Add more internal metrics

[An-DJ]

Is your feature request related to a problem? Please describe.
The pgagroal lacks some metrics of internal resource state, such as network traffic and so on.

Describe the solution you'd like
This feature includes new metrics as follows:

• pgagroal_network_sent which characterizes the network traffic sent
• pgagroal_network_received which characterizes the network traffic received
• pgagroal_client_sockets which characterizes the number of sockets client used
• pgagroal_self_sockets which characterizes the number of sockets pgagroal used itself

[An-DJ]

@jesperpedersen I have pushed a new commit which contains the implementation of pgagroal_client_sockets and pgagroal_self_sockets.

I am considering whether to remove metric pgagroal_server_sockets, since it is always equal to the total number of connections. Could you give me some advice?

[jesperpedersen]

pgagroal_client_sockets is the important one. pgagroal_self_sockets should stay more or less constant.

For pgagroal_server_sockets I would just use the existing metrics for connections, and display that as server - e.g. not a new metric.

[An-DJ]

pgagroal_client_sockets is the important one. pgagroal_self_sockets should stay more or less constant.

For pgagroal_server_sockets I would just use the existing metrics for connections, and display that as server - e.g. not a new metric.

Server sockets has been added into dashboard in new commit in #165. PTAL~ @jesperpedersen

[An-DJ]

@jesperpedersen Besides, I notice that the main memory size is fixed by configuration file like shared memory created by mmap and so on. I have no idea whether pgagroal_mem_alloc should be added into pgagroal.

Could you give me some Suggestions?

[jesperpedersen]

There is a very minimal allocation rate which happens during startup of a connection, or with the client.

The pipeline uses a fixed buffer for communication so that doesn't change.

As such I don't think we need a metric for allocations, and there is no portable system function for checking memory usage.

[An-DJ]

There is a very minimal allocation rate which happens during startup of a connection, or with the client.

The pipeline uses a fixed buffer for communication so that doesn't change.

As such I don't think we need a metric for allocations, and there is no portable system function for checking memory usage.

I think so, too. Thanks!