From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <vdavydov.dev@gmail.com>
Date: Mon, 25 Feb 2019 14:34:38 +0300
From: Vladimir Davydov <vdavydov.dev@gmail.com>
Subject: Re: [tarantool-patches] [PATCH] iproto: report active connections
 number
Message-ID: <20190225113438.7ppiqajf6rk5hd4t@esperanza>
References: <20190224235225.21885-1-i.kosarev@tarantool.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190224235225.21885-1-i.kosarev@tarantool.org>
To: Ilya Kosarev <i.kosarev@tarantool.org>
Cc: tarantool-patches@freelists.org, georgy@tarantool.org, i.kosarev@corp.mail.ru
List-ID: <tarantool-patches.dev.tarantool.org>

On Mon, Feb 25, 2019 at 02:52:25AM +0300, Ilya Kosarev wrote:
> Now there is new member in box.stat.net() called "CONNECTIONS" which is a number of active iproto connections.
> 
> Closes #3905
> ---
> Branch: https://github.com/tarantool/tarantool/tree/i.kosarev/gh-3905-report-connections-number
> Issue: https://github.com/tarantool/tarantool/issues/3905

My comments to "[PATCH] Close socket explicitly before wal_dir at exit"
regarding the commit meesage are valid for this patch as well. Please
fix accordingly.

> 
>  src/box/iproto.cc          | 7 ++++++-
>  test/box/stat_net.result   | 8 ++++++++
>  test/box/stat_net.test.lua | 2 ++
>  3 files changed, 16 insertions(+), 1 deletion(-)
> 
> diff --git a/src/box/iproto.cc b/src/box/iproto.cc
> index a08c8c5cb..a20a7b005 100644
> --- a/src/box/iproto.cc
> +++ b/src/box/iproto.cc
> @@ -267,10 +267,11 @@ struct rmean *rmean_net;
>  enum rmean_net_name {
>  	IPROTO_SENT,
>  	IPROTO_RECEIVED,
> +	IPROTO_CONNECTIONS,
>  	IPROTO_LAST,
>  };
>  
> -const char *rmean_net_strings[IPROTO_LAST] = { "SENT", "RECEIVED" };
> +const char *rmean_net_strings[IPROTO_LAST] = { "SENT", "RECEIVED", "CONNECTIONS" };

No-no-no. RPS doesn't make any sense at all for the number of active
connections. Please don't extend it. Add a simple counter instead.
Perhaps, you don't even need to do that and it's possible to extract
the number of connections from iproto_connection_pool stats. Check it
out, please.

>  
>  static void
>  tx_process_destroy(struct cmsg *m);
> @@ -579,6 +580,8 @@ iproto_connection_close(struct iproto_connection *con)
>  		/* Make evio_has_fd() happy */
>  		con->input.fd = con->output.fd = -1;
>  		close(fd);
> +		/* Count statistics */
> +		rmean_collect(rmean_net, IPROTO_CONNECTIONS, -1);
>  		/*
>  		 * Discard unparsed data, to recycle the
>  		 * connection in net_send_msg() as soon as all
> @@ -1853,6 +1856,8 @@ iproto_on_accept(struct evio_service * /* service */, int fd,
>  	struct iproto_connection *con = iproto_connection_new(fd);
>  	if (con == NULL)
>  		return -1;
> +	/* Count statistics */
> +	rmean_collect(rmean_net, IPROTO_CONNECTIONS, 1);
>  	/*
>  	 * Ignore msg allocation failure - the queue size is
>  	 * fixed so there is a limited number of msgs in
> diff --git a/test/box/stat_net.result b/test/box/stat_net.result
> index b3e3db11f..be7f2ff8a 100644
> --- a/test/box/stat_net.result
> +++ b/test/box/stat_net.result
> @@ -46,6 +46,10 @@ box.stat.net.RECEIVED.total > 0
>  ---
>  - true
>  ...
> +box.stat.net.CONNECTIONS.total > 0
> +---
> +- true
> +...

The test is poor. I want to see that you actually check that the number
of connections is accounted correctly. Please use net.box module to
establish several connections and check that the number reported in the
stats is correct.