Description:

------------

To avoide too many concurrency threads，InnoDB use

innodb_thread_concurrency to control the number of operating system

threads concurrently inside InnoDB. When a thread is permitted to

enter InnoDB, it is given a number of "free tickets" equal to the

value of innodb_concurrency_tickets, and the thread can enter and

leave InnoDB freely until it has used up its tickets.

innodb_concurrency_tickets is a fixed number, and it is hard to

choose a proper value fit for both small transactions and large

transactons. With a small innodb_concurrency_tickets value, large

transactons will acquire tickets many times, which extends the length

of time required to complete their task. With a large

innodb_concurrency_tickets value, small transactons may get starved

by too many concurrent large transactions.

This patch provides a method to adaptively adjust tickets assigned to

readonly SELECT SQL statement. For large SELECT query, it may acquire

tickets many times if the previous assigned tickets are exhausted.

To avoid starving smally quries, the tickets assiged to large query are

decreased exponentially according to the number of times it acquire

tickets, i.e. the more times large query acquire tickets, the smaller

tickets it will be assigned. To avoid it getting a too small value,

innodb_rds_min_concurrency_tickets is used as the lower bound of the

tickets number assigned.

Two global InnoDB variables are added,

1. innodb_rds_adaptive_tickets_algo: control whether adaptive algorithm

   is enabled.

2. innodb_rds_min_concurrency_tickets: the lower bound of tickets can

   be assiged under adaptive adjusting.