Stephan Szabo <sszabo@megazone23.bigpanda.com> writes:
> The question is why did it get confused and end up linking to a filename
> that already existed?
The message comes from InstallXLogFileSegment(), which is careful to
ensure that the link() cannot fail, either by unlinking the previous
file, or searching for an unused name. But it failed anyway.
It seems to me that there are only two possible explanations: a race
condition (but holding ControlFileLock should prevent that) or
BasicOpenFile() failed for a reason other than nonexistence of the file.
Hmm ... I wonder if Noah's machine could have been running out of kernel
file table slots, or something like that? It does seem that it'd be
more robust to use something like stat(2) to probe for an existing file.
regards, tom lane