Awesome
yab 
yab (Yet Another Benchmarker) is a tool to benchmark YARPC services. It
currently supports making Thrift requests to both HTTP and TChannel services, as well as Protobuf requests to gRPC services.
yab is currently in beta status.
Installing
If you have go installed, simply run the following to install the latest version:
GO111MODULE=on go get -u github.com/yarpc/yab
This will install yab to $GOPATH/bin/yab.
Optionally, you can get precompiled binaries from Releases.
Usage
Usage:
yab [<service> <method> <body>] [OPTIONS]
yab is a benchmarking tool for TChannel and HTTP applications. It's primarily
intended for Thrift applications but supports other encodings like JSON and
binary (raw). It can be used in a curl-like fashion when benchmarking features
are disabled.
yab includes a full man page (man yab), which is also available online:
http://yarpc.github.io/yab/man.html
Application Options:
-v Enable more detailed logging. Repeats increase
the verbosity, ie. -vvv
--version Displays the application version
Request Options:
-e, --encoding= The encoding of the data, options are: Thrift,
proto, JSON, raw. Defaults to Thrift if the
method contains '::' or a Thrift file is
specified. Defaults to proto if the method
contains '/' or a proto filedescriptorset is
specified
-t, --thrift= Path of the .thrift file
-F, --file-descriptor-set-bin= A binary file containing a compiled protobuf
FileDescriptorSet.
--procedure= The full method name to invoke (Thrift:
Svc::Method, Proto: package.Service/Method).
-m, --method= Alias for procedure
-r, --request= The request body, in JSON or YAML format
-f, --file= Path of a file containing the request body in
JSON or YAML
-H, --header= Individual application header as a key:value
pair per flag
--headers= The headers in JSON or YAML format
--headers-file= Path of a file containing the headers in JSON
or YAML
-B, --baggage= Individual context baggage header as a
key:value pair per flag
--health Hit the health endpoint, Meta::health
--timeout= The timeout for each request. E.g., 100ms,
0.5s, 1s. If no unit is specified,
milliseconds are assumed. (default: 1s)
-y, --yaml-template= Send a tchannel request specified by a YAML
template
-A, --arg= A list of key-value template arguments,
specified as -A foo:bar -A user:me
--disable-thrift-envelope Disables Thrift envelopes (disabled by default
for TChannel and gRPC)
--multiplexed-thrift Enables the Thrift TMultiplexedProtocol used
by services that host multiple Thrift services
on a single endpoint.
--stream-interval= Interval between consecutive stream message sends,
applicable separately to every stream request
opened on a connection.
--stream-delay-close-send= Delay the closure of send stream once all the
request messages have been sent.
Transport Options:
-s, --service= The TChannel/Hyperbahn service name
-p, --peer= The host:port of the service to call
-P, --peer-list= Path or URL of a JSON, YAML, or flat file
containing a list of host:ports. -P? for
supported protocols.
--caller= Caller will override the default caller name
(which is yab-$USER).
--rk= The routing key overrides the service name
traffic group for proxies.
--rd= The routing delegate overrides the routing key
traffic group for proxies.
--sk= The shard key is a transport header that clues
where to send a request within a clustered
traffic group.
--jaeger Use the Jaeger tracing client to send Uber
style traces and baggage headers
--force-jaeger-sample If Jaeger tracing is enabled with --jaeger, force all requests
to be sampled.
-T, --topt= Transport options for TChannel, protocol
headers for HTTP
--http-method= The HTTP method to use (default: POST)
--grpc-max-response-size= Maximum response size for gRPC requests. Default value is 4MB
Benchmark Options:
-n, --max-requests= The maximum number of requests to make. 0
implies no limit. (default: 0)
-d, --max-duration= The maximum amount of time to run the
benchmark for. 0 implies no duration limit.
(default: 0s)
--cpus= The number of OS threads
--connections= The number of TCP connections to use
--warmup= The number of requests to make to warmup each
connection (default: 10)
--concurrency= The number of concurrent calls per connection
(default: 1)
--rps= Limit on the number of requests per second.
The default (0) is no limit. (default: 0)
--statsd= Optional host:port of a StatsD server to
report metrics
--per-peer-stats Whether to emit stats by peer rather than
aggregated
Help Options:
-h, --help Show this help message
Making a single request
Thrift
The following examples assume that the Thrift service running looks like:
service KeyValue {
string get(1: string key)
}
If a TChannel service was running with name keyvalue on localhost:12345, you can
make a call to the get method by running:
yab -t ~/keyvalue.thrift -p localhost:12345 keyvalue KeyValue::get -r '{"key": "hello"}'
Proto
The following examples assume that the Proto service running looks like:
service KeyValue {
rpc GetValue(Request) returns (Response) {}
rpc GetValueStream(stream Request) returns (stream Response) {}
}
message Request {
required string key = 1;
}
message Response {
required string value = 1;
}
If a gRPC service was running with name KeyValue on localhost:12345 with proto package name pkg.keyvalue and
binary file containing a compiled protobuf FileDescriptorSet as keyValue.proto.bin, you can
make a call to the GetValue method by running:
yab keyvalue pkg.keyvalue/GetValue --file-descriptor-set-bin=keyValue.proto.bin -r '{"key": "hello"}' -p localhost:12345
You can make a call to the bi-directional stream method GetValueStream with multiple requests by running:
yab keyvalue pkg.keyvalue/GetValueStream --file-descriptor-set-bin=keyValue.proto.bin -r '{"key": "hello1"} {"key": "hello2"}' -p localhost:12345
You can also interactively pass request data (JSON or YAML) on STDIN to the bi-directional stream method GetValueStream by setting option -request='-':
yab keyvalue pkg.keyvalue/GetValueStream --file-descriptor-set-bin=keyValue.proto.bin -r '-' -p localhost:12345
{"key": "hello1"} // STDIN request-1
{...} //Response-1
{"key": "hello2"} // STDIN request-2
{...} //Response-2
Note: YAML requests on STDIN must be delimited by --- and followed by a newline.
Protobuf FileDescriptorSet can be generated by running:
protoc --include_imports --descriptor_set_out=keyValue.proto.bin keyValue.proto
Note : If Server Reflection is enabled which provides information about publicly-accessible gRPC services on a server, then there is no need to specify the FileDescriptorSet binary:
yab keyvalue pkg.keyvalue/GetValue -r '{"key": "hello"} -p localhost:12345'
Specifying Peers
A single host:port is specified using -p, but you can also specify multiple peers
by passing the -p flag multiple times:
yab -t ~/keyvalue.thrift -p localhost:12345 -p localhost:12346 keyvalue KeyValue::get -r '{"key": "hello"}'
yab keyvalue pkg.keyvalue/GetValue -r '{"key": "hello"} -p localhost:12345 -p localhost:12346'
If you have a file containing a list of host:ports (either JSON or new line separated), you can
specify the file using -P:
yab -t ~/keyvalue.thrift -P ~/hosts.json keyvalue KeyValue::get -r '{"key": "hello"}'
yab keyvalue pkg.keyvalue/GetValue -r '{"key": "hello"} -P ~/hosts.json'
yab also supports HTTP, instead of the peer being a single host:port, you would use a URL:
yab -t ~/keyvalue.thrift -p "http://localhost:8080/rpc" keyvalue KeyValue::get -r '{"key": "hello"}'
Benchmarking
To benchmark an endpoint, you need all the command line arguments to describe the request,
followed by benchmarking options. You need to set at least --maxDuration (or -d) to
set the maximum amount of time to run the benchmark.
You can set values such as 3s for 3 seconds, or 1m for 1 minute. Valid time units are:
msfor millisecondssfor secondsmfor minutes.
You can also control rate limit the benchmark (--rps), or customize the number of
connections (--connections) or control the amount of concurrent calls per
connection (--concurrency).
yab -t ~/keyvalue.thrift -p localhost:12345 keyvalue KeyValue::get -r '{"key": "hello"}' -d 5s --rps 100 --connections 4
In a gRPC stream method benchmark, a stream benchmark request is considered successful when a stream sends all the requests and receives response messages successfully. Example stream benchmark command and output:
> yab keyvalue pkg.keyvalue/GetValueStream -r '{"key": "hello1"} {"key": "hello2"}' -p localhost:12345 --duration=1s
Benchmark parameters:
CPUs: 12
Connections: 24
Concurrency: 1
Max requests: 10000
Max duration: 1s
Max RPS: 0
Latencies:
0.5000: 779.01µs
0.9000: 2.034852ms
0.9500: 2.932846ms
0.9900: 11.698821ms
0.9990: 15.839751ms
0.9995: 16.651223ms
1.0000: 17.198644ms
Elapsed time (seconds): 0.49
Total requests: 10000
RPS: 20190.25
Total stream messages sent: 20000
Total stream messages received: 20000