Re: [luciddb-users] Horizontal Partitioning Across Distributed Servers

[Nicholas G. <ngo...@dy...>]

On Mar 30, 2011, at 1:34 PM, Jeremy Lemaire wrote:

> Questions
> 
> 1. Is there a workaround for this other than not using Java
> PreparedStatement with parameters?
It might be related to the UNION ALL portions and the PreparedStatement.  Not JUST the PS and remote server by itself.

What does the Explain Plan for: 
SELECT SUM("COUNT") as "COUNT", f.npa, f.filled, f."LANGUAGE" 
	FROM "INVENTORY_ANALYZER_SCHEMA"."INVENTORY_BY_LANGUAGE_2011_Q1" f 
		LEFT JOIN ad_inventory_warehouse.publisher_dimension pub USING (source_id) 
			WHERE f.source_id IN (?) AND 
				pub.publisher_id IN (?) AND 
				f.DATETIME >= timestamp '2011-03-29 00:00:00' AND 
				f.DATETIME <= timestamp '2011-03-29 23:59:59' 
GROUP BY f.npa, f.filled, f."LANGUAGE"
When run as a Prepared Statement?
Also, if you can submit the Explain plans (which will have the information on the remote SQL being executed (with implementation) for both the good/slow ones, that'd be great. 
> 2. I am currently on LucidDb v0.9.2, will a software upgrade fix the
> problem?
Maybe... JVS did a bunch of commits as part of his work with Firewater that might help with some of this.  Actually, for the type of partitioning you're doing, you almost certain want to use that technology.

We are currently preparing it as an "add on" extension to deploy on top of an existing LucidDB installation.  We intend to have a Firewater "mod" available with the 0.9.4 release.
http://p4webhost.eigenbase.org:8080/@md=d&cd=//open/dy/dev/firewater/&c=rOd@/14145?ac=10
Not quite ready yet, but getting there.

> 3. Assuming this is a problem with the LucidDbClient.jar, can I safely use a
> 0.9.3 LucidDbClient.jar with a 0.9.2 server installation?   
We have two customers that have attempted this (and are also subscribed to this email list) which can attest to the issues they faced.  vJDBC was ugpraded during these release to fix issues with Explain plan, etc.  We were able to ultimately get a 0.9.3 installation to talk to a 0.9.2 LucidDB but reintroduced the bugs to do so.  This is, not recommended.  :)

Follow on questions:
 - Can you (or will you) ultimately use Firewater?  It's designed for PRECISELY the use case you're articulating.
 - What is your degree of parallelism set to on your coordinator node?