I don't have answers to all of your questions, but I'll reply to the ones I know or have opinions on.
On Tue, Jan 28, 2020 at 6:20 AM Zoltan Chovan <[email protected]> wrote: > Hey, > > Additionally, I've been looking at TxnHandler and CompactionTxnHandler > classes, these only use directSql, which makes testing any changes to them > quite difficult. Are there any plans to introduce JDO/Datanucleus in those > parts? > I didn't use JDO when I wrote this because an object model is a really bad abstraction for transaction and locks. The solution for this is not to move it into the object model but rather to move it out of constant database operations and into something that will be faster and lighter weight. I believe Olli Draese has some thoughts on how to do this, something along the lines of an in memory cache for reads and a WAL for writing out locks and transaction open, commit, abort. > > Also how committed are we to Datanucleus as the JDO/ORM lib? Is there or > were there any plans to switch? I know about Hibernate and MyBatis as > possible alternatives. Sidenote: afaik Hibernate has LGPL license which > might be a problem, is that correct? MyBatis on the other hand is Apache > License 2.0. > LGPL excludes Hibernate as an option. > > I am aware that these changes would be quite enormous, but on the long run > it might give us better performance and testability, also most likely > easier/quicker development in the related parts. My intention is to start a > discussion on the topic. > Given the huge impact this change would have on developers and the existing user base, the expected benefit would have to be very large for it to make sense to pay the cost. Alan. > > What are your thoughts? > > Best regards, > Zoltan > > [1] https://github.com/apache/hive/blob/master/pom.xml#L131 > [2] http://www.datanucleus.org/documentation/products.html > [3] http://www.datanucleus.org/products/accessplatform_5_1/migration.html >
