Re-initialize the reply parser for every new connection. If a connection is terminated,
the parser could be left in a bad state. After the auto-reconnect magic kicks in, it
tries to reuse the old parser, which will not work.
This change is visible to client programs if you depend on client.reply_parser.name being
set immediately. It will now only be set after a connection is established.
Thanks to @jhurliman for reporting and @pietern for the fix suggestion.
Simply and speed up command argument processing logic.
Commands now return the true/false value from the underlying socket write(s).
Implement command_queue high water and low water for more better control of queueing.
When a Redis server starts up, it might take a while to load the dataset into memory.
During this time, the server will accept connections, but will return errors for all non-INFO
commands. Now node_redis will send an INFO command whenever it connects to a server.
If the info command indicates that the server is not ready, the client will keep trying until
the server is ready. Once it is ready, the client will emit a "ready" event as well as the
"connect" event. The client will queue up all commands sent before the server is ready, just
like it did before. When the server is ready, all offline/non-ready commands will be replayed.
This should be backward compatible with previous versions.
To disable this ready check behavior, set `options.no_ready_check` when creating the client.
As a side effect of this change, the key/val params from the info command are available as
`client.server_options`. Further, the version string is decomposed into individual elements
in `client.server_options.versions`.
