so that we can get some idea whether there is an additional bug that's
only biting you and not the previous reporters.
BTW, if vc_abort is involved then the occurrence of the problem probably
depends on whether any other backends are using the database and what
they are doing. (The vc_abort bug escaped notice up till last week
because it doesn't trigger when vacuum is the only thing running.)
Do you have other clients running when you do the vacuum? What are
they doing?
regards, tom lane