Обсуждение: AW: backup and restore

> Well, that would only be part og what I'm looking for. The thing I like about 
> informix is that I can make a Level 0 backup of all the data (equal to the 
> pg_dumpall), and then leave the logical logs downloading continuosly, so that 
> if in one moment the system breaks, I restore the Level 0 backup and then 
> apply the logical logs, which are the small changes that have been done to 
> the database in each transaction, administration, etc.
> 
> Could this be added? I am willing to help with the coding.

This is what Version 7.1 WAL is all about. 
There might be some help wanted in one of the possible backup methods:1. a pg_dumpall restore, and a subsequent restore
oflogs2. a restore of a "physical backup of db files" + subsequent restore of logs
 

I think Vadim has the 1st way in his works.
The 2nd way would need some work and testing, and probably some utility to 
backup the files in the correct order 
(Could be something that calls tar with appropriate arguments).
I am still pretty sure that a physical backup without synchronization with the
postmaster is possible with a little extra work. E.g. checking index validity
after restore and rebuild if bogus. The better way would probably be to not backup 
index files at all and rebuild them after restore.

A distinct suffix for different file types would definitely help in this area (.dat, .idx, .tmp ...).
I think this would be a good idea overall. 

Andreas

At 16:21 12/10/00 +0200, Zeugswetter Andreas SB wrote:
>> 
>> Could this be added? I am willing to help with the coding.
>
>This is what Version 7.1 WAL is all about. 
>There might be some help wanted in one of the possible backup methods:
>    1. a pg_dumpall restore, and a subsequent restore of logs
>    2. a restore of a "physical backup of db files" + subsequent restore of logs
>
>I think Vadim has the 1st way in his works.

My guess is that there are some issues to resolve: I presume the WAL uses
OID to identify rows, which will mean that BLOB restoration won't work
because it changes the OID of the BLOB. Also, pg_dump will restore row OIDs
for table data, but not the OID for such things as sequences. Guessing
again, I'd say that sequence updates in the WAL will also be OID based.

It's definitely true that the WAL is an essential first step to proper
backup, but there's probably the need to write a backup utility as well.
Unless of course Vadim has done that as well...


----------------------------------------------------------------
Philip Warner                    |     __---_____
Albatross Consulting Pty. Ltd.   |----/       -  \
(A.B.N. 75 008 659 498)          |          /(@)   ______---_
Tel: (+61) 0500 83 82 81         |                 _________  \
Fax: (+61) 0500 83 82 82         |                 ___________ |
Http://www.rhyme.com.au          |                /           \|                                |    --________--
PGP key available upon request,  |  /
and from pgp5.ai.mit.edu:11371   |/