06-05-08
--------
- DB base backup
- Bug fix

DB backup
---------

1. Snapshot works on hmidb. 

I used it to prepare a base backup on the db. The scheduled db
downtime on May 28 went well. Because of the snapshot utility, the
downtime is only a few minutes. And that's for me to change the
location of WAL archive (from d/ to c/). Future base backup does not
require database to be shutdown. Two things to archive: 

. data directory
. the WALs generated between pg_start_backup() and pg_end_backup() 

These two together can bring up a consistent database.

2. Wrote a step-by-step instruction for backup/restore. To to presented.

3. Tested the sanity of the base backup.

I ran into a series of problems due to lack of experience, which put
the database (not the production db) into all sorts of precarious
states, including missing checkpoints. Learned to use pg_controldata
to check the state of the database. A few things to watch out:

. pg_xlog/ must be cleaned
. new WAL archive location, better not be the same as the WAL files
  saved from base backup
. recovery.conf must exists.

Lesson learned: make plans to do the recovery second or third time in
case something goes wrong.

Currently working on

- script to do base backup. In addition to snapshot and writing to
  tape, this involves psql and some parsing of the db backup file to
  generate a list of WALs to archive.
- reading on warm standby

Bug fix
-------

Fixed ug in drms_record.c::ParseRecSetDesc() where an extra ']' was added
when parsing [??] clause. This prevented multiple [??] clauses in the
dataset name.