cached entries are check for validity with ping command.
but this check happens EVERY time send_message is called on client, which, considering bottleneck may be accumulated latency between message servers, may double small message delivery time (and all we have are small messages).
potential solution:
- introduce last ping time for cached entries and do not ping within a configurable time interval from it.
potential problems:
- investigate/test how client reacts to lost connections when it's basically returned broken (if not ping-checked) by open_sending_stream (probably it will be fine - there's error catching on every corner)
cached entries are check for validity with ping command.
but this check happens EVERY time send_message is called on client, which, considering bottleneck may be accumulated latency between message servers, may double small message delivery time (and all we have are small messages).
potential solution:
potential problems: