How do I measure request and response times at once using cURL?

Question

I have a web service that receives data in JSON format, processes the data, and then returns the result to the requester.

I want to measure the request, response, and total time using cURL.

My example request looks like:

curl -X POST -d @file server:port

and I currently measure this using the time command in Linux:

time curl -X POST -d @file server:port

The time command only measures total time, though - which isn't quite what I am looking for.

Is there any way to measure request and response times using cURL?

Unstrung

Answer 1

From this brilliant blog post... https://blog.josephscott.org/2011/10/14/timing-details-with-curl/

cURL supports formatted output for the details of the request (see the cURL manpage for details, under -w, –write-out <format>). For our purposes we’ll focus just on the timing details that are provided. Times below are in seconds.

1. Create a new file, curl-format.txt, and paste in:

        time_namelookup:  %{time_namelookup}s\n
           time_connect:  %{time_connect}s\n
        time_appconnect:  %{time_appconnect}s\n
       time_pretransfer:  %{time_pretransfer}s\n
          time_redirect:  %{time_redirect}s\n
     time_starttransfer:  %{time_starttransfer}s\n
                        ----------\n
             time_total:  %{time_total}s\n
   

2. Make a request:

    curl -w "@curl-format.txt" -o /dev/null -s "http://wordpress.com/"
   

   Or on Windows, it's...

    curl -w "@curl-format.txt" -o NUL -s "http://wordpress.com/"
   

What this does:

-w "@curl-format.txt" tells cURL to use our format file
-o /dev/null redirects the output of the request to /dev/null
-s tells cURL not to show a progress meter
"http://wordpress.com/" is the URL we are requesting. Use quotes particularly if your URL has "&" query string parameters

And here is what you get back:

   time_namelookup:  0.001s
      time_connect:  0.037s
   time_appconnect:  0.000s
  time_pretransfer:  0.037s
     time_redirect:  0.000s
time_starttransfer:  0.092s
                   ----------
        time_total:  0.164s

I have not yet seen an option to output the results in microseconds, but if you're aware of one, post in the comments below.

Make a Linux/Mac shortcut (alias)

alias curltime="curl -w \"@$HOME/.curl-format.txt\" -o /dev/null -s "

Then you can simply call...

curltime wordpress.org

Thanks to commenter Pete Doyle!

Make a Linux/Mac stand-alone script

This script does not require a separate .txt file to contain the formatting.

Create a new file, curltime, somewhere in your executable path, and paste in:

#!/bin/bash

curl -w @- -o /dev/null -s "$@" <<'EOF'
    time_namelookup:  %{time_namelookup}\n
       time_connect:  %{time_connect}\n
    time_appconnect:  %{time_appconnect}\n
   time_pretransfer:  %{time_pretransfer}\n
      time_redirect:  %{time_redirect}\n
 time_starttransfer:  %{time_starttransfer}\n
                    ----------\n
         time_total:  %{time_total}\n
EOF

Then call it the same way as the alias:

curltime wordpress.org

Make a Windows shortcut (aka BAT file)

Create a new text file called curltime.bat in the same folder as curl.exe and curl-format.txt, and paste in the following line:

curl -w "@%~dp0curl-format.txt" -o NUL -s %*

Then from the command line you can simply call:

curltime wordpress.org

(Make sure the folder is listed in your Windows PATH variable to be able to use the command from any folder.)

Snipe

In the windows BAT file it only sends the first parameter, change to this to pass all parameters and to disable echo the command it self: @curl -w "@%~dp0curl-format.txt" -o NUL -s %* Great answer – Gnu

For Linux, I made a dotfile and an alias and it seems to work well: alias curltime="curl -w \"@$HOME/.curl-format.txt\" -o NUL -s ". Likely works on MacOS, too. – Trivium

For zsh I you can use this alias with arguments: alias curltime='curl -w "@$HOME/bin/curl-format.txt" -o /dev/null -s $1' which makes you able to do curltime http://someurl.com :) – Warfore

Would it be possible to run it multiple times to a have an average result? – Rabbinism

not working on win11 – Bencher

Curl always changes its commands. For this command to work, you must now pass the -k parameter, for example: curl -k -w "@data.txt" -o /dev/null -s "your url" – Finback

Answer 2

Option 1: to measure total time:

curl -o /dev/null -s -w 'Total: %{time_total}s\n'  https://www.google.com

Sample output:

Total: 0.441094s

Option 2: to get time to establish connection, time to first byte (TTFB) and total time:

curl -o /dev/null -s -w 'Establish Connection: %{time_connect}s\nTTFB: %{time_starttransfer}s\nTotal: %{time_total}s\n'  https://www.google.com

Sample output:

Establish Connection: 0.020033s
TTFB: 0.417907s
Total: 0.435486s

Ref: Get response time with curl

Adjure

Cool, but it seems cannot check bad URL. – Haily

Answer 3

Here is the answer:

curl -X POST -d @file server:port -w %{time_connect}:%{time_starttransfer}:%{time_total}

All of the variables used with -w can be found in man curl.

Unstrung

It is better for user experience to add new lines: "\n%{time_connect}:%{time_starttransfer}:%{time_total}\n" – Sulfuric

For me it was not working without quotes. I would suggest adding quotes while specifying the format /h/a/c/haproxy # ❯❯❯ curl -w "%{time_total}\n" google.com -o /dev/null -s 0.055 – Cervantez

@Cervantez It generally makes sense to show errors when operating in silent mode (-sS). – Supplementary

Answer 4

The following is inspired by Simon's answer. It's self-contained (doesn't require a separate format file), which makes it great for inclusion into .bashrc.

curl_time() {
    curl -so /dev/null -w "\
   namelookup:  %{time_namelookup}s\n\
      connect:  %{time_connect}s\n\
   appconnect:  %{time_appconnect}s\n\
  pretransfer:  %{time_pretransfer}s\n\
     redirect:  %{time_redirect}s\n\
starttransfer:  %{time_starttransfer}s\n\
-------------------------\n\
        total:  %{time_total}s\n" "$@"
}

Futhermore, it should work with all arguments that curl normally takes, since the "$@" just passes them through. For example, you can do:

curl_time -X POST -H "Content-Type: application/json" -d '{"key": "val"}' https://postman-echo.com/post

Output:

   namelookup:  0,125000s
      connect:  0,250000s
   appconnect:  0,609000s
  pretransfer:  0,609000s
     redirect:  0,000000s
starttransfer:  0,719000s
-------------------------
        total:  0,719000s

Audiology

Answer 5

A shortcut you can add to your .bashrc etc, based on other answers here:

function perf {
  curl -o /dev/null -s -w "%{time_connect} + %{time_starttransfer} = %{time_total}\n" "$1"
}

Usage:

> perf stackoverflow.com
0.521 + 0.686 = 1.290

Aeonian

I use a variation that displays the number of bytes downloaded during the measured time: curl -o /dev/null -s -w "time_total: %{time_total} sec\nsize_download: %{size_download} bytes\n" https://www.google.com – Hayfork

Answer 6

If you want to analyze or summarize the latency you can try apache bench:

ab -n [number of samples] [url]

For example:

ab -n 100 http://www.google.com/

It will show:

This is ApacheBench, Version 2.3 <$Revision: 1757674 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking www.google.com (be patient).....done


Server Software:        gws
Server Hostname:        www.google.com
Server Port:            80

Document Path:          /
Document Length:        12419 bytes

Concurrency Level:      1
Time taken for tests:   10.700 seconds
Complete requests:      100
Failed requests:        97
   (Connect: 0, Receive: 0, Length: 97, Exceptions: 0)
Total transferred:      1331107 bytes
HTML transferred:       1268293 bytes
Requests per second:    9.35 [#/sec] (mean)
Time per request:       107.004 [ms] (mean)
Time per request:       107.004 [ms] (mean, across all concurrent requests)
Transfer rate:          121.48 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:       20   22   0.8     22      26
Processing:    59   85 108.7     68     911
Waiting:       59   85 108.7     67     910
Total:         80  107 108.8     90     932

Percentage of the requests served within a certain time (ms)
  50%     90
  66%     91
  75%     93
  80%     95
  90%    105
  95%    111
  98%    773
  99%    932
 100%    932 (longest request)

Stedmann

This is a fantastic answer. And ab handily accepts a lot of the same flags as curl e.g. -H for headers. I used this command to benchmark the response times of a third party API (supplying the bearer token in an Authorization header). Worked brilliantly. – Cadency

Answer 7

You can use curl -v --trace-time to add timestamps to the trace/verbose output. This must be done in verbose mode or trace mode to do anything.

Fatma

This is awesome! It helped me diagnose a problem with my IPv6. I couldn't see this information from the top-voted answer. 👍 – Boswall

Answer 8

Another way is configuring ~/.curlrc like this

-w "\n\n==== cURL measurements stats ====\ntotal: %{time_total} seconds \nsize: %{size_download} bytes \ndnslookup: %{time_namelookup} seconds \nconnect: %{time_connect} seconds \nappconnect: %{time_appconnect} seconds \nredirect: %{time_redirect} seconds \npretransfer: %{time_pretransfer} seconds \nstarttransfer: %{time_starttransfer} seconds \ndownloadspeed: %{speed_download} byte/sec \nuploadspeed: %{speed_upload} byte/sec \n\n"

So the output of curl is

❯❯ curl -I https://google.com
HTTP/2 301
location: https://www.google.com/
content-type: text/html; charset=UTF-8
date: Mon, 04 Mar 2019 08:02:43 GMT
expires: Wed, 03 Apr 2019 08:02:43 GMT
cache-control: public, max-age=2592000
server: gws
content-length: 220
x-xss-protection: 1; mode=block
x-frame-options: SAMEORIGIN
alt-svc: quic=":443"; ma=2592000; v="44,43,39"



==== cURL measurements stats ====
total: 0.211117 seconds
size: 0 bytes
dnslookup: 0.067179 seconds
connect: 0.098817 seconds
appconnect: 0.176232 seconds
redirect: 0.000000 seconds
pretransfer: 0.176438 seconds
starttransfer: 0.209634 seconds
downloadspeed: 0.000 byte/sec
uploadspeed: 0.000 byte/sec

Industrialist

Would you show me the reference to more detail documents about that? – Babara

@TrầnĐứcTâm details in curl official book ec.haxx.se/usingcurl-writeout.html – Industrialist

Oh this is a great solution, works well for i.e. the nvim-rest plugin (which uses cURL under the hood) – Epirogeny

Answer 9

Another option that is perhaps the simplest one in terms of the command line is adding the built-in --trace-time option:

curl -X POST -d @file server:port --trace-time

Even though it technically does not output the timings of the various steps as requested by the OP, it does display the timestamps for all steps of the request as shown below. Using this, you can (fairly easily) calculate how long each step has taken.

$ curl https://www.google.com --trace-time -v -o /dev/null
13:29:11.148734 * Rebuilt URL to: https://www.google.com/
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
  0     0    0     0    0     0      0      0 --:--:-- --:--:-- --:--:--     013:29:11.149958 *   Trying 172.217.20.36...
13:29:11.149993 * TCP_NODELAY set
13:29:11.163177 * Connected to www.google.com (172.217.20.36) port 443 (#0)
13:29:11.164768 * ALPN, offering h2
13:29:11.164804 * ALPN, offering http/1.1
13:29:11.164833 * successfully set certificate verify locations:
13:29:11.164863 *   CAfile: none
  CApath: /etc/ssl/certs
13:29:11.165046 } [5 bytes data]
13:29:11.165099 * (304) (OUT), TLS handshake, Client hello (1):
13:29:11.165128 } [512 bytes data]
13:29:11.189518 * (304) (IN), TLS handshake, Server hello (2):
13:29:11.189537 { [100 bytes data]
13:29:11.189628 * TLSv1.2 (IN), TLS handshake, Certificate (11):
13:29:11.189658 { [2104 bytes data]
13:29:11.190243 * TLSv1.2 (IN), TLS handshake, Server key exchange (12):
13:29:11.190277 { [115 bytes data]
13:29:11.190507 * TLSv1.2 (IN), TLS handshake, Server finished (14):
13:29:11.190539 { [4 bytes data]
13:29:11.190770 * TLSv1.2 (OUT), TLS handshake, Client key exchange (16):
13:29:11.190797 } [37 bytes data]
13:29:11.190890 * TLSv1.2 (OUT), TLS change cipher, Client hello (1):
13:29:11.190915 } [1 bytes data]
13:29:11.191023 * TLSv1.2 (OUT), TLS handshake, Finished (20):
13:29:11.191053 } [16 bytes data]
13:29:11.204324 * TLSv1.2 (IN), TLS handshake, Finished (20):
13:29:11.204358 { [16 bytes data]
13:29:11.204417 * SSL connection using TLSv1.2 / ECDHE-ECDSA-CHACHA20-POLY1305
13:29:11.204451 * ALPN, server accepted to use h2
13:29:11.204483 * Server certificate:
13:29:11.204520 *  subject: C=US; ST=California; L=Mountain View; O=Google LLC; CN=www.google.com
13:29:11.204555 *  start date: Oct  2 07:29:00 2018 GMT
13:29:11.204585 *  expire date: Dec 25 07:29:00 2018 GMT
13:29:11.204623 *  subjectAltName: host "www.google.com" matched cert's "www.google.com"
13:29:11.204663 *  issuer: C=US; O=Google Trust Services; CN=Google Internet Authority G3
13:29:11.204701 *  SSL certificate verify ok.
13:29:11.204754 * Using HTTP2, server supports multi-use
13:29:11.204795 * Connection state changed (HTTP/2 confirmed)
13:29:11.204840 * Copying HTTP/2 data in stream buffer to connection buffer after upgrade: len=0
13:29:11.204881 } [5 bytes data]
13:29:11.204983 * Using Stream ID: 1 (easy handle 0x55846ef24520)
13:29:11.205034 } [5 bytes data]
13:29:11.205104 > GET / HTTP/2
13:29:11.205104 > Host: www.google.com
13:29:11.205104 > User-Agent: curl/7.61.0
13:29:11.205104 > Accept: */*
13:29:11.205104 > 
13:29:11.218116 { [5 bytes data]
13:29:11.218173 * Connection state changed (MAX_CONCURRENT_STREAMS == 100)!
13:29:11.218211 } [5 bytes data]
13:29:11.251936 < HTTP/2 200 
13:29:11.251962 < date: Fri, 19 Oct 2018 10:29:11 GMT
13:29:11.251998 < expires: -1
13:29:11.252046 < cache-control: private, max-age=0
13:29:11.252085 < content-type: text/html; charset=ISO-8859-1
13:29:11.252119 < p3p: CP="This is not a P3P policy! See g.co/p3phelp for more info."
13:29:11.252160 < server: gws
13:29:11.252198 < x-xss-protection: 1; mode=block
13:29:11.252228 < x-frame-options: SAMEORIGIN
13:29:11.252262 < set-cookie: 1P_JAR=2018-10-19-10; expires=Sun, 18-Nov-2018 10:29:11 GMT; path=/; domain=.google.com
13:29:11.252297 < set-cookie: NID=141=pzXxp1jrJmLwFVl9bLMPFdGCtG8ySQKxB2rlDWgerrKJeXxfdmB1HhJ1UXzX-OaFQcnR1A9LKYxi__PWMigjMBQHmI3xkU53LI_TsYRbkMNJNdxs-caQQ7fEcDGE694S; expires=Sat, 20-Apr-2019 10:29:11 GMT; path=/; domain=.google.com; HttpOnly
13:29:11.252336 < alt-svc: quic=":443"; ma=2592000; v="44,43,39,35"
13:29:11.252368 < accept-ranges: none
13:29:11.252408 < vary: Accept-Encoding
13:29:11.252438 < 
13:29:11.252473 { [5 bytes data]
100 12215    0 12215    0     0   112k      0 --:--:-- --:--:-- --:--:--  112k
13:29:11.255674 * Connection #0 to host www.google.com left intact

Pat

This is actually a great answer which is probably going to fit most of the use-cases people here are looking for. The other answers are great for thorough, in-depth solutions, but this is good for quickly checking round-trip times. – Fellmonger

Thanks @ChrisVandevelde. Yeah, I was aware that there was "something" like this (had used this parameter before), then I googled my way to this SO post and found the more sophisticated form, but... I had a feeling there was another way also. :) Like you say, it's kind of neat in its simplicity and is sometimes well enough for simpler use cases. – Pat

See the timestamps to the left of each step. I overlooked it and was looking for an actual explicit output. – Nelle

Answer 10

Hey is better than Apache Bench, has fewer issues with SSL

./hey https://google.com -more
Summary:
  Total:    3.0960 secs
  Slowest:  1.6052 secs
  Fastest:  0.4063 secs
  Average:  0.6773 secs
  Requests/sec: 64.5992

Response time histogram:
  0.406 [1] |
  0.526 [142]   |∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
  0.646 [1] |
  0.766 [6] |∎∎
  0.886 [0] |
  1.006 [0] |
  1.126 [0] |
  1.246 [12]    |∎∎∎
  1.365 [32]    |∎∎∎∎∎∎∎∎∎
  1.485 [5] |∎
  1.605 [1] |

Latency distribution:
  10% in 0.4265 secs
  25% in 0.4505 secs
  50% in 0.4838 secs
  75% in 1.2181 secs
  90% in 1.2869 secs
  95% in 1.3384 secs
  99% in 1.4085 secs

Details (average, fastest, slowest):
  DNS+dialup:    0.1150 secs, 0.0000 secs, 0.4849 secs
  DNS-lookup:    0.0032 secs, 0.0000 secs, 0.0319 secs
  req write:     0.0001 secs, 0.0000 secs, 0.0007 secs
  resp wait:     0.2068 secs, 0.1690 secs, 0.4906 secs
  resp read:     0.0117 secs, 0.0011 secs, 0.2375 secs

Status code distribution:
  [200] 200 responses

References

• https://github.com/rakyll/hey

Schoolgirl

Answer 11

Generic solution to measuring command execution time

Linux / Mac OS / Unix

On Linux you can measure the time taken for a command to execute with the time command

time curl https://en.wikipedia.org/wiki/Cat 

Output:

ect...
</body>
</html>

real    0m0.565s
user    0m0.011s
sys     0m0.024s

Here the overall time / response time is the real time.

This solution is generic and should work with any Unix command

time sleep 5

Output:

real    0m5.001s
user    0m0.001s
sys 0m0.000s

Windows Powershell

You can use Measure-Command

Floatage

Answer 12

Here's a Bash one-liner to hit the same server repeatedly:

for i in {1..1000}; do curl -s -o /dev/null -w "%{time_total}\n" http://server/get_things; done

Mechling

Answer 13

I made a friendly formatter for sniffing curl requests to help with debugging ( see comments for usage ). It contains's every known output parameter you can write out in an easy to read format.

https://gist.github.com/manifestinteractive/ce8dec10dcb4725b8513

Krimmer

Answer 14

here is the string you can use with -w, contains all options that curl -w supports.

{"contentType":"%{content_type}","filenameEffective":"%{filename_effective}","ftpEntryPath":"%{ftp_entry_path}","httpCode":"%{http_code}","httpConnect":"%{http_connect}","httpVersion":"%{http_version}","localIp":"%{local_ip}","localPort":"%{local_port}","numConnects":"%{num_connects}","numRedirects":"%{num_redirects}","proxySslVerifyResult":"%{proxy_ssl_verify_result}","redirectUrl":"%{redirect_url}","remoteIp":"%{remote_ip}","remotePort":"%{remote_port}","scheme":"%{scheme}","size":{"download":"%{size_download}","header":"%{size_header}","request":"%{size_request}","upload":"%{size_upload}"},"speed":{"download":"%{speed_download}","upload":"%{speed_upload}"},"sslVerifyResult":"%{ssl_verify_result}","time":{"appconnect":"%{time_appconnect}","connect":"%{time_connect}","namelookup":"%{time_namelookup}","pretransfer":"%{time_pretransfer}","redirect":"%{time_redirect}","starttransfer":"%{time_starttransfer}","total":"%{time_total}"},"urlEffective":"%{url_effective}"}

outputs JSON.

Fasta

Prepending \n helps separate the timing when body doesn't end with newline: curl -w '\n{"contentType":"..."}... – Staffan

Answer 15

This is a modified version of Simons answer which makes the multi-lined output a single line. It also introduces the current timestamp so it's easier to follow each line of output.

Sample format fle

$ cat time-format.txt
time_namelookup:%{time_namelookup} time_connect:%{time_connect} time_appconnect:%{time_appconnect} time_pretransfer:%{time_pretransfer} time_redirect:%{time_redirect} time_starttransfer:%{time_starttransfer} time_total:%{time_total}\n

example cmd

$ while [ 1 ];do echo -n "$(date) - " ; curl -w @time-format.txt -o /dev/null -s https://myapp.mydom.com/v1/endpt-http; sleep 1; done | grep -v time_total:0

results

Mon Dec 16 17:51:47 UTC 2019 - time_namelookup:0.004 time_connect:0.015 time_appconnect:0.172 time_pretransfer:0.172 time_redirect:0.000 time_starttransfer:1.666 time_total:1.666
Mon Dec 16 17:51:50 UTC 2019 - time_namelookup:0.004 time_connect:0.015 time_appconnect:0.175 time_pretransfer:0.175 time_redirect:0.000 time_starttransfer:3.794 time_total:3.795
Mon Dec 16 17:51:55 UTC 2019 - time_namelookup:0.004 time_connect:0.017 time_appconnect:0.175 time_pretransfer:0.175 time_redirect:0.000 time_starttransfer:1.971 time_total:1.971
Mon Dec 16 17:51:58 UTC 2019 - time_namelookup:0.004 time_connect:0.014 time_appconnect:0.173 time_pretransfer:0.173 time_redirect:0.000 time_starttransfer:1.161 time_total:1.161
Mon Dec 16 17:52:00 UTC 2019 - time_namelookup:0.004 time_connect:0.015 time_appconnect:0.166 time_pretransfer:0.167 time_redirect:0.000 time_starttransfer:1.434 time_total:1.434
Mon Dec 16 17:52:02 UTC 2019 - time_namelookup:0.004 time_connect:0.015 time_appconnect:0.177 time_pretransfer:0.177 time_redirect:0.000 time_starttransfer:5.119 time_total:5.119
Mon Dec 16 17:52:08 UTC 2019 - time_namelookup:0.004 time_connect:0.014 time_appconnect:0.172 time_pretransfer:0.172 time_redirect:0.000 time_starttransfer:30.185 time_total:30.185
Mon Dec 16 17:52:39 UTC 2019 - time_namelookup:0.004 time_connect:0.014 time_appconnect:0.164 time_pretransfer:0.164 time_redirect:0.000 time_starttransfer:30.175 time_total:30.176
Mon Dec 16 17:54:28 UTC 2019 - time_namelookup:0.004 time_connect:0.015 time_appconnect:3.191 time_pretransfer:3.191 time_redirect:0.000 time_starttransfer:3.212 time_total:3.212
Mon Dec 16 17:56:08 UTC 2019 - time_namelookup:0.004 time_connect:0.015 time_appconnect:1.184 time_pretransfer:1.184 time_redirect:0.000 time_starttransfer:1.215 time_total:1.215
Mon Dec 16 18:00:24 UTC 2019 - time_namelookup:0.004 time_connect:0.015 time_appconnect:0.181 time_pretransfer:0.181 time_redirect:0.000 time_starttransfer:1.267 time_total:1.267

I used the above to catch slow responses on the above endpoint.

Fantastic