When zabbix_sender is used with the -i flag to supply multiple key/values, I noticed it pretty-prints JSON on the TCP layer, wasting bandwidth (although making things more readable).

tcpdump of zabbix_sender sending multiple-values using the -i flag:

"10:41:26.642318 IP 10.10.14.30.39445 > tools01.example.com.zabbix-trapper: Flags [.], seq 5:2901, ack 1, win 115, options [nop,nop,TS val 1677996083 ecr 849199326], length 2896
@[email protected].

..

`..'C.d.+{9Y....s;......
d.032....9......{
"request":"sender data",
"data":[

{ "host":"app30.example.com", "key":"test_s1r1_index_deletedDocs", "value":"0"}

,

{ "host":"app30.example.com", "key":"test_s1r1_index_sizeInBytes", "value":"65"}

,

{ "host":"app30.example.com", "key":"test_s1r1_index_numDocs", "value":"0"}

,

{ "host":"app30.example.com", "key":"test_s1r1_index_maxDoc", "value":"0"}

,

{ "host":"app30.example.com", "key":"test_s1r1_select_medReqTime", "value":"37.77"}

,

{ "host":"app30.example.com", "key":"test_s1r1_select_avgTPerReq", "value":"37.77"}

,

{ "host":"app30.example.com", "key":"test_s1r1_select_95thPcReqTime", "value":"37.77"}

,

{ "host":"app30.example.com", "key":"test_s1r1_select_errors", "value":"0"}

,

{ "host":"app30.example.com", "key":"test_s1r1_select_requests", "value":"1"}

,

{ "host":"app30.example.com", "key":"test_s1r1_select_timeouts", "value":"0"}

,

{ "host":"app30.example.com", "key":"test_s1r1_select_75thPcReqTime", "value":"37.77"}

,

{ "host":"app30.example.com", "key":"test_s1r1_select_avgReqPerSec", "value":"0.00"}

,

{ "host":"app30.example.com", "key":"test_s1r1_select_99thPcReqTime", "value":"37.77"}

,
....
"

I suggest that zabbix_sender DOES NOT pretty print JSON on the wire, reducing overhead. Thoughts?

Tim Vaillancourt