fortio

README

Φορτίο

Φορτίο (fortio) is Istio's load testing tool. Fortio runs at a specified query per second (qps) and records an histogram of execution time and calculates percentiles (e.g. p99 ie the response time such as 99% of the requests take less than that number (in seconds, SI unit))

The name fortio comes from greek φορτίο which is load/burden.

Fortio is a reusable, embeddable go library as well as a command line tool and server process, the server includes a simple web UI and graphical representation of the results.

Installation

1. Install go (golang 1.8 or later)
2. go get -u istio.io/fortio
3. you can now run fortio (from your gopath bin/ directory)

Or use docker, for instance:

docker run -p 8080:8080 -p 8079:8079 istio/fortio server & # For the server
docker run istio/fortio load http://www.google.com/ # For a test run

Or download the binary distribution, for instance:

curl -L https://github.com/istio/fortio/releases/download/0.4.0/fortio-linux_x64-0.4.0.tgz \
 | sudo tar -C / -xvzpf -

You can visit the new web UI at http://localhost:8080/fortio/

Command line arguments

Fortio can be an http or grpc load generator, gathering statistics using the load subcommand, or start simple http and grpc ping servers, as well as a basic web UI and result graphing, with the server command or issue grpc ping messages using the grpcping command. It can also fetch a single URL's content using the -curl flag to the load command. Lastly if you saved JSON results (using the web UI or directly from the command line), you can browse and graph those results using the report command.

$ fortio
Φορτίο 0.5.1 usage:
	fortio command [flags] target
where command is one of: load (load testing), server (starts grpc ping and http echo/ui servers), grpcping (grpc client)
or report (report only UI server), where target is a url (http load tests) or host:port (grpc health test)
and flags are:
  -H value
    	Additional Header(s)
  -a	Automatically save JSON result with filename based on labels and timestamp
  -allow-initial-errors
    	Allow and don't abort on initial warmup errors
  -c int
    	Number of connections/goroutine/threads (default 4)
  -compression
    	Enable http compression
  -curl
    	Just fetch the content once
  -data-dir string
    	Directory where JSON results are stored/read (default ".")
  -echo-debug-path string
    	http echo server URI for debug, empty turns off that part (more secure) (default "/debug")
  -gomaxprocs int
    	Setting for runtime.GOMAXPROCS, <1 doesn't change the default
  -grpc
    	Use GRPC (health check) for load testing
  -grpc-port int
    	grpc port (default 8079)
  -halfclose
    	When not keepalive, whether to half close the connection (only for fast http)
  -health
    	client mode: use health instead of ping
  -healthservice string
    	which service string to pass to health check
  -http-port int
    	http echo server port (default 8080)
  -http1.0
    	Use http1.0 (instead of http 1.1)
  -httpbufferkb int
    	Size of the buffer (max data size) for the optimized http client in kbytes (default 128)
  -httpccch
    	Check for Connection: Close Header
  -httpreqtimeout duration
      	Http request timeout value (default 15s)
  -json string
    	Json output to provided file or '-' for stdout (empty = no json output, unless -a is used)
  -keepalive
    	Keep connection alive (only for fast http 1.1) (default true)
  -labels string
    	Additional config data/labels to add to the resulting JSON, defaults to target URL and hostname
  -logcaller
    	Logs filename and line number of callers to log (default true)
  -loglevel value
    	loglevel, one of [Debug Verbose Info Warning Error Critical Fatal] (default Info)
  -logprefix string
    	Prefix to log lines before logged messages (default "> ")
  -n int
    	how many ping(s) the client will send (default 1)
  -p string
    	List of pXX to calculate (default "50,75,99,99.9")
  -payload string
    	Payload string to send along
  -profile string
    	write .cpu and .mem profiles to file
  -qps float
    	Queries Per Seconds or 0 for no wait/max qps (default 8)
  -r float
    	Resolution of the histogram lowest buckets in seconds (default 0.001)
  -static-dir string
    	Absolute path to the dir containing the static files dir
  -stdclient
    	Use the slower net/http standard client (works for TLS)
  -t duration
    	How long to run the test or 0 to run until ^C (default 5s)
  -ui-path string
    	http server URI for UI, empty turns off that part (more secure) (default "/fortio/")

Example use and output

• Start the internal servers:

$ fortio server &
UI starting - visit:
http://localhost:8080/fortio/
Fortio 0.3.6 echo server listening on port 8080
Fortio 0.3.6 grpc ping server listening on port 8079

• Simple grpc ping:

$ fortio grpcping localhost
02:29:27 I pingsrv.go:116> Ping RTT 305334 (avg of 342970, 293515, 279517 ns) clock skew -2137
Clock skew histogram usec : count 1 avg -2.137 +/- 0 min -2.137 max -2.137 sum -2.137
# range, mid point, percentile, count
>= -4 < -2 , -3 , 100.00, 1
# target 50% -2.137
RTT histogram usec : count 3 avg 305.334 +/- 27.22 min 279.517 max 342.97 sum 916.002
# range, mid point, percentile, count
>= 250 < 300 , 275 , 66.67, 2
>= 300 < 350 , 325 , 100.00, 1
# target 50% 294.879

• Load (low default qps/threading) test:

$ fortio load http://www.google.com
Fortio 0.3.6 running at 8 queries per second, 8->8 procs, for 5s: http://www.google.com
19:10:33 I httprunner.go:84> Starting http test for http://www.google.com with 4 threads at 8.0 qps
Starting at 8 qps with 4 thread(s) [gomax 8] for 5s : 10 calls each (total 40)
19:10:39 I periodic.go:314> T002 ended after 5.056753279s : 10 calls. qps=1.9775534712220633
19:10:39 I periodic.go:314> T001 ended after 5.058085991s : 10 calls. qps=1.9770324224999916
19:10:39 I periodic.go:314> T000 ended after 5.058796046s : 10 calls. qps=1.9767549252963101
19:10:39 I periodic.go:314> T003 ended after 5.059557593s : 10 calls. qps=1.9764573910247019
Ended after 5.059691387s : 40 calls. qps=7.9056
Sleep times : count 36 avg 0.49175757 +/- 0.007217 min 0.463508712 max 0.502087879 sum 17.7032725
Aggregated Function Time : count 40 avg 0.060587641 +/- 0.006564 min 0.052549016 max 0.089893269 sum 2.42350566
# range, mid point, percentile, count
>= 0.052549 < 0.06 , 0.0562745 , 47.50, 19
>= 0.06 < 0.07 , 0.065 , 92.50, 18
>= 0.07 < 0.08 , 0.075 , 97.50, 2
>= 0.08 <= 0.0898933 , 0.0849466 , 100.00, 1
# target 50% 0.0605556
# target 75% 0.0661111
# target 99% 0.085936
# target 99.9% 0.0894975
Code 200 : 40
Response Header Sizes : count 40 avg 690.475 +/- 15.77 min 592 max 693 sum 27619
Response Body/Total Sizes : count 40 avg 12565.2 +/- 301.9 min 12319 max 13665 sum 502608
All done 40 calls (plus 4 warmup) 60.588 ms avg, 7.9 qps

• Report only UI

If you have json files saved from running the full UI, you can serve just the reports:

$ fortio report
Browse only UI starting - visit:
http://localhost:8080/

Implementation details

Fortio is written in the Go language and includes a scalable semi log histogram in stats.go and a periodic runner engine in periodic.go with specializations for http and grpc. The http/ package includes a very high performance specialized http 1.1 client. You may find fortio's logger useful as well.

You can run the histogram code standalone as a command line in histogram/, a basic echo http server in echosrv/, or both the http echo and GRPC ping server through fortio server, the fortio command line interface lives in this top level directory fortio_main.go

Another example output

With 5k qps: (includes envoy and mixer in the calls)

$ time fortio load -qps 5000 -t 60s -c 8 -r 0.0001 -H "Host: perf-cluster" http://benchmark-2:9090/echo
2017/07/09 02:31:05 Will be setting special Host header to perf-cluster
Fortio running at 5000 queries per second for 1m0s: http://benchmark-2:9090/echo
Starting at 5000 qps with 8 thread(s) [gomax 4] for 1m0s : 37500 calls each (total 300000)
2017/07/09 02:32:05 T004 ended after 1m0.000907812s : 37500 calls. qps=624.9905437680746
2017/07/09 02:32:05 T000 ended after 1m0.000922222s : 37500 calls. qps=624.9903936684861
2017/07/09 02:32:05 T005 ended after 1m0.00094454s : 37500 calls. qps=624.9901611965524
2017/07/09 02:32:05 T006 ended after 1m0.000944816s : 37500 calls. qps=624.9901583216429
2017/07/09 02:32:05 T001 ended after 1m0.00102094s : 37500 calls. qps=624.9893653892883
2017/07/09 02:32:05 T007 ended after 1m0.001096292s : 37500 calls. qps=624.9885805003184
2017/07/09 02:32:05 T003 ended after 1m0.001045342s : 37500 calls. qps=624.9891112105419
2017/07/09 02:32:05 T002 ended after 1m0.001044416s : 37500 calls. qps=624.9891208560392
Ended after 1m0.00112695s : 300000 calls. qps=4999.9
Aggregated Sleep Time : count 299992 avg 8.8889218e-05 +/- 0.002326 min -0.03490402 max 0.001006041 sum 26.6660543
# range, mid point, percentile, count
< 0 , 0 , 8.58, 25726
>= 0 < 0.001 , 0.0005 , 100.00, 274265
>= 0.001 < 0.002 , 0.0015 , 100.00, 1
# target 50% 0.000453102
WARNING 8.58% of sleep were falling behind
Aggregated Function Time : count 300000 avg 0.00094608764 +/- 0.0007901 min 0.000510522 max 0.029267604 sum 283.826292
# range, mid point, percentile, count
>= 0.0005 < 0.0006 , 0.00055 , 0.15, 456
>= 0.0006 < 0.0007 , 0.00065 , 3.25, 9295
>= 0.0007 < 0.0008 , 0.00075 , 24.23, 62926
>= 0.0008 < 0.0009 , 0.00085 , 62.73, 115519
>= 0.0009 < 0.001 , 0.00095 , 85.68, 68854
>= 0.001 < 0.0011 , 0.00105 , 93.11, 22293
>= 0.0011 < 0.0012 , 0.00115 , 95.38, 6792
>= 0.0012 < 0.0014 , 0.0013 , 97.18, 5404
>= 0.0014 < 0.0016 , 0.0015 , 97.94, 2275
>= 0.0016 < 0.0018 , 0.0017 , 98.34, 1198
>= 0.0018 < 0.002 , 0.0019 , 98.60, 775
>= 0.002 < 0.0025 , 0.00225 , 98.98, 1161
>= 0.0025 < 0.003 , 0.00275 , 99.21, 671
>= 0.003 < 0.0035 , 0.00325 , 99.36, 449
>= 0.0035 < 0.004 , 0.00375 , 99.47, 351
>= 0.004 < 0.0045 , 0.00425 , 99.57, 290
>= 0.0045 < 0.005 , 0.00475 , 99.66, 280
>= 0.005 < 0.006 , 0.0055 , 99.79, 380
>= 0.006 < 0.007 , 0.0065 , 99.82, 92
>= 0.007 < 0.008 , 0.0075 , 99.83, 15
>= 0.008 < 0.009 , 0.0085 , 99.83, 5
>= 0.009 < 0.01 , 0.0095 , 99.83, 1
>= 0.01 < 0.012 , 0.011 , 99.83, 8
>= 0.012 < 0.014 , 0.013 , 99.84, 35
>= 0.014 < 0.016 , 0.015 , 99.92, 231
>= 0.016 < 0.018 , 0.017 , 99.94, 65
>= 0.018 < 0.02 , 0.019 , 99.95, 26
>= 0.02 < 0.025 , 0.0225 , 100.00, 139
>= 0.025 < 0.03 , 0.0275 , 100.00, 14
# target 50% 0.000866935
# target 75% 0.000953452
# target 99% 0.00253875
# target 99.9% 0.0155152
Code 200 : 300000
Response Body Sizes : count 300000 avg 0 +/- 0 min 0 max 0 sum 0

Or you can get the data in JSON format (using -json result.json)

Or graphically (through the http://localhost:8080/fortio/ web UI):

Simple form/UI:

Sample requests with responses delayed by 250us and 0.5% of 503 and 1.5% of 429 simulated http errors.

Run result:

Code 200 : 2939 (98.0 %)
Code 429 : 47 (1.6 %)
Code 503 : 14 (0.5 %)

Contributing

Contributions whether throuh issues, documentation, bug fixes, or new features are most welcome !

Please also see Contributing to Istio

And make sure to go format and run those commands successfully before sending your PRs:

make test
make lint
make webtest

When modifying Javascript:

standard --fix fortio_chart.js

See also

Our wiki and the Fortio FAQ (including for instance differences between fortio and wrk or httpbin)