TCPInfo Plugin

This global plugin logs TCP metrics at various points in the HTTP processing pipeline. The TCP information is retrieved by the getsockopt(2) function using the TCP_INFO option. This is only supported on systems that support the TCP_INFO option, currently Linux and BSD.

Plugin Options

The following options may be specified in plugin.config:

--hooks=NAMELIST

This option specifies when TCP information should be logged. The argument is a comma-separated list of the event names listed below. TCP information will be sampled and logged each time the specified set of events occurs.

Event Name

Triggered when

send_resp_hdr

The server begins sending an HTTP response.

ssn_start

A new TCP connection is accepted.

txn_close

A HTTP transaction is completed.

txn_start

A HTTP transaction is initiated.

--log-file=NAME

This specifies the base name of the file where TCP information should be logged. If this option is not specified, the name tcpinfo is used. Traffic Server will automatically append the .log suffix.

--log-level=LEVEL

The log level can be either 1 to log only the round trip time estimate, or 2 to log the complete set of TCP information.

The following fields are logged when the log level is 1:

Field Name

Description

timestamp

Event timestamp

event

Event name (one of the names listed above)

client

Client IP address

server

Server IP address

rtt

Estimated round trip time in microseconds

The following fields are logged when the log level is 2:

Field Name

Description

timestamp

Event timestamp

event

Event name (one of the names listed above)

client

Client IP address

server

Server IP address

rtt

Estimated round trip time in microseconds

rttvar

last_sent

last_recv

snd_cwnd

snd_ssthresh

rcv_ssthresh

unacked

sacked

lost

retrans

fackets

all_retrans

In addition, on newer Linux system, the following two fields are appended in log level 2: ============== ================================================== Field Name Description ============== ================================================== data_segs_in Number of incoming data segments data_segs_out Number of outgoing data segments

Note: Features such as TSO (TCP Segment Offload) might skew the numbers here. That’s true as well for e.g. the retransmit metrics, i.e. anything that deals with segments.

--sample-rate=COUNT

This is the number of times per 1000 requests that the data will be logged. A pseudo-random number generator is used to determine if a request will be logged. The default value is 1000 and this option is not required to be in the configuration file. To achieve a log rate of 1% you would set this value to 10.

--rolling-enabled=VALUE

This logfile option allows you to set logfile rolling behaviour of the output log file without making any changes to the global logging configurations. This option overrides the proxy.config.output.logfile.rolling_enabled setting in records.config for the tcpinfo plugin. The setting may range from 0 to 3. 0 disables logfile rolling. 1 is the default and enables logfile rolling at specific intervals set by --rolling-interval-sec discussed below. 2 enables logfile rolling by logfile size, see --rolling-size-mb below. Finally a value of 3 enables logfile rolling at specific intervals or size, whichever occurs first using the interval or size settings discussed below.

--rolling-offset-hr=VALUE

Set the hour 0 to 23 at which the output log file will roll when using interval rolling. Default value is 0.

--rolling-interval-sec=VALUE

Set the rolling interval in seconds for the output log file. May be set from 60 to 86400 seconds, Defaults to 86400.

--rolling-size=VALUE

Set the size in MB at which the output log file will roll when using log size rolling. Minimum value is 10, defaults to 1024. In your config file, you may use the K, M, or G suffix as in:

--rolling-size=10M

Examples:

This example logs the simple TCP information to tcp-metrics.log at the start of a TCP connection and once for each HTTP transaction thereafter:

tcpinfo.so --log-file=tcp-metrics --log-level=1 --hooks=ssn_start,txn_start

The file tcp-metrics.log will contain the following log format (with client and server both on 127.0.0.1):

timestamp event client server rtt
20140414.17h40m14s ssn_start 127.0.0.1 127.0.0.1 153859
20140414.17h40m14s txn_start 127.0.0.1 127.0.0.1 181018
20140414.17h40m16s ssn_start 127.0.0.1 127.0.0.1 86869
20140414.17h40m16s txn_start 127.0.0.1 127.0.0.1 19088
20140414.17h40m16s ssn_start 127.0.0.1 127.0.0.1 85718
20140414.17h40m16s txn_start 127.0.0.1 127.0.0.1 38059