HailDB Cursor not resetting this->cursor to clustered index after secondary index operations (causing sysbench failure)

[Lee Bieber]

Initial symptom was Sysbench failing with HailDB.

http://hudson.drizzle.org/view/Drizzle-staging/job/drizzle-staging-sysbench-readonly-01-haildb/
http://hudson.drizzle.org/view/Drizzle-staging/job/drizzle-staging-sysbench-readwrite-02-haildb/

Investigation showed that we weren't resetting the cursor we track in the HailDBCursor object (to help avoid lots of malloc/free cycles we keep a cursor around) back to the clustered index after an operation on a secondary index (e.g. index scan). This mean we were attempting the UPDATE on the wrong index, ending up with a DB_ERROR from ib_cursor_update_row(), which translates to -1 (hence "error -1 from engine").

[Stewart Smith (stewart)]

all branches needed for htis should now be in trunk.

[Lee Bieber (kalebral-deactivatedaccount)]

Still seeing failures with readwrite - http://hudson.drizzle.org/job/drizzle-staging-sysbench-readwrite-02-haildb/6/

[Stewart Smith (stewart)]

I now know the cause of this, working on a fix

[Stewart Smith (stewart)]

There is another bug that I'm tracking before rw sysbench is working 100%, but shouldn't be too long.