Fixes #10806 - cache local and remote socket address in EndPoint for benefit of RequestLog

[joakime]

The EndPoint local-address and remote-address are inaccessible once the EndPoint is closed due to "ensureOpen" checks by the Java implementation itself.

This PR merely caches the local-address and remote-address in the EndPoint classes to be accessed if unable to get the attempt to get the SocketAddress from the Channel fails due to a ClosedChannelException.

Fixes: #10806

[gregw]

Amazed we have not noticed this before!

[gregw]

@joakime I've pushed a bit more adventurous cleanup of local/remote addresses to this PR. Thoughts?

I've contributed now

[joakime]

@joakime I've pushed a bit more adventurous cleanup of local/remote addresses to this PR. Thoughts?

Interesting, kinda like it.
How well does the UnixDomainSocket stuff work with this new approach?

[sbordet]

I don't think this is right.
What about the case of socket address reuse?

If it's just a problem of the request log, it probably should be fixed there rather than returning cached addresses that may be used in completely different contexts where returning a cached address is not right.

For example clients may behave differently from servers.

Also, in QUIC the remote address may change dynamically, so we don't want to cache.

The RequestLog is passed _request.getLoggedRequest(), so we can easily wrap the "logged" request and cache the socket addresses there, rather than everywhere else.

[joakime]

If it's just a problem of the request log, it probably should be fixed there rather than returning cached addresses that may be used in completely different contexts where returning a cached address is not right.

The problem is way down in the Java Impl.
If the socket is closed, you cannot get the LocalAddress or RemoteAddress out from it due to an ensureOpen() check in the implementation.

[joakime]

I'm working on an alternative approach suggested by a conversation with @sbordet, i'll submit that as a new PR.

[gregw]

I think there might be some goodness in the cleanup in this PR even if we don't make addresses immutable in end points. So, before closing this one, please look to see if any of the changes make sense if we don't make the addresses immutable.

[joakime]

Dropping in favor of #10867