Skip to content

gse (Go Scheduler Exporter) exports the Go runtime scheduler traces to Prometheus.

License

Notifications You must be signed in to change notification settings

MadhavJivrajani/gse

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

13 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

gse

gse stands for Go Scheduler Exporter and contains the source and setup required to run the demo that was done as part of the talk Queues, Fairness, and The Go Scheduler. The slides for the talk can be found here.

Working

On a high level:

  • gse exports scheduler traces for a particular binary to Prometheus to help observe and visualize the Go runtime scheduler.
  • It takes a config file which looks as follows:
    path: "cmd to execute binary"
    prometheus:
      endpoint: metrics
      port: 8080
    sched:
      interval: 10 # frequency in ms at which scheduler traces will be emitted.
    Please see the comments here for more details on these fields.
  • The metrics exported are as follows:
    • Run queue lengths of local runqueues
    • Run queue length of global runqueue
    • No. of idle procs
    • No. of threads
    • No. of spinning threads
    • No. of idle threads
  • These metrics can now be viewed in grafana.

Running gse

Get and build gse

git clone https://github.com/MadhavJivrajani/gse.git && cd gse
go build .
# optional
mv ./gse /usr/local/bin
if permission denied then use export path instead 
export PATH=$PATH:$PWD/gse

Run gse

An example program and config file can be found here.

Compile the program

go build -o ./example/gophercon ./example/main.go

Running gse

Before actually running gse, make sure you have prometheus running whose config matches the prometheus part of the config provided to gse. The config I used and passed to a locally running prometheus was:

global:
  scrape_interval: 10ms

scrape_configs:
  - job_name: "sched"
    static_configs:
      - targets: ["localhost:8080"]

The 10ms scrape interval is provided since 10ms is also the timeslice given to a gorotuine, post which async preemption is attempted, and hence visualization becomes significantly clearer.

Actually run gse now:

gse -f example/test.yaml

Once running, you can open up grafana and add prometheus as a data source and visualize the exported metrics:

grafana

Tracing SIGURG

This part is orthogonal to gse, but here for reference! To trace when a long running Gorutine is asynchronously preempted using the SIGURG signal, we can make use of the Linux tracing subsystem to get events whenever the SIGURG signal is issued.

To set things up, you need to be a root user first. Once you are, run the following:

./example/preemption_tracing/setup.sh

In our example, the name of the binary was gophercon. If we search through the trace log for gophercon, we should be able to see the preemption events.

cat /sys/kernel/debug/tracing/trace | grep gophercon

About

gse (Go Scheduler Exporter) exports the Go runtime scheduler traces to Prometheus.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published