[ 
https://issues.apache.org/jira/browse/CALCITE-1190?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15228679#comment-15228679
 ] 

Josh Elser commented on CALCITE-1190:
-------------------------------------

Actually, maybe a good first stab is to just implement an HSQLDB avatica server 
impl that we can provide avatica jars at runtime? Same approach, but we 
wouldn't have to depend on Phoenix/Drill/etc to wire it up.

> Cross-Version Compatibility Test Harness
> ----------------------------------------
>
>                 Key: CALCITE-1190
>                 URL: https://issues.apache.org/jira/browse/CALCITE-1190
>             Project: Calcite
>          Issue Type: Test
>          Components: avatica
>            Reporter: Josh Elser
>            Assignee: Josh Elser
>             Fix For: avatica-1.8.0
>
>
> One thing that the Protobuf serialization aimed to provide was a library 
> which provides us the tools to make Avatica compatible across versions. 
> However, Protobuf is just a tool and the developers can still misuse protobuf 
> in such a way that breaks compatibility across versions. Not to mention, 
> compatibility isn't even remotely certain without any tests.
> Because of Avatica's position as a library less than a product, we have to 
> defer some logic to the concrete product being tested (e.g. Phoenix or 
> Drill). I'm thinking something like the following:
> The user provides pairs of client and server "definitions" for a given 
> version of compatibility. This would include some version of Avatica and some 
> backing database. For example, Avatica-1.7.1 and Phoenix-4.7.0 or 
> Avatica-1.8.0-SNAPSHOT and HSQLDB-2.3.1.
> The client half would be some template for the appropriate JDBC url to use 
> (sans the URL to the Avatica server) and a JAR file containing the necessary 
> classes to run the j.s.Driver. The server half would just be a URL to the 
> Avatica server instance.
> The test harness itself can provide the logic to test the remote driver 
> against the other servers, enumerating the possible combinations of 
> client-server communication. Starting the server for each version to test is 
> likely too difficult to automate well given the unknown of what the server 
> will be, so that will be left as a prerequisite.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to