Re: Why we really need timelines *now* in PITR

Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> writes:

> Anybody see any holes in this design?

God help the DBA who deletes a history file with needed information. Or edits
it inappropriately.

Why can't every log file contain a header that says which timeline it's part
of and which timeline the preceding log file was part of? That would avoid
having a file outside PGDATA and would mean the backups would always contain
enough information to be self-sufficient.

Then if you want to restore from a cold backup and apply PITR up to segment
0020 in timeline X. You read the header of X.0020, find out it followed X.0019
and so on. If X.0010 branched from Y.0009 you'll find out and be able to
continue threading back until you find a segment that matches the current
segment in the cold backup.

The only problem I see is that if your backups are stored on tape it might be
awkward to have to read the headers of all log segments in reverse order to
backtrack to the right place to start.

--
greg