WL#1292: Autotest of MySQL Cluster

Affects: Server-7.1   —   Status: On-Hold

Use the 12-machine cluster in Uppsala to run autotests of MySQL Cluster.


TODO:
* Setup 12 machine cluster in Uppsala. DONE [Ove]
   - RH9 auto install. 
   - Raid1 on two disks.
* Setup 1 "pinky" machine as the autotest controller. 
    - install it in  cluster 
    - set it up as autotest.mysql.com 
    - set up as webserver for cluster
    - set up as nfs server for cluster
    - this is where one should log in to use the cluster.
    - commands can be mapped onto all machines in the cluster. This simplifies 
      cluster admninistration. Ove will show how.
    - backup it up. This is the only cluster machine that wil have backup.
* Setup autotest script to check out source from bitkeeper. CVS will not be
supported. 
  (v21x can be tested on the minicluster until discontinued.)
* Setup autotest script to build mysql and ndb.
* Setup autotest script to start MySQL cluster. I.e start mgmtsrvr, ndb and 
  mysqld processes.
* Make new version of HUGO for SQL. Keep old version.
* Make test programs that uses HUGO for SQL. In this way we can for example:
  - Insert data using INSERT INTO... via MySql API. Then connect to ndb via 
    NDBAPI and verify data.
  - Stop one node, then UPDATE ... data via MySql API. Verify data in Ndb. 
    Start the stopped node. Verify data both via SQL and NDBAPI.
  - Stop one MySql node. Create table through other MySql node. Start stopped
    MySql node. Test that table can be used from that node.
* Setup test to run regularly in different configurations:
  - 1, 2, 4, 8-node etc.
  - TCP and SHM-transporters
  - Multiple MySql nodes. 
* Setup autotest script to generate html report on autotest.mysql.com 
* Add possibility to show status of a running test.
* Add possibility to stop a running test.
* Reduce use of NFS in autotest scripts. Use scp or rsync instead.
* Reduce use of rsh  kill . Use cpcd instead 
* Add support for multiple cluster to enable testing or replication