rpc-perf was created to help measure the performance of caching systems. We've found this tool to be useful for validating performance of cache backends, effects of kernel version and system tuning, as well as testing new hardware platforms and network changes.
BEWARE rpc-perf can write to its target and can generate many requests
- run only if data in the server can be lost/destroyed/corrupted/etc
- run only if you understand the impact of sending high-levels of traffic across your network
- Getting rpc-perf
- Sample Usage
- Sample Output
- Future Work
rpc-perf is built through the
cargo command which ships with rust. If you don't have Rust
installed, you can use rustup to manage your Rust installation. Otherwise, follow the
instructions on rust-lang.org to get Rust and Cargo installed.
rpc-perf targets stable Rust.
Build from source
With rust installed, clone this repo, and cd into this folder:
git clone https://github.com/twitter/rpc-perf.git cd rpc-perf/rpc-perf cargo build --release
If you need TLS support, you'll need to use nightly Rust:
git clone https://github.com/twitter/rpc-perf.git cd rpc-perf/rpc-perf rustup override set nightly cargo build --release --features tls
This will produce a binary at
../target/release/rpc-perf which can be run in-place or copied to a
more convenient location on your system.
rpc-perf is configured through a combination of a TOML config file and command line parameters. If
an option is specified in both the config file and on the command line, the command line wins. See
--help and the example configurations in
rpc-perf/configs to learn more about configuration.
BEWARE use caution when running rpc-perf
# display help rpc-perf --help # use a config file and specify an endpoint rpc-perf --config some_config.toml --endpoint 127.0.0.1:11211 # use a config file and override the request rate rpc-perf --config some_config.toml --endpoint 127.0.0.1:11211 --request-rate 200000 # use a config file and override the protocol rpc-perf --config some_config.toml --endpoint 127.0.0.1:6379 --protocol redis # generate a waterfall plot of request latency rpc-perf --config some_config.toml --endpoint 127.0.0.1:11211 --interval 60 --windows 5 --waterfall waterfall.png
listen option in the
general section of your TOML
config to enable HTTP based stats exposition. This will allow for scraping the
metrics provided by rpc-perf into Prometheus or other compatible observability
stack. A typical use case would be for long-running tests where you wish to
correlate client metrics with system or service metrics.
admin option in the
general section of your TOML config
to enable a HTTP admin endpoint. You can use this endpoint to change the request
PUT requests. For example, if configured with port
the admin port:
curl -X PUT -d 100 127.0.0.1:40404/ratelimit/request would
update the current rate to 100 requests per second. To use this, you must set a
request ratelimit when launching rpc-perf.
- Start with a short test before moving on to tests spanning larger periods of time
- If comparing latency between two setups, be sure to set a ratelimit that's achievable on both
--clientsbelow the number of cores on the machine generating workload
--poolsizeas necessary to simulate production-like connection numbers
- You may need to use multiple machines to generate enough workload and/or connections to the target
- Log your configuration and results to make repeating and sharing experiments easy
- Use waterfalls to help visualize latency distribution over time and see anomalies
- high-resolution latency metrics
- supports memcache and redis protocols
- mio for async networking
- optional waterfall visualization of latencies
- powerful workload configuration
Create a new issue on GitHub.
We feel that a welcoming community is important and we ask that you follow Twitter's Open Source Code of Conduct in all interactions with the community.
- Brian Martin firstname.lastname@example.org
A full list of contributors can be found on GitHub.
Follow @TwitterOSS on Twitter for updates.
Copyright 2015-2019 Twitter, Inc.
Licensed under the Apache License, Version 2.0: https://www.apache.org/licenses/LICENSE-2.0
Please report sensitive security issues via Twitter's bug-bounty program (https://hackerone.com/twitter) rather than GitHub.