Jan Wieck wrote:
> On 7/11/2006 1:36 PM, Richard Yen wrote:
>   
>> Hi,
>>
>> I recently did a MOVE SET switchover from node 1 to node 3 on a 3- 
>> node system.  So, all that remains are nodes 2 and 3, while we run  
>> hardware tests on node 1.
>>
>> What I'm trying to do now is to add a node 4 to serve as a relay to  
>> an offsite node.
>>
>> However, when I try to run a slonik script, I get the following error  
>> message:
>> Error: No admin conninfo provided for node 1
>>
>> Since node 1 is inactive (and no longer in the sl_nodes, sl_path,  
>> etc. tables), how can I run a slonik script?
>>     
>
> Add "event node = 3" to the commands that fail. The default event node is 1.
>   
Node 2 could probably also work, but yeah, 3's probably preferred...
> Also, are you going to have node 1 offline for so long that it warrants 
> a complete rebuild as opposed to a catch up? If the DB isn't heavily 
> updated, it might have been possible to just take it offline, let it 
> later catch up and switch back.
>   
Good point...

Let me point out the "counter"...  I recently had a test node in a
cluster that got into an unfortunate state, such that it was down for ~4
days.

I never could get that node to catch up, I believe because it kept
reading in 4 days worth of events, where there was some query trying to
read *all* the events that was timing out...

[1]> (* 4 24 360)
34560

Falling behind by a day seems to be untroublesome, but falling behind by
a sufficient volume of SYNCs becomes prohibitive, even if most of them
are virtually empty.  At what point you get to "too far behind" seems to
be dependent on volume of updates.  A few users with HEAVY updates have
reported that getting 100 SYNCs behind will "kill" replication, but
clearly those with lesser volume can survive far more.

It seems to me that there are some queries in Slony-I that represent
inherent bottlenecks; we should probably take a look at that as a
post-1.2.0 action...