Non JSON error message at the end of JSON output at the client-side of the measurement

[zokl]

• Version of iperf3: master, 787919c

• Hardware: x86_64

• Operating system (and distribution, if any): Linux, OpenWRT, iperf compiler from github repo

Bug Report

• Expected Behavior

• Actual Behavior

Non JSON error message at the end of JSON output at the client-side of the measurement. For now, I found only described behavior for this message: "iperf3: error - unable to send control message: Broken pipe". Server-side produces valid JSON output.

Wrong JSON output on client-side

client-4.iperf3.gz

			"remote_total":	0,
			"remote_user":	0,
			"remote_system":	0
		},
		"sender_tcp_congestion":	"cubic"
	}
}
iperf3: error - unable to send control message: Broken pipe

Good JSON output on server-side

server-4.iperf3.gz

				"bytes":	0,
				"bits_per_second":	0,
				"omitted":	false,
				"sender":	false
			}
		}],
	"end":	{
	},
	"error":	"error - select failed: Bad file descriptor"
}

• Steps to Reproduce

/usr/bin/iperf3 -c 1.2.3.4 --connect-timeout 1000 -t 3600 --logfile test-4.iperf3 -p 5201 -J --parallel 5 -i 1 --window 32k --set-mss 1400
/usr/bin/iperf3 -c 1.2.3.4 --connect-timeout 1000 -t 3600 --logfile test-2.iperf3 -p 5200 -J -R --parallel 5 -i 1 --window 32k --set-mss 1400

[davidBar-On]

I tried to run the non-reverse test ("test-4.iperf3" log) on my local machine and there was no problem. Maybe try running using the latest iperf3 code (lately some changes were done that that may be related). If that won't help, maybe debug info (-V -d) can help.

The error probably happened in client when for some reason when trying to send IPERF_DONE state to the server fails. Since this is done after the JSON output is created, the error is shown outside of the JSON.

[zokl]

The described problem occurs if the connection interface goes down during the test. Typical situation is a mobile interface and its restart because of a problem with signal etc. Version of iperf3 ismaster, 787919c.

For testing is good to use the following scenarios:

1. On the server: iperf3 -s
2. On client: /usr/bin/iperf3 -c 1.2.3.4 --connect-timeout 1000 -t 60 --logfile test.iperf3 -p 5201 -J --parallel 5 -i 1 --window 32k --set-mss 1400
3. On the client: wait 5 seconds for test start
4. On the client: ifconfig wwan0 down
5. Wait for ending iperf3 (it take several minutes)

JSON output in test.iperf3 is not correct because it end by

                            "host_total":   0.32808521971039462,
                        "host_user":    0.065823820341056063,
                        "host_system":  0.26226304830393965,
                        "remote_total": 0,
                        "remote_user":  0,
                        "remote_system":        0
                }​​​​​,
                "sender_tcp_congestion":        "cubic"
        }​​​​​
}​​​​​
iperf3: error - unable to send control message: Broken pipe

and this is not a correct JSON format.

The same JSON format problem occurs if the test ends by „iperf3: error - idle timeout for receiving data”. This could be reproduced by the following scenario:

1. On the server: iperf3 -s
2. On client: /usr/bin/iperf3 -c 1.2.3.4 --connect-timeout 1000 -t 60 --logfile test.iperf3 -p 5201 -J --parallel 5 -i 1 --window 32k --set-mss 1400 -R --rcv-timeout 5000
3. On the client: wait 5 seconds for test start
4. On the client: ifconfig wwan0 down
5. Wait for ending iperf3 (it take several seconds)

JSON output in test.iperf3 is not correct too:

                                "end":  11.001215,
                                "seconds":      0.999882996082306,
                                "bytes":        0,
                                "bits_per_second":      0,
                                "omitted":      false,
                                "sender":       false
                        }
                }],
        "end":  {
        }
}
iperf3: error - idle timeout for receiving data

[davidBar-On]

Thanks for the additional information - helped to understand what is the problem. Submitted PR #1146 with a suggested fix.

[zokl]

Thanks for the additional information - helped to understand what is the problem. Submitted PR #1146 with a suggested fix.

Thank you for fix.

[bmah888]

Another way to reproduce this problem is just to kill the server side during the test.