nats module subscriber issue when connected to more "nats_urls"

[smititelu]

Description, Troubleshooting, Reproduction

I have 3 go nats servers running and the following kamailio config for nats:

modparam("nats", "nats_url", "nats://127.0.0.1:4222")
modparam("nats", "nats_url", "nats://127.0.0.2:4222")
modparam("nats", "nats_url", "nats://127.0.0.3:4222")

modparam("nats", "subject_queue_group", "foo:2020")
...

event_route[nats:foo]
{
                xlog("L_ERR", "==============> received $natsData");
}

I try to publish using some nats publisher (e.g. nats.c example/libuv-pub.c), and nothing gets printed in event route:

./pub -s nats://127.0.0.1:4222 -txt "aaaaaaaaaaaa" -count 1 // event route log NOT printed
./pub -s nats://127.0.0.2:4222 -txt "aaaaaaaaaaaa" -count 1 // event route log NOT printed

However, if I try to publish on the last server "127.0.0.3", log is printed:

./pub -s nats://127.0.0.3:4222 -txt "aaaaaaaaaaaa" -count 1 // event route log IS printed

I expected to see 1 log message for each of the above publish(es). Found this while implementing and testing #2915

Thanks,
Stefan

[smititelu]

Found in nats.h file, from nats.c git repo:

 * By default, #natsConnection_Connect() attempts to connect to a server
 * specified in provided list of servers. If it cannot connect and the list has been
 * fully tried, the function returns an error.

So the lib behavior is to connect to a single server from the list, not to all servers in the list. Right now, the subscriber part of kamailio nats module creates a single connection with the list of servers from "nats_url" modparams.

To be decided later if a single connection per "nats_url" is needed for subscriber part of kamailio nats module.

Closing this, as it is expected lib behavior. Will document this in "nats_url" modparam.

Thanks,
Stefan

[eschmidbauer]

This sounds like a nice improvement, I submitted a PR to move the connection handling into a new struct. We can extend it into a linked-list by doing something like this:

I think we should leave the default behavior of relying on libnats to do connection failover based on the server list it is given, and define a "name" on the connection to create simultaneous connections.
perhaps using something similar to htable like

modparam("nats", "nats_url", "nats://127.0.0.1:4222;name=connection1")
modparam("nats", "nats_url", "nats://127.0.0.1:5222;name=connection1")
modparam("nats", "nats_url", "nats://127.0.0.1:6222;name=connection2")

This would create two sets of connections.
"connection1" would contain server list of nats://127.0.0.1:4222 and nats://127.0.0.1:5222
"connection2" would contain server list of nats://127.0.0.1:6222
We could then reference the connections by name in other modparams or functions.