nudnik 0.0.12

gRPC load-balancing testing tool

Nudnik

Goals

Allow easy testing of load-balancing and failures in gRPC and REST based service-mesh

Features

• Rate limiting
• Fake IO load from either client or server side
• Export timing metrics to a file or InfluxDB with any custom formatting

Quick Start

• Install Nudnik:

# Install package
pip install nudnik

• Run example Server:

nudnik --server

• Run a client that identifies itself as barbaz, fork to 2 threads, each sending 5 gRPC messages every 3 seconds.

nudnik --name barbaz --streams 2 --interval 3 --rate 5

Docker - Quick Start

• Running on your local/remote docker daemon using docker-compose: docker-compose up

Under The Hood

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 the streams arg.
• message_id: At every interval, the client sends rate 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 field ptime, 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.

Local Development

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 [e-mail]

# "Compile" the entity protobuf
cd nudnik
python -m grpc_tools.protoc --proto_path=./nudnik/ --python_out=./nudnik/ --grpc_python_out=./nudnik/ ./entity.proto

Configure

Via Nudnik Command line args:

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>

Via config file:

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 to 20 threads, each sending 5 gRPC messages every 3 seconds.

nudnik --name foobar --streams 20 --interval 3 --rate 5

• Run a client that identifies itself as FakeFixedLatency, fork to 3 threads, each sending 1 gRPC messages every 10 seconds, and also make the server wait for 0.01 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 to 3 threads, each sending 1 gRPC messages every 10 seconds, and also make the server wait for a random value between 0 and 0.5 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 to 1 threads, each sending 100 gRPC messages every 1 seconds, and also make the server wait for 2ms and fake-load the CPU for 0.5 seconds before replying. Export timing metrics to a output.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