Jon Culverson posted to -admin on 2 June with a PITR bug that seems to
go back at least 2 releases, i.e. it occurs in 8.2 and 8.1 current
production releases as well as CVS HEAD.
The issue is that some archive_status files are left in an invalid
state, when the last recovered file is from the archive rather than the
local directory. The archive_status files show that a file that came
from the archive is ready to be archived (again). This causes archiving
to fail on a newly recovered server if you have an archive_command that
prevents duplicate inserts into the archive and you watch the logs long
enough to notice the failure, which can occur some seconds after the
server is operational. [That precise timing window is related to a
second bug report by Kevin Grittner on -hackers on 22 June.]
Recovery itself does not have a bug, so there is no need to take
additional backups as a result of this. The workaround is to set
archive_command to a command that allows duplicate inserts.
Culverson's bug has been patched, as of 8 June and confirmed resolved,
but has not been applied. Please review the patch and apply/reject.
Grittner's bug has not yet been patched by me, expected soon.
--
Simon Riggs
EnterpriseDB http://www.enterprisedb.com