istio/fortio

istio/fortio repository overview

⁠Fortio

Fortio (Φορτίο) started as, and is, Istio⁠'s load testing tool and now graduated to be its own project.

Fortio is also used by, among others, Meshery⁠

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)). It can run for a set duration, for a fixed number of calls, or until interrupted (at a constant target QPS, or max speed/load per connection/thread).

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

Fortio is a fast, small (3Mb docker image, minimal dependencies), 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 (both a single latency graph and a multiple results comparative min, max, avg, qps and percentiles graphs).

Fortio also includes a set of server side features (similar to httpbin) to help debugging and testing: request echo back including headers, adding latency or error codes with a probability distribution, tcp echoing, tcp proxying, http fan out/scatter and gather proxy server, GRPC echo/health in addition to http, etc...

Fortio is quite mature and very stable with no known major bugs (lots of possible improvements if you want to contribute though!), and when bugs are found they are fixed quickly, so after 1 year of development and 42 incremental releases, we reached 1.0 in June 2018.

Fortio components can be used a library even for unrelated projects, for instance the log, stats, or fhttp utilities both client and server. As well as the newly integrated Dynamic Flags⁠ support (greatly inspired/imported initially from https://github.com/mwitkow/go-flagz⁠)

⁠Installation

Install go⁠ (golang 1.16 or later)
go get fortio.org/fortio
you can now run fortio (from your gopath bin/ directory)

Or use docker, for instance:

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

Or download one of the binary distributions, from the releases⁠ assets page or for instance:

curl -L https://github.com/fortio/fortio/releases/download/v1.17.0/fortio-linux_x64-1.17.0.tgz \
 | sudo tar -C / -xvzpf -
# or the debian package
wget https://github.com/fortio/fortio/releases/download/v1.17.0/fortio_1.17.0_amd64.deb
dpkg -i fortio_1.17.0-1_amd64.deb
# or the rpm
rpm -i https://github.com/fortio/fortio/releases/download/v1.17.0/fortio-1.17.0-1.x86_64.rpm

On a MacOS you can also install Fortio using Homebrew⁠:

brew install fortio

On Windows, download https://github.com/fortio/fortio/releases/download/v1.17.0/fortio_win_1.17.0.zip⁠ and extract all to some location then using the Windows Command Prompt:

cd fortio
fortio.exe server

(at the prompt, allow the windows firewall to let connections in)

Once fortio server is running, you can visit its web UI at http://localhost:8080/fortio/⁠

You can get a preview of the reporting/graphing UI at https://fortio.istio.io/⁠ and on istio.io/docs/performance-and-scalability/synthetic-benchmarks/⁠

⁠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, result graphing, tcp/udp echo, proxies, https redirector, with the server command or issue grpc ping messages using the grpcping command. It can also fetch a single URL's for debugging when using the curl command (or the -curl flag to the load command). Likewise you can establish a single TCP (or unix domain or UDP (use udp:// prefix)) connection using the nc command (like the standalone netcat package). You can run just the redirector with redirect or just the tcp echo with tcp-echo. 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. The version command will print version and build information, fortio version -s just the version. Lastly, you can learn which flags are available using help command.

Most important flags for http load generation:

Flag	Description, example
`-qps rate`	Queries Per Seconds or 0 for no wait/max qps
`-c connections`	Number of parallel simultaneous connections (and matching go routine)
`-t duration`	How long to run the test (for instance `-t 30m` for 30 minutes) or 0 to run until ^C, example (default 5s)
`-n numcalls`	Run for exactly this number of calls instead of duration. Default (0) is to use duration (-t).
`-r resolution`	Resolution of the histogram lowest buckets in seconds (default 0.001 i.e 1ms), use 1/10th of your expected typical latency
`-H "header: value"`	Can be specified multiple times to add headers (including Host:)
`-a`	Automatically save JSON result with filename based on labels and timestamp
`-json filename`	Filename or `-` for stdout to output json result (relative to `-data-dir` by default, should end with .json if you want `fortio report` to show them; using `-a` is typicallly a better option)
`-labels "l1 l2 ..."`	Additional config data/labels to add to the resulting JSON, defaults to target URL and hostname

You can switch from http GET queries to POST by setting -content-type or passing one of the -payload-* option.

Full list of command line flags (fortio help):

Φορτίο 1.17.0 usage:
        fortio command [flags] target
where command is one of: load (load testing), server (starts ui, http-echo,
redirect, proxies, tcp-echo and grpc ping servers), tcp-echo (only the tcp-echo
server), report (report only UI server), redirect (only the redirect server),
proxies (only the -M and -P configured proxies), grpcping (grpc client),
or curl (single URL debug), or nc (single tcp or udp:// connection).
where target is a url (http load tests) or host:port (grpc health test).
flags are:
  -H header
        Additional header(s)
  -L    Follow redirects (implies -std-client) - do not use for load test
  -M value
        Http multi proxy to run, e.g -M "localport1 baseDestURL1 baseDestURL2"
-M ...
  -P value
        Tcp proxies to run, e.g -P "localport1 dest_host1:dest_port1" -P
"[::1]:0 www.google.com:443" ...
  -a    Automatically save JSON result with filename based on labels & timestamp
  -abort-on code
        Http code that if encountered aborts the run. e.g. 503 or -1 for socket
errors.
  -allow-initial-errors
        Allow and don't abort on initial warmup errors
  -base-url URL
        base URL used as prefix for data/index.tsv generation. (when empty, the
url from the first request is used)
  -c int
        Number of connections/goroutine/threads (default 4)
  -cacert Path
        Path to a custom CA certificate file to be used for the TLS client
connections, if empty, use https:// prefix for standard internet/system CAs
  -cert Path
        Path to the certificate file to be used for client or server TLS
  -compression
        Enable http compression
  -config path
        Config directory path to watch for changes of dynamic flags (empty for
no watch)
  -content-type string
        Sets http content type. Setting this value switches the request method
from GET to POST.
  -curl
        Just fetch the content once
  -data-dir Directory
        Directory where JSON results are stored/read (default ".")
  -echo-debug-path URI
        http echo server URI for debug, empty turns off that part (more secure)
(default "/debug")
  -echo-server-default-params value
        Default parameters/querystring to use if there isn't one provided
explicitly. E.g "status=404&delay=3s"
  -gomaxprocs int
        Setting for runtime.GOMAXPROCS, <1 doesn't change the default
  -grpc
        Use GRPC (health check by default, add -ping for ping) for load testing
  -grpc-max-streams uint
        MaxConcurrentStreams for the grpc server. Default (0) is to leave the
option unset.
  -grpc-ping-delay duration
        grpc ping delay in response
  -grpc-port port
        grpc server port. Can be in the form of host:port, ip:port or port or
/unix/domain/path or "disabled" to not start the grpc server. (default "8079")
  -halfclose
        When not keepalive, whether to half close the connection (only for fast
http)
  -health
        grpc ping client mode: use health instead of ping
  -healthservice string
        which service string to pass to health check
  -http-port port
        http echo server port. Can be in the form of host:port, ip:port, port
or /unix/domain/path. (default "8080")
  -http1.0
        Use http1.0 (instead of http 1.1)
  -httpbufferkb kbytes
        Size of the buffer (max data size) for the optimized http client in
kbytes (default 128)
  -httpccch
        Check for Connection: Close Header
  -https-insecure
        Long form of the -k flag
  -jitter
        set to true to de-synchronize parallel clients' requests
  -json path
        Json output to provided file path or '-' for stdout (empty = no json
output, unless -a is used)
  -k    Do not verify certs in https connections
  -keepalive
        Keep connection alive (only for fast http 1.1) (default true)
  -key Path
        Path to the key file matching the -cert
  -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 "> ")
  -max-echo-delay value
        Maximum sleep time for delay= echo server parameter. dynamic flag.
(default 1.5s)
  -maxpayloadsizekb Kbytes
        MaxPayloadSize is the maximum size of payload to be generated by the
EchoHandler size= argument. In Kbytes. (default 256)
  -multi-mirror-origin
        Mirror the request url to the target for multi proxies (-M) (default
true)
  -multi-serial-mode
        Multi server (-M) requests one at a time instead of parallel mode
  -n int
        Run for exactly this number of calls instead of duration. Default (0)
is to use duration (-t). Default is 1 when used as grpc ping count.
  -nc-dont-stop-on-eof
        in netcat (nc) mode, don't abort as soon as remote side closes
  -p string
        List of pXX to calculate (default "50,75,90,99,99.9")
  -payload string
        Payload string to send along
  -payload-file path
        File path to be use as payload (POST for http), replaces -payload when
set.
  -payload-size int
        Additional random payload size, replaces -payload when set > 0, must be
smaller than -maxpayloadsizekb. Setting this switches http to POST.
  -ping
        grpc load test: use ping instead of health
  -profile file
        write .cpu and .mem profiles to file
  -qps float
        Queries Per Seconds or 0 for no wait/max qps (default 8)
  -quiet
        Quiet mode: sets the loglevel to Error and reduces the output.
  -r float
        Resolution of the histogram lowest buckets in seconds (default 0.001)
  -redirect-port port
        Redirect all incoming traffic to https URL (need ingress to work
properly). Can be in the form of host:port, ip:port, port or "disabled" to
disable the feature. (default "8081")
  -resolve string
        Resolve CN of cert to this IP, so that we can call https://cn directly
  -s int
        Number of streams per grpc connection (default 1)
  -static-dir path
        Deprecated/unused path.
  -stdclient
        Use the slower net/http standard client (works for TLS)
  -sync URL
        index.tsv or s3/gcs bucket xml URL to fetch at startup for server modes.
  -sync-interval duration
        Refresh the url every given interval (default, no refresh)
  -t duration
        How long to run the test or 0 to run until ^C (default 5s)
  -tcp-port port
        tcp echo server port. Can be in the form of host:port, ip:port, port or
/unix/domain/path or "disabled". (default "8078")
  -timeout duration
        Connection and read timeout value (for http) (default 3s)
  -udp-async
        if true, udp echo server will use separate go routine to reply
  -udp-port port
        udp echo server port. Can be in the form of host:port, ip:port, port or
"disabled". (default "8078")
  -udp-timeout duration
        Udp timeout (default 750ms)
  -ui-path URI
        http server URI for UI, empty turns off that part (more secure)
(default "/fortio/")
  -unix-socket path
        Unix domain socket path to use for physical connection
  -user user:password
        User credentials for basic authentication (for http). Input data format
should be user:password

See also the FAQ entry about fortio flags for best results⁠

⁠Example use and output

⁠Start the internal servers

$ fortio server &
14:11:05 I fortio_main.go:171> Not using dynamic flag watching (use -config to set watch directory)
Fortio 1.17.0 tcp-echo server listening on [::]:8078
Fortio 1.17.0 grpc 'ping' server listening on [::]:8079
Fortio 1.17.0 https redirector server listening on [::]:8081
Fortio 1.17.0 echo server listening on [::]:8080
Data directory is /Users/ldemailly/go/src/fortio.org/fortio
UI started - visit:
http://localhost:8080/fortio/
(or any host/ip reachable on this server)
14:11:05 I fortio_main.go:233> All fortio 1.17.0 release go1.16.5 servers started!

⁠Change the port / binding address

By default, Fortio's web/echo servers listen on port 8080 on all interfaces. Use the -http-port flag to change this behavior:

$ fortio server -http-port 10.10.10.10:8088
UI starting - visit:
http://10.10.10.10:8088/fortio/
Https redirector running on :8081
Fortio 1.17.0 grpc ping server listening on port :8079
Fortio 1.17.0 echo server listening on port 10.10.10.10:8088

⁠Unix domain sockets

You can use unix domain socket for any server/client:

$ fortio server --http-port /tmp/fortio-uds-http &
Fortio 1.17.0 grpc 'ping' server listening on [::]:8079
Fortio 1.17.0 https redirector server listening on [::]:8081
Fortio 1.17.0 echo server listening on /tmp/fortio-uds-http
UI started - visit:
fortio curl -unix-socket=/tmp/fortio-uds-http http://localhost/fortio/
14:58:45 I fortio_main.go:217> All fortio 1.17.0 unknown go1.16.5 servers started!
$ fortio curl -unix-socket=/tmp/fortio-uds-http http://foo.bar/debug
15:00:48 I http_client.go:428> Using unix domain socket /tmp/fortio-uds-http instead of foo.bar http
HTTP/1.1 200 OK
Content-Type: text/plain; charset=UTF-8
Date: Wed, 08 Aug 2018 22:00:48 GMT
Content-Length: 231

Φορτίο version 1.17.0 unknown go1.16.5 echo debug server up for 2m3.4s on ldemailly-macbookpro - request from

GET /debug HTTP/1.1

headers:

Host: foo.bar
User-Agent: fortio.org/fortio-1.17.0

body:

⁠TCP

Start the echo-server alone and run a load (use tcp:// prefix for the load test to be for tcp echo server)

$ fortio tcp-echo &
Fortio 1.17.0 tcp-echo TCP server listening on [::]:8078
19:45:30 I fortio_main.go:238> All fortio 1.17.0 release go1.16.5 servers started!
$ fortio load -qps -1 -n 100000 tcp://localhost:8078
Fortio 1.17.0 running at -1 queries per second, 16->16 procs, for 100000 calls: tcp://localhost:8078
20:01:31 I tcprunner.go:218> Starting tcp test for tcp://localhost:8078 with 4 threads at -1.0 qps
Starting at max qps with 4 thread(s) [gomax 16] for exactly 100000 calls (25000 per thread + 0)
20:01:32 I periodic.go:558> T003 ended after 1.240585427s : 25000 calls. qps=20151.77629520873
20:01:32 I periodic.go:558> T002 ended after 1.241141084s : 25000 calls. qps=20142.75437521493
20:01:32 I periodic.go:558> T001 ended after 1.242066385s : 25000 calls. qps=20127.7486468648
20:01:32 I periodic.go:558> T000 ended after 1.24227731s : 25000 calls. qps=20124.331176909283
Ended after 1.242312567s : 100000 calls. qps=80495
Aggregated Function Time : count 100000 avg 4.9404876e-05 +/- 1.145e-05 min 2.7697e-05 max 0.000887051 sum 4.94048763
# range, mid point, percentile, count
>= 2.7697e-05 <= 0.000887051 , 0.000457374 , 100.00, 100000
# target 50% 0.00045737
# target 75% 0.00067221
# target 90% 0.000801115
# target 99% 0.000878457
# target 99.9% 0.000886192
Sockets used: 4 (for perfect no error run, would be 4)
Total Bytes sent: 2400000, received: 2400000
tcp OK : 100000 (100.0 %)
All done 100000 calls (plus 0 warmup) 0.049 ms avg, 80495.0 qps

⁠UDP

Start the udp-echo server alone and run a load (use tcp:// prefix for the load test to be for tcp echo server)

$ fortio udp-echo &
Fortio 1.17.0 udp-echo UDP server listening on [::]:8078
21:54:52 I fortio_main.go:273> Note: not using dynamic flag watching (use -config to set watch directory)
21:54:52 I fortio_main.go:281> All fortio 1.17.0 release go1.16.5 servers started!
$ fortio load -qps -1 -n 100000 udp://localhost:8078/
Fortio 1.17.0 running at -1 queries per second, 16->16 procs, for 100000 calls: udp://localhost:8078/
21:56:48 I udprunner.go:222> Starting udp test for udp://localhost:8078/ with 4 threads at -1.0 qps
Starting at max qps with 4 thread(s) [gomax 16] for exactly 100000 calls (25000 per thread + 0)
21:56:49 I periodic.go:558> T003 ended after 969.635695ms : 25000 calls. qps=25782.879208051432
21:56:49 I periodic.go:558> T000 ended after 969.906228ms : 25000 calls. qps=25775.687667818544
21:56:49 I periodic.go:558> T002 ended after 970.543935ms : 25000 calls. qps=25758.751457243405
21:56:49 I periodic.go:558> T001 ended after 970.737665ms : 25000 calls. qps=25753.610786287973
Ended after 970.755702ms : 100000 calls. qps=1.0301e+05
Aggregated Function Time : count 100000 avg 3.8532238e-05 +/- 1.7e-05 min 2.0053e-05 max 0.000881827 sum 3.85322376
# range, mid point, percentile, count
>= 2.0053e-05 <= 0.000881827 , 0.00045094 , 100.00, 100000
# target 50% 0.000450936
# target 75% 0.000666381
# target 90% 0.000795649
# target 99% 0.000873209
# target 99.9% 0.000880965
Sockets used: 4 (for perfect no error run, would be 4)
Total Bytes sent: 2400000, received: 2400000
udp OK : 100000 (100.0 %)
All done 100000 calls (plus 0 warmup) 0.039 ms avg, 103012.5 qps

⁠GRPC

⁠Simple grpc ping

$ fortio grpcping -n 5 localhost
22:36:55 I pingsrv.go:150> Ping RTT 212000 (avg of 259000, 217000, 160000 ns) clock skew -10500
22:36:55 I pingsrv.go:150> Ping RTT 134333 (avg of 170000, 124000, 109000 ns) clock skew 5000
22:36:55 I pingsrv.go:150> Ping RTT 112000 (avg of 111000, 122000, 103000 ns) clock skew 5000
22:36:55 I pingsrv.go:150> Ping RTT 157000 (avg of 136000, 158000, 177000 ns) clock skew 6000
22:36:55 I pingsrv.go:150> Ping RTT 108333 (avg of 118000, 106000, 101000 ns) clock skew 1000
Clock skew histogram usec : count 5 avg 1.3 +/- 6.145 min -10.5 max 6 sum 6.5
# range, mid point, percentile, count
>= -10.5 <= -10 , -10.25 , 20.00, 1
> 0 <= 2 , 1 , 40.00, 1
> 4 <= 6 , 5 , 100.00, 3
# target 50% 4.33333
RTT histogram usec : count 15 avg 144.73333 +/- 44.48 min 101 max 259 sum 2171
# range, mid point, percentile, count
>= 101 <= 110 , 105.5 , 26.67, 4
> 110 <= 120 , 115 , 40.00, 2
> 120 <= 140 , 130 , 60.00, 3
> 140 <= 160 , 150 , 73.33, 2
> 160 <= 180 , 170 , 86.67, 2
> 200 <= 250 , 225 , 93.33, 1
> 250 <= 259 , 254.5 , 100.00, 1
# target 50% 130

⁠Change the target port for grpc

The value of -grpc-port (default 8079) is used when specifying a hostname or an IP address in grpcping. Add :port to the grpcping destination to change this behavior:

$ fortio grpcping 10.10.10.100:8078 # Connects to gRPC server 10.10.10.100 listening on port 8078
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

⁠`grpcping` using TLS

First, start Fortio server with the -cert and -key flags:

/path/to/fortio/server.crt and /path/to/fortio/server.key are paths to the TLS certificate and key that you must provide.

$ fortio server -cert /path/to/fortio/server.crt -key /path/to/fortio/server.key
UI starting - visit:
http://localhost:8080/fortio/
Https redirector running on :8081
Fortio 1.17.0 grpc ping server listening on port :8079
Fortio 1.17.0 echo server listening on port localhost:8080
Using server certificate /path/to/fortio/server.crt to construct TLS credentials
Using server key /path/to/fortio/server.key to construct TLS credentials

Next, use grpcping with the -cacert flag:

/path/to/fortio/ca.crt is the path to the CA certificate that issued the server certificate for localhost. In our example, the server certificate is /path/to/fortio/server.crt:

$ fortio grpcping -cacert /path/to/fortio/ca.crt localhost
Using server certificate /path/to/fortio/ca.crt to construct TLS credentials
16:00:10 I pingsrv.go:129> Ping RTT 501452 (avg of 595441, 537088, 371828 ns) clock skew 31094
Clock skew histogram usec : count 1 avg 31.094 +/- 0 min 31.094 max 31.094 sum 31.094
# range, mid point, percentile, count
>= 31.094 <= 31.094 , 31.094 , 100.00, 1
# target 50% 31.094
RTT histogram usec : count 3 avg 501.45233 +/- 94.7 min 371.828 max 595.441 sum 1504.357
# range, mid point, percentile, count
>= 371.828 <= 400 , 385.914 , 33.33, 1
> 500 <= 595.441 , 547.721 , 100.00, 2
# target 50% 523.86

⁠GRPC to standard https service

grpcping can connect to a non-Fortio TLS server by prefacing the destination with https://:

$ fortio grpcping https://fortio.istio.io
11:07:55 I grpcrunner.go:275> stripping https scheme. grpc destination: fortio.istio.io. grpc port: 443
Clock skew histogram usec : count 1 avg 12329.795 +/- 0 min 12329.795 max 12329.795 sum 12329.795
# range, mid point, percentile, count
>= 12329.8 <= 12329.8 , 12329.8 , 100.00, 1
# target 50% 12329.8

⁠Simple load test

Load (low default qps/threading) test:

$ fortio load http://www.google.com
Fortio 1.17.0 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

Tag summary

Recent tags

Content type

Image

Digest

Size

4.4 MB

Last updated

about 5 years ago

docker pull istio/fortio

This week's pulls

Pulls:

196

Last week

Learn more⁠