This helps reduce the effect that a hanging connection has on the whole
system.
Making this configurable shouldn't be needed as long as the rate of
authentication failures doesn't exceed an acceptable level. In most
practical cases allowing two seconds for authentication to complete should
be enough.
By delaying the processing of a client DCB in the zombie queue, we avoid
initiating the shutdown process of the backend DCBs too early. This solves
the problem of the previous implementation where the session shutdown
proceeded too far before the delaying process started.
The fix isn't fully functional as closed DCBs never process the events
they receive. The mechanism would work if certain events would be
processed but they'll have to be fixed in a later release.
By deferring the closing of a DCB until the protocol tells that it's in a
stable state, we avoid closing the connection mid-authentication. This
makes sure that all connections have reached a stable state before they
are closed which in turn prevents the connections from counting towards
aborted connects (or failed authentications like it did with the old fix).
When a RoutingWorker is destructed, all existing local data will
unconditionally be destroyed.
The case in point is the vector<SFilterDef>s that are held as routing
worker local data by the Service. Unless the local data is deleted, the
filter defs will not be deleted at shutdown.
This is somewhat brute-force (but good to have in place nonetheless) as
the deletion of the local data should be done by the Service and before
the routing worker thread has exited.
The code that selects which worker to assign the DCB to is now completely
in the Listener class. This removes the need to change the ownership of a
DCB after it has been allocated.
Storing all the runtime errors makes it possible to return all of them
them via the REST API. MaxAdmin will still only show the latest error but
MaxCtrl will now show all errors if more than one error occurs.
Added an overload to execute_concurrently that takes an std::function as a
parameter and added a const version of operator* for rworker_local. Also
removed the std::move of the return value in rworker_local::values as it
can prevent RVO from taking place.
If the startup of the listeners requires communication with all of the
workers, the workers must be up and running for that to happen.
Due to the fact that the main thread is still a worker thread, the
initialization code is not extra straightforward. By queuing an event to
the main worker, the startup of all listeners is done at a fully
operational state with all workers fully functional.
The service initialization code was also flawed in the sense that it would
cause a deadlock if any of the threads would have to check for the user
permissions. This is mainly a problem with the authenticator modules but
the benefits of the per service pre-loading of users is most likely
superficial. In theory startup will be faster as each thread now queries
the users in parallel.
Allocating DCB with new allows the use of C++ objects in the DCB
struct. Also the explicit poll field can be replaced by inheriting from
MXB_POLL_DATA.
