Strest client and server implementations for gRPC.
To run the client and server locally, first start the server.
$ strest-grpc server
starting gRPC server on :11111
Next run the client. By default, the client will send as many request as it can on a single connection. Kill it to see a final performance report.
$ strest-grpc client --address localhost:11111 --latencyUnit us
2018-08-10T10:45:10-07:00 0 0.0B 39594/0 10s L: 132 [388 629 ] 7383 J: 0 0
{
"good": 42074,
"bad": 0,
"bytes": 0,
"latency": {
"p50": 215,
"p75": 250,
"p90": 317,
"p95": 397,
"p99": 634,
"p999": 7383
},
"jitter": {
"p50": 0,
"p75": 0,
"p90": 0,
"p95": 0,
"p99": 0,
"p999": 0
}
}
$ strest-grpc --help
A load tester for stress testing grpc intermediaries.
Find more information at https://github.com/buoyantio/strest-grpc.
Usage:
strest-grpc [command]
Available Commands:
client run the strest-grpc client
help Help about any command
max-rps compute max RPS
ref-client run a gRPC reference client
ref-server run a gRPC reference server
server run the strest-grpc server
Flags:
-h, --help help for strest-grpc
-l, --logLevel string log level, must be one of: panic, fatal, error, warn, info, debug (default "info")
Use "strest-grpc [command] --help" for more information about a command.
$ strest-grpc client --help
run the strest-grpc client
Usage:
strest-grpc client [flags]
Flags:
--address string address of strest-grpc service or intermediary (default "localhost:11111")
--clientTimeout duration timeout for unary client requests. Default: no timeout
--connections uint number of concurrent connections (default 1)
--errorRate float the chance to return an error
-h, --help help for client
--interval duration reporting interval (default 10s)
--iterations uint number of iterations for the experiment. Exits gracefully after `iterations * interval` (default 0, meaning infinite)
--latencyPercentiles string response latency percentile distribution. (e.g. 50=10,100=100) (default "100=0")
--latencyUnit string latency units [ms|us|ns] (default "ms")
--lengthPercentiles string response body length percentile distribution. (e.g. 50=100,100=1000) (default "100=0")
--metricAddr string address to serve metrics on
--noFinalReport do not print a final JSON output report
--noIntervalReport only print the final report, nothing intermediate
--streaming use the streaming features of strest server
--streamingRatio string the ratio of streaming requests/responses (default "1:1")
--streams uint number of concurrent streams per connection (default 1)
--tlsTrustChainFile string the path to the certificate used to validate the remote's signature
--totalRequests uint total number of requests to send. default: infinite
--totalTargetRps uint target requests per second
-u, --unix use Unix Domain Sockets instead of TCP
Global Flags:
-l, --logLevel string log level, must be one of: panic, fatal, error, warn, info, debug (default "info")
$ strest-grpc server --help
run the strest-grpc server
Usage:
strest-grpc server [flags]
Flags:
--address string address to serve on (default ":11111")
-h, --help help for server
--latencyPercentiles string response latency percentile distribution added to client latencies. (e.g. 50=10,100=100) (default "100=0")
--metricAddr string address to serve metrics on
--tlsCertFile string the path to the trust certificate
--tlsPrivKeyFile string the path to the server's private key
-u, --unix use Unix Domain Sockets instead of TCP
Global Flags:
-l, --logLevel string log level, must be one of: panic, fatal, error, warn, info, debug (default "info")
$ strest-grpc max-rps --help
compute max RPS
Usage:
strest-grpc max-rps [flags]
Flags:
--address string hostname:port of strest-grpc service or intermediary (default "localhost:11111")
--concurrencyLevels string levels of concurrency to test with (default "1,5,10,20,30")
-h, --help help for max-rps
--timePerLevel duration how much time to spend testing each concurrency level (default 1s)
Global Flags:
-l, --logLevel string log level, must be one of: panic, fatal, error, warn, info, debug (default "info")
$ strest-grpc ref-client --help
run a gRPC reference client
Usage:
strest-grpc ref-client [flags]
Flags:
-h, --help help for ref-client
Global Flags:
-l, --logLevel string log level, must be one of: panic, fatal, error, warn, info, debug (default "info")
$ strest-grpc ref-server --help
run a gRPC reference server
Usage:
strest-grpc ref-server [flags]
Flags:
-h, --help help for ref-server
Global Flags:
-l, --logLevel string log level, must be one of: panic, fatal, error, warn, info, debug (default "info")
It's also possible to run the client and server in Kubernetes, using the provided configs in the config/ directory.
First create the strest-server deployment and service.
kubectl apply -f config/server.yml
Then create the strest-client deployment.
kubectl apply -f config/client.yml
View the logs from the client pod to see the strest performance report.
kubectl logs -c strest-client deploy/strest-client
To build the strest-grpc binaries and archives, run:
./bin/release.sh [VERSION TAG]
That will create strest-grpc binaries and archives in ./release.
To build a docker image, run:
$ docker build -t buoyantio/strest-grpc:latest .
Replace latest with whatever tag you are trying to build.
strest-grpc max-rps will calculate the maximum RPS that a strest-server or intermediary
can sustain. The maxConcurrency score lets you know what at which point adding more
clients no longer improves overall throughput. This calculation is based on the
Universal Scalability Law and can change dramatically based on the environment and
resources available.
To release:
- Update and submit a PR with a changelog for the release in CHANGES.md.
- Merge the changelog PR.
- Build the strest-grpc archives as described in the Building section above.
- Use Github to create a new release.
- Add the archives that you built as attachments to the release.
- The docker image is built automatically once the release is created.