[ADMIN] Question about PITR backup

[Kevin Grittner]

I have a question about PITR backup in a single server, the method
is make a base backup, and backup the WAL archive log(eg, every
day at 11:30 pm). But if the OS' harddisk is broken(eg,14:00
pm),the system can't start, we have to recover the database on a
another server using the base backup and the WAL backup,so in this
case suppose that we only can restore the database to yesterday's
11:30 pm state, am I right?

You can't recover to a point past your last available WAL record.

I want to backup the pg_xlog folder every minute by crontab but my
concern is that the data inconsistent, because the lastest log in
pg_xlog is being updated all the time, am I right? Any
suggestions?

Use streaming replication? Or at the very least, set a short
archive_timeout value and copy from the archive target location
frequently.

-Kevin

[René Romero Benavides]

Dear admin,
I have a question about PITR backup in a single server, the method is
make a base backup, and backup the WAL archive log(eg, every day at
11:30 pm). But if the OS' harddisk is broken(eg,14:00 pm),the system
can't start, we have to recover the database on a another server using
the base backup and the WAL backup,so in this case suppose that we
only can restore the database to yesterday's 11:30 pm state, am I right?
How can we recover the database to 14:00 pm? We are missing the newest
pg_xlog.
I want to backup the pg_xlog folder every minute by crontab but my
concern is that the data inconsistent, because the lastest log in
pg_xlog is being updated all the time, am I right? Any suggestions?

Yes, the wal segments are updated constantly. You'll have a consistent
state in your data files up to the time when the last checkpoint
happened. Should a crash occur the server will replay the log segments
in the WAL files starting from the one that was marked during the last
checkpoint recorded.
So,provided all the WAL files generated by the server after the last
checkpoint are available, it will always be able to bring the data
files to a consistent state. Those WAL files are always kept in the
pg_xlog directory.

Though I'm also in favor of leveraging a streaming replication
configuration as it's more reliable and the window for data loss is very
small.