Allow easy testing of load-balancing and failures in gRPC and REST based service-mesh
- Rate limiting
- Fake IO load from either client or server side
- Export timing metrics to a file or InfluxDB with any custom formatting
- Install Nudnik:
# Install package
pip install nudnik
- Run example Server:
nudnik --server
- Run a client that identifies itself as
barbaz
, fork to2
threads, each sending5
gRPC messages every3
seconds.
nudnik --name barbaz --streams 2 --interval 3 --rate 5
- Running on your local/remote docker daemon using docker-compose:
docker-compose up
The server binds to the host:port that were specified, and listens to incoming gRPC messages. Every message consists of several fields:
- name: The name of the client, arbitrary string
- stream_id: The stream id, the streams are numbered from
0
to whatever was configureed via thestreams
arg. - message_id: At every
interval
, the client sendsrate
messages, this id is an autoincrement index for that message. - ctime: The timestamp at which the request was created
- rtime: The timestamp at which the request was retransmitted (0 if not applicapble)
- rcount: The amount of times this message was retransmitted (0 if not applicapble)
- meta: Just another string field that does nothing, you may send additional arbitrary data to increase message size or specify
random
to generate random data with every request - load: This field may be repeated several times, it instructs the server to create fake load of some sort before replying.
Upon recieveing a message, the server will:
- parse the
load
field - perform the required load
- print a log message using the configured
out_format
- reply an
OK
to the client, also add the fieldptime
, which represents the time at which the request was parsed at the server, this allows the client to calculate exact RTT, even if NTP is not synchronized.
As a rule of thumb wed'e recommend using virtualenv. Requirement are Python 3.7 + requirements.txt file
- Clone and initialize Nudnik:
# Install python requirements
pip install grpcio grpcio-tools requests-unixsocket
# Clone and configure the repository
git clone https://github.com/salosh/nudnik.git
git config --global push.default matching
git config --global user.name "Your Name"
git config --global user.email [email protected]
# "Compile" the entity protobuf
cd nudnik
python -m grpc_tools.protoc --proto_path=./nudnik/ --python_out=./nudnik/ --grpc_python_out=./nudnik/ ./entity.proto
nudnik -h
usage: nudnik [-h] [--config-file CONFIG_FILE] [--host HOST] [--port PORT]
[--server] [--name NAME] [--name-mismatch-error] [--meta META]
[--streams STREAMS] [--interval INTERVAL] [--rate RATE]
[--load load_type load_value] [--retry-count RETRY_COUNT]
[--fail-ratio FAIL_RATIO] [--metrics {stdout,file,influxdb}]
[--file-path FILE_PATH]
[--influxdb-socket-path INFLUXDB_SOCKET_PATH]
[--influxdb-database-name INFLUXDB_DATABASE_NAME] [--debug]
[--verbose] [--version]
Nudnik - gRPC load tester
optional arguments:
-h, --help show this help message and exit
--config-file CONFIG_FILE, -f CONFIG_FILE
Path to YAML config file
--host HOST, -H HOST host
--port PORT, -p PORT port
--server, -S Operation mode (default: client)
--name NAME, -n NAME Parser name
--name-mismatch-error
Fail request on name mismatch (default: False)
--meta META, -M META Send this extra data with every request
--streams STREAMS, -s STREAMS
Number of streams (Default: 1)
--interval INTERVAL, -i INTERVAL
Number of seconds per stream message cycle (Default:
1)
--rate RATE, -r RATE Number of messages per interval (Default: 10)
--load load_type load_value, -l load_type load_value
Add artificial load [rtt, rttr, cpu, mem] (Default:
None)
--retry-count RETRY_COUNT
Number of times to re-send failed messages (Default:
-1, which means infinite times)
--fail-ratio FAIL_RATIO
Percent of requests to intentionally fail (Default: 0)
--metrics {stdout,file,influxdb}, -m {stdout,file,influxdb}
Enable metrics outputs (Default: None)
--file-path FILE_PATH, -F FILE_PATH
Path to exported metrics file (Default:
./nudnikmetrics.out)
--influxdb-socket-path INFLUXDB_SOCKET_PATH
Absolute path to InfluxDB Unix socket (Default:
/var/run/influxdb/influxdb.sock)
--influxdb-database-name INFLUXDB_DATABASE_NAME
InfluxDB database name (Default: nudnikmetrics)
--debug, -d Debug mode (default: False)
--verbose, -v Verbose mode (default: None)
--version, -V Display Nudnik version
2018 (C) Salo Shp <https://github.com/salosh/nudnik.git>
nano ./config.yml
- Run example Server that fails
14%
of all incoming requests:
nudnik --server --fail-ratio 14
- Run a client that identifies itself as
foobar
, fork to20
threads, each sending5
gRPC messages every3
seconds.
nudnik --name foobar --streams 20 --interval 3 --rate 5
- Run a client that identifies itself as
FakeFixedLatency
, fork to3
threads, each sending1
gRPC messages every10
seconds, and also make the server wait for0.01
seconds before replying.
nudnik --name FakeFixedLatency --streams 3 --interval 10 --rate 1 --load rtt 0.01
- Run a client that identifies itself as
FakeRandomLatency
, fork to3
threads, each sending1
gRPC messages every10
seconds, and also make the server wait for a random value between0
and0.5
seconds before replying. Export timing metrics to InfluxDB.
nudnik --name FakeRandomLatency --streams 3 --interval 10 --rate 1 --load rttr 0.5 --metrics influxdb
- Run a client that identifies itself as
FakeLatencyAndCPU
, fork to1
threads, each sending100
gRPC messages every1
seconds, and also make the server wait for2ms
and fake-load the CPU for0.5
seconds before replying. Export timing metrics to aoutput.csv
.
nudnik --name FakeLatencyAndCPU --streams 1 --interval 1 --rate 100 --load rtt 0.002 --load cpu 0.5 --metrics file --file-path ./output.csv
Visit our website at https://salosh.org