From SQL to NDB API and the MySQL Cluster Data Nodes and Back Again

Understanding MySQL Cluster query execution while converting SQL to NDB API. In this session we will start with two SQL queries (a key lookup and a join) and follow them through to what NDB API code is run by the MySQL Server, what we’d write if we were being optimal (NDB API in C++/Java) and then how the MySQL Cluster Data Nodes execute this code.

Come to this talk if: - you’re interested in using MySQL Cluster through the NDB API OR - you’re interested in how MySQL Cluster executes queries OR - you’re interested in NDB API Connectors

You’re expected to: - not be scared of SQL (much) - Be okay with seeing C++ and/or Java code

s/Java/$LANGUAGE_OF_CHOICE/ as the NDB API Connectors are awesome.

OUTLINE: – Pkey lookup SQL: CREATE TABLE t1 (pk int PRIMARY KEY, v VARCHAR) ENGINE=NDB; INSERT INTO t1 values (?,?); SELECT v FROM t1 WHERE pk=1; NDB API: Transaction/Operation intro NDB KERNEL: Where signals go (and come from) When data is sent/received from kernel Performance: NDBAPI (C++) versus NDBAPI (Java). API versus SQL.

- (Simple) Join
       SQL:
               select b.name, f.insult from bstrd b, fckr f 
               where b.insult_id=f.id and b.id=%
   NDB API:
       batched key access, TC selection hints
   NDB KERNEL: 
       how the latency is bad through MySQL but not so much through API 
               How this is changing with MySQL Server Batched Key Access for joins in (hopefully) demoable code
   Performance
       NDBAPI (C++) versus NDBAPI(Java) versus SQL