eunomia-exporter

eunomia-exporter

An prometheus and OpenTelemetry exporter for custom eBPF metrics, written in async rust: eunomia-exporter

This is a single binary exporter, you don't need to install BCC/LLVM when you use it. The only thing you will need to run the exporter on another machine is the config file and pre-compiled eBPF code.

example

This is an adapted version of opensnoop from bcc/libbpf-tools, you can check our source code here: bpftools/examples/opensnoop

You can just download the pre-compiled opensnoop package.json.

Or you can compile the opensnoop like this:

$ cd bpftools/examples/opensnoop
$ docker run -it -v /userpath/eunomia-bpf/bpftools/examples/opensnoop:/src yunwei37/ebpm:latest

userpath needs to be replaced with your own repo path.

After compile the eBPF code, you can define a config file like this:

programs:
- name: opensnoop
  metrics:
    counters:
    - name: eunomia_file_open_counter
      description: test
      labels:
      - name: pid
      - name: comm
      - name: filename
        from: fname
  compiled_ebpf_filename: package.json

use the path to package.json as compiled_ebpf_filename in the config file. You can find the example at config.yaml.

Then, you can start the exporter:

$ ls
config.yaml  eunomia-exporter package.json
$ sudo ./eunomia-exporter

Running ebpf program opensnoop takes 46 ms
Listening on http://127.0.0.1:8526
running and waiting for the ebpf events from perf event...
Receiving request at path /metrics

Different from the bcc ebpf_exporter, the only thing you need to run on the deployment machine is the config file and package.json. There is no need to install LLVM/CLang for BCC.

The result is:

manage eBPF tracing program via API

start an eBPF exporter via web API:

$ curl -X POST http://127.0.0.1:8526/start -H "Content-Type: application/json" -d @examples/opensnoop/curl_post_example.json

{"id":1}

see curl_post_example.json for the example of the request body.

list all running eBPF programs:

$ curl http://127.0.0.1:8526/list

[{"id":0,"name":"bootstrap"},{"id":1,"name":"opensnoop"}]

stop an eBPF program:

$ curl -X POST http://127.0.0.1:8526/stop -H "Content-Type: application/json" -d '{"id": 1}'

build

Notice: You must compile eunomia-bpf before build eunomia-exporter. Details in build.md

This component was written by rust,Please install rust before build.

$ curl -sSf https://static.rust-lang.org/rustup.sh | sh

You can compile the rust code in this folder:

$ cargo build --release
$ target/release/eunomia-exporter -h
eunomia-exporter 0.1.0

USAGE:
    eunomia-exporter [OPTIONS]

OPTIONS:
    -c, --config <CONFIG>    Sets a custom config file [default: config.yaml]
    -h, --help               Print help information
    -V, --version            Print version information

benchmark

Take opensnoop from bcc/libbpf-tools as an example. starting with BCC, you will need about 0.8s to start the exporter and attach to the probe. With out implement, you only need about 50-70ms which is significantly faster.

$ ps -aux | grep eunomia
root      171562  0.0  0.0  15176  4576 pts/6    S+   01:08   0:00 sudo ./eunomia-exporter
root      171605  0.1  0.0 350540  7740 pts/6    Sl+  01:08   0:00 ./eunomia-exporter

The memory usage and CPU usage is also low.

Supported scenarios

Currently the only supported way of getting data out of the kernel is via maps (we call them tables in configuration).