[mb-devel] Database replication?

[Steve Wyles]

On Sat, 22 Jul 2006, Elliot Chance wrote:

> Ahhh thats makes sense. Few more questions
>
> 1. So in the Pending, the first id matches with the data in PendingData, but
> does that id have any bearing on the table?

Taking Pending first

Rowid
tablename the changes are for
type (u,d,i)
Transaction sequence

Now PendingData

rowid
boolean IsKey
packed data in key/value pairs (the code in dbmirror.pm deals with this)

>
> 2. What is the second id in Pending for?
>
The transaction sequence. For any changes to the data on the main server a 
begin transaction, process changes, commit transaction procedure is 
followed. The transaction number is incremented each time a new 
transaction is started. Within a transaction there can be any number of 
related DB changes. If one change fails, the whole transaction up to that 
point is reversed.

An overview of the replication packet processing is:

download and untar the next replication packet

check the replication sequence and database schema version

load the Pending and PendingData files into the Pending and PendingData 
db tables

process through the data by transaction id taking each record within the 
transaction.

Delete all records for that transaction from Pending/PendingData (this is 
so if replication fails, it can restart from the point it left off)

get next transaction and associated records

when no more tranactions are left, see if another replication packet is 
available. If not, exit

> 3. In PendingData, what does the 't' and 'f' represent?
>

true and false whether it refers to a db key, it is used to order the 
select of the records from PendingData

Steve