tcpretrans_example.txt

Demonstrations of tcpretrans, the Linux eBPF/bcc version.


This tool traces the kernel TCP retransmit function to show details of these
retransmits. For example:

# ./tcpretrans
TIME     PID    IP LADDR:LPORT          T> RADDR:RPORT          STATE
01:55:05 0      4  10.153.223.157:22    R> 69.53.245.40:34619   ESTABLISHED
01:55:05 0      4  10.153.223.157:22    R> 69.53.245.40:34619   ESTABLISHED
01:55:17 0      4  10.153.223.157:22    R> 69.53.245.40:22957   ESTABLISHED
[...]

This output shows three TCP retransmits, the first two were for an IPv4
connection from 10.153.223.157 port 22 to 69.53.245.40 port 34619. The TCP
state was "ESTABLISHED" at the time of the retransmit. The on-CPU PID at the
time of the retransmit is printed, in this case 0 (the kernel, which will
be the case most of the time).

Retransmits are usually a sign of poor network health, and this tool is
useful for their investigation. Unlike using tcpdump, this tool has very
low overhead, as it only traces the retransmit function. It also prints
additional kernel details: the state of the TCP session at the time of the
retransmit.


A -l option will include TCP tail loss probe attempts:

# ./tcpretrans -l
TIME     PID    IP LADDR:LPORT          T> RADDR:RPORT          STATE
01:55:45 0      4  10.153.223.157:22    R> 69.53.245.40:51601   ESTABLISHED
01:55:46 0      4  10.153.223.157:22    R> 69.53.245.40:51601   ESTABLISHED
01:55:46 0      4  10.153.223.157:22    R> 69.53.245.40:51601   ESTABLISHED
01:55:53 0      4  10.153.223.157:22    L> 69.53.245.40:46444   ESTABLISHED
01:56:06 0      4  10.153.223.157:22    R> 69.53.245.40:46444   ESTABLISHED
01:56:06 0      4  10.153.223.157:22    R> 69.53.245.40:46444   ESTABLISHED
01:56:08 0      4  10.153.223.157:22    R> 69.53.245.40:46444   ESTABLISHED
01:56:08 0      4  10.153.223.157:22    R> 69.53.245.40:46444   ESTABLISHED
01:56:08 1938   4  10.153.223.157:22    R> 69.53.245.40:46444   ESTABLISHED
01:56:08 0      4  10.153.223.157:22    R> 69.53.245.40:46444   ESTABLISHED
01:56:08 0      4  10.153.223.157:22    R> 69.53.245.40:46444   ESTABLISHED
[...]

See the "L>" in the "T>" column. These are attempts: the kernel probably
sent a TLP, but in some cases it might not have been ultimately sent.

To spot heavily retransmitting flows quickly one can use the -c flag. It will
count occurring retransmits per flow.

# ./tcpretrans.py -c
Tracing retransmits ... Hit Ctrl-C to end
^C
LADDR:LPORT              RADDR:RPORT             RETRANSMITS
192.168.10.50:60366  <-> 172.217.21.194:443         700
192.168.10.50:666    <-> 172.213.11.195:443         345
192.168.10.50:366    <-> 172.212.22.194:443         211
[...]

This can ease to quickly isolate congested or otherwise awry network paths
responsible for clamping tcp performance.

TCP sequence numbers can be included via -s, except in count mode. These numbers
are useful for identifying specific retransmissions in large packet caputes.
Note, lossprobe -l output will display 0 for the sequence number for L type.

# ./tcpretrans.py -s
TIME     PID    IP LADDR:LPORT          T> RADDR:RPORT          STATE        SEQ
18:03:46 0      4  192.168.10.50:41976  R> 172.217.21.194:443   SYN_SENT     2879306108
18:03:49 0      4  192.168.10.50:41976  R> 172.217.21.194:443   SYN_SENT     2879306108

USAGE message:

# ./tcpretrans -h
usage: tcpretrans.py [-h] [-s] [-l] [-c] [-4 | -6]

Trace TCP retransmits

optional arguments:
  -h, --help       show this help message and exit
  -s, --sequence   display TCP sequence numbers
  -l, --lossprobe  include tail loss probe attempts
  -c, --count      count occurred retransmits per flow
  -4, --ipv4       trace IPv4 family only
  -6, --ipv6       trace IPv6 family only

examples:
    ./tcpretrans           # trace TCP retransmits
    ./tcpretrans -l        # include TLP attempts
    ./tcpretrans -4        # trace IPv4 family only
    ./tcpretrans -6        # trace IPv6 family only