[go-nuts] How to set socket option for a "TCPListener" before bind?

On Tuesday, June 25, 2013 7:35:23 AM UTC+8, Ewan Chou wrote:

I can get the "syscall.Handle" from a "TCPListener" by reflection, and
call "syscall.SetsockoptInt".

But the type of the only field of "TCPListener" is "netFD" which is
unexported, so the only way to get a "TCPListener" is to call
"net.ListenTCP".

But the listener is already bound.

My need is to set "SO_EXCLUSIVEADDRUSE" option for a TCP listener on
WINDOWS, so more than one processes can listen to the same port.

that option can be set only before bind.

How can I do that?

On Tue, Jun 25, 2013 at 10:38 AM, Ewan Chou wrote:
I tried to call “syscall.GetsockoptInt(fd, syscall.SOL_SOCKET,
^syscall.SO_REUSEADDR)” directly, but I got the error "not supported by
windows".

--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to golang-nuts...@googlegroups.com <javascript:>.
For more options, visit https://groups.google.com/groups/opt_out.

On Tuesday, June 25, 2013 8:47:39 AM UTC+8, Dave Cheney wrote:

If you're really keen on doing this you'll find a hint in the
go.net/ipv4. From my reading SO_EXCLUSIVEADDRUSE does exactly the
opposite of what you want.

On Tuesday, June 25, 2013 10:38:40 AM UTC+8, Dave Cheney wrote:

I can make two processes listen on the same port now, but only through
"syscall".

Why do you need to do such a thing ? The only use cases for this
facility that I am aware of is listening to a UDP port for multicast
messages (ie, mDNS)

On Tue, Jun 25, 2013 at 12:46 PM, Ewan Chou wrote:
If two process can listen on the same port, then the application can do
zero-downtime restart.

--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to golang-nuts...@googlegroups.com <javascript:>.
For more options, visit https://groups.google.com/groups/opt_out.

On Tuesday, June 25, 2013 10:50:52 AM UTC+8, Dave Cheney wrote:

Ok, so now we're getting to the actual problem you are trying to solve.

Some comments

1. Do you really need zero downtime, how about 1 second downtime ? 5
seconds, 10 seconds ?
2. If you really need zero downtime, then you can't achieve that with
a single process, so you're really talking about how to cleanly
shutdown one worker at a time for a rolling upgrade scenario.
3. You should research how products like nginx do process restarts,
they don't use socket tricks like this, they just pass the accepting
socket down to a child process.

So, some questions

1. what platform(s) are you targeting ?
2. what is your SLA ?
3. how are you currently managing your process at the moment, and how
quickly can it restart the process if it fails ?

Cheers

Dave

On Tue, Jun 25, 2013 at 12:46 PM, Ewan Chou
<coo...@gmail.com<javascript:>>
wrote:
through