Those per se are not that annoying, but the random messages generated
by the backend while pg_arman is waiting for segments to be archived
are annoying. This commit gets rid of them by using client_min_messages
set to warning on the connection used.
Report by Yury Zhuravlev.
If the target path is too long, an error needs to be emitted as well.
The buffer is correctly null-terminated, this will just avoid running
into weird problems should the target have a too long name.
IsDir was somewhat optimized on systems where DT_DIR (dirent.d_type
being not part of the POSIX spec) is present on some systems leading
to more complex logic depending on the file system used, particularly
on XFS this routine was actually broken. Having a call to stat()
should not be that expensive to check if a path is a directory or
not and this is proving to be far more stable coding, so just rely
on that, that's more portable anyway and will avoid future surprises.
Report by Yury Zhuravlev, though I did not use his patch.
This restriction has been inherited from the upstream project pg_rman, however
the new differential facility is proving to be able to address that need quite
nicely after more testing regarding that.
Mentioned by Thomas Reiss.
All the ERROR_* fields are removed in favor of a more simple layer
made of ERROR, FATAL, PANIC. The two last ones are not actually used
yet, thought there should be some code paths that would need more
polishing on this matter.
The error message emitted before leaving should be fine to let the
user know what is happening.
During the execution of the "delete" command, the oldest start LSN is saved
and used to determine what is the oldest WAL segment that needs to be kept
for the existing backups. Like pg_archivecleanup, the implemented logic
ignores the timeline part of the WAL segment name, ensuring that a segment
from a parent timeline will not be removed prematurely. This logic could
be made more complicated regarding this matter, but in order to take backups
from a node on a given timeline this is far enough.
This commit improves the performance of page-level, or differential
backup, by not having to scan anymore all the pages of a relation file,
something that can be very long on large data sets, but by scanning the
list of blocks changed by WAL records since the last full or differential
backup.
As a restriction and to avoid potential data corruption should hint-bit
updates occur on a page, backups can only be taken from a server that has
wal_log_hints or data checksums enabled.
Base patch by Yury Zhuravlev, heavily modified by me.
