add impl of Test::mysqld::Multi

[shogo82148]

I often uses Test::mysqld with App::Prove::Plugin::MySQLPool.
App::Prove::Plugin::MySQLPool launches -j number of Test::mysqld instances first,
while the launching process is very slow if the number is large.

I think that Test::mysqld should have an interface for creating some Test::mysqld instances at once.
Your kind consideration of this matter would be sincerely appreciated.

[kazuho]

Thank you for the PR. Can you make the following changes?

• use the functions provided by Test::mysqld to control the database servers
  • the PR should not includes calls like exec or kill
• please move the functions into Test::mysqld if it is possible to implement the feature only by adding functions

[shogo82148]

Thank you for reviewing.

please move the functions into Test::mysqld if it is possible to implement the feature only by adding functions

I did.

use the functions provided by Test::mysqld to control the database servers

I want to call just exec, but Test::mysqld has no such function.
(start function calls exec, while it calls waitpid too. I want not to call waitpid)

[kazuho]

please move the functions into Test::mysqld if it is possible to implement the feature only by adding functions
I did.

Thank you for the changes.

use the functions provided by Test::mysqld to control the database servers

I want to call just exec, but Test::mysqld has no such function.
(start function calls exec, while it calls waitpid too. I want not to call waitpid)

I understand that your code spawns all the MySQL instances, and then waits for the instances to start up. Does the approach result in non-marginal speed-up compared against when using the start function (i.e. doing fork-exec and wait for each instance)?

If not, please use the start function.

If yes, please refactor the start function, by splitting its two blocks (i.e. spawn and wait-for-startup) to two functions, and call the functions from your multi-daemon startup code. Having same logic at two places of the module is a no-go.

[shogo82148]

Does the approach result in non-marginal speed-up compared against when using the start function (i.e. doing fork-exec and wait for each instance)?

Yes.
I refactored.