[mb-datafeed] replication packet 4637 is zero length

[Robert Kaye]

On Jul 4, 2006, at 9:28 PM, Robert Kaye wrote:

> Not sure what happened, but replication packet #4637 is zero length  
> and the replication process is... up in the air. I've taken one of  
> the mirrors (catbus))  and had it skip this packet, to see what  
> happens. Please stay tuned.

It turns out that the main web server ran out of disk space while  
creating a data snapshot and packet 4637 was lost.

I have taken two mirror servers and skipped this packet and SO FAR  
nothing has broken. Chances are that at some point something will  
break when a new piece of data coming down the feed expects some  
piece of data to exist that was in packet 4637. I would recommend  
that you skip packet 4637 (see below for instructions) and then plan  
to reimport the MB data at your earliest convenience.

To skip packet 4637, enter the postgres shell and issue this sql query:

    update replication_control set current_replication_sequence = 4637;

IMPORTANT: Only do this if you haven't tinkered with the replication  
in any other way.

For restarting your data stream, start from this snapshot:

    http://ftp.musicbrainz.org/pub/musicbrainz/data/fullexport/ 
20060705-202700/

Dave has taken a number of steps to prevent a future disk space fill- 
up from loosing data packets in the future. These new protections  
will be included in the new server (non schema change) release that  
will be coming up soon.

I sincerely apologize for the trouble that this mishap may cause you.  
Sorry!

--

--ruaok      Somewhere in Texas a village is *still* missing its idiot.

Robert Kaye     --     rob at eorbit.net     --    http://mayhem-chaos.net