better diagnostics

[Antoine Martin]

When things go wrong we rely on "xpra info" to capture the state of the system so that we can figure out what is going wrong, we may need to add more information in there:

• more network layer state
• system load, etc

[Antoine Martin]

Done in:

• r18002 captures more socket information (varies depending on the type of socket), ie for TCP, SSL, websockets or secure websockets:

  $ xpra info | grep socket.options
  client.connection.socket.options.IP.IP_MULTICAST_IF=0
  client.connection.socket.options.IP.IP_MULTICAST_LOOP=1
  client.connection.socket.options.IP.IP_MULTICAST_TTL=64
  client.connection.socket.options.IP.IP_OPTIONS=0
  client.connection.socket.options.IP.IP_TOS=0
  client.connection.socket.options.IP.IP_TTL=64
  client.connection.socket.options.SOCKET.SO_BROADCAST=0
  client.connection.socket.options.SOCKET.SO_DONTROUTE=0
  client.connection.socket.options.SOCKET.SO_ERROR=0
  client.connection.socket.options.SOCKET.SO_KEEPALIVE=0
  client.connection.socket.options.SOCKET.SO_LINGER=0
  client.connection.socket.options.SOCKET.SO_OOBINLINE=0
  client.connection.socket.options.SOCKET.SO_RCVBUF=1062000
  client.connection.socket.options.SOCKET.SO_RCVLOWAT=1
  client.connection.socket.options.SOCKET.SO_RCVTIMEO=0
  client.connection.socket.options.SOCKET.SO_REUSEADDR=1
  client.connection.socket.options.SOCKET.SO_REUSEPORT=0
  client.connection.socket.options.SOCKET.SO_SNDBUF=2626560
  client.connection.socket.options.SOCKET.SO_SNDTIMEO=0
  client.connection.socket.options.SOCKET.SO_TYPE=1
  client.connection.socket.options.TCP.TCP_MAXSEG=65483
  client.connection.socket.options.TCP.TCP_NODELAY=1
  

• r18003: server load (load value multiplied by 1000), ie:

  $ xpra info | grep server.load
  server.load=(630, 640, 680)
  

• r18004: platform fix for win32 to avoid warnings there

@maxmylyn: this may help diagnose performance issues.

[J. Max Mena]

This is all very useful, but I'd like to request another metric to add - iowait - it should be fairly trivial to calculate, and would offer a good metric for users with older/slower Hard Drives. One of my test boxes has a very very slow disk, and as such when I'm doing IO intensive actions, the iowait will go up (as viewed via ​https://netdata.io - very useful visuals), while the CPU utilization stays notably low. This would be useful in pointing out slow storage (or other devices).

[Antoine Martin]

it should be fairly trivial to calculate

I don't think it is at all, please file a separate ticket for that.

[J. Max Mena]

Very well, closing this one.