Sébastien Blin authored 2 years ago and Adrien Béraud committed 2 years ago

Make the proxy able to support the UnifiedPush platform, so that
applications using the proxy server can use any UnifiedPush provider.

Documentation: https://unifiedpush.org/

How to test:
+ Install a UnifiedPush provider on a device (e.g. https://ntfy.sh)
+ Run dhtnode with a proxyserver and --unifiedpush https://ntfy.sh
+ Subscribe to a random topic on ntfy (e.g. "dhtproxytest")
+ Subscribe to a hash on the DHT:
    curl -X SUBSCRIBE http://my.dht.proxy:8000/key/foo \
         -d '{"key":"dhtproxytest", "platform":"unifiedpush"}'
+ Send data on the subscribed hash

Note: the topic should be completely random.

Sébastien Blin authored 2 years ago and Adrien Béraud committed 2 years ago

If the proxy got TLS enabled, this doesn't mean that the REST
endpoint for push notifications is also using TLS.

Also, fix an incorrect log.

Sébastien Blin authored 2 years ago and Adrien Béraud committed 2 years ago

when searching for a value to put, the proxy server returned
a json with value->id = 0, causing the client to not refreshing
correctly the values later (when timeout-ing, the value to refresh
is not found).

Sébastien Blin authored 2 years ago and Adrien Béraud committed 2 years ago

When a second permanent put request was received, we didn't call
async_wait after changing the expiration, causing push to not be
sent the second day and presence to expire on some devices after
~3 days.

http: modify timeout behaviour

Sébastien Blin authored 2 years ago and Adrien Béraud committed 2 years ago

The current master is broken with clang, this fix the issue.

Sébastien Blin authored 2 years ago and Adrien Béraud committed 2 years ago

When switching between two network interfaces, local address will
change but not the public ip. This generally cause the client to
not correctly updates its candidates.
In this patch, we inject in the proxy client the local candidates
and use it to check if the network changed.

In the previous version, the timeout for the connection could be
set only after async_connect while the request is sending.
This means if async_connect is slow (it can happen as if there is
an ipv6 endpoint, it can try to resolve it during dozens of seconds
before falling back on ipv4).

In this patch, it moves the timeout callback in the request. This
callback will be set just before async_connect, to be sure that we
are using the callback while connecting.