build/perf
Michael Anthony Knyszek e107a4861d perf: detangle benchseries comparison from influx upload
Right now the benchseries comparison performed on benchmark data and the
influx upload code is deeply intertwined. This makes it hard to test the
benchseries comparison code, and also makes it hard to just try out
one-off comparisons for debugging.

This change splits the logic of these two using a Go 1.23 iterator
(though they're not available in this repository yet).

For #69594.

Change-Id: Ibc104a7abf711b89e5f3a2dced91f6cd7dc9e882
Reviewed-on: https://go-review.googlesource.com/c/build/+/615137
Auto-Submit: Michael Knyszek <mknyszek@google.com>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
2024-09-30 16:36:56 +00:00
..
app perf: detangle benchseries comparison from influx upload 2024-09-30 16:36:56 +00:00
Dockerfile
Makefile
README.md
deployment-prod.yaml
main.go

README.md

Performance analysis system

This package contains the https://perf.golang.org/ benchmark result analysis system. It serves as a front-end to the benchmark result storage system at https://perfdata.golang.org/.

The storage system is designed to have a standardized REST API at https://perfdata.golang.org/, and we encourage additional analysis tools to be written against the API. An example client can be found in the perfdata package.

Local

Both storage and analysis can be run locally; the following commands will run the complete stack on your machine with an in-memory datastore.

To run the storage system:

$ go install golang.org/x/build/perfdata/localperfdata@latest
$ localperfdata -addr=:8081 -view_url_base=http://localhost:8080/search?q=upload: &

To run the analysis frontend:

$ make docker-prod
$ docker run --rm --net=host gcr.io/symbolic-datum-552/perf:latest -listen-http=:8080 -perfdata=http://localhost:8081

Browse to https://localhost:8080 (note that the instance uses a self-signed certificate).

To run the analysis frontend connected to a local InfluxDB instance, first follow the instructions to set one up. Then, run the above command but with the -influx-host and -influx-token flags set to their appropriate values (as specified by the InfluxDB setup instructions).

Send an HTTP request to https://localhost:8080/cron/sync-influx to sync the local instances together.

Google Cloud

One-time setup:

  1. IAM setup, based on https://cloud.google.com/kubernetes-engine/docs/how-to/workload-identity#authenticating_to:

a. Create GCP service account:

$ gcloud iam service-accounts create perf-prod \
    --description="Runs golang.org/x/build/perf"

c. Allow Kubernetes service account (created by deployment-prod.yaml) to impersonate the GCP service account:

$ gcloud iam service-accounts add-iam-policy-binding \
    perf-prod@<PROJECT>.iam.gserviceaccount.com \
    --role roles/iam.workloadIdentityUser \
    --member "serviceAccount:<PROJECT>.svc.id.goog[prod/perf-prod]"

Deployment

See the documentation on deployment.