You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Measure the test account cost for two weeks after 2026-09-26 and compare it with the two weeks before.
Cost Explorer (us-west-2, before tax), per day:
S3 PutObject request count and cost
Athena SQL cost and Spark DPU-hours
Glue Data Catalog requests
The number of Test workflow runs per day, split by event (pull_request, schedule, workflow_dispatch) and conclusion, and the number of AWS jobs per run. This gives the cost per run and per job.
After: 2026-09-29 to 2026-10-12, with no further cost changes planned. Measure on or after 2026-10-13.
Because some of the earlier changes landed inside the "before" window, it is a mixed baseline. Also compare against 2026-09-01 to 2026-09-14, which predates all of #834.
What to compare, from Cost Explorer with daily granularity, for the test account:
S3: PutObject requests and cost
Athena: SQL and Spark DPU
Glue requests
Total
Normalize by activity with the Test workflow runs in each window (gh run list --workflow test.yaml): the number of runs and of AWS jobs, split by event (pull_request, schedule, dispatch, release). Report the cost per run as well as the totals.
Measurement results (Cost Explorer, daily granularity, test account).
The test account started incurring test costs on 2026-09-18 (2026-09-02 to 2026-09-17 are close to $0 per day), so the 2026-09-01 to 2026-09-14 comparison does not apply. The "before" window is 2026-09-18 to 2026-09-28, and the "after" window is the same length, 2026-09-29 to 2026-10-09 (measured on 2026-10-10 instead of waiting for 2026-10-12).
About an 87% reduction per day, from roughly $530 to roughly $70 per month.
The highest day in the "after" window was $9.0 (2026-10-03), compared with $30 to $37 on the busiest "before" days.
Test workflow runs went from 243 to 758, mostly Draft PRs that now run lint only without AWS. AWS jobs were not counted per run, so no cost-per-run figure is given.
The reduction is clear, so this closes the measurement.
Activity
laughingman7743 commentedon Sep 28, 2026
Measurement window, now that the last cost-related changes of #834 have merged:
Because some of the earlier changes landed inside the "before" window, it is a mixed baseline. Also compare against 2026-09-01 to 2026-09-14, which predates all of #834.
What to compare, from Cost Explorer with daily granularity, for the test account:
Normalize by activity with the Test workflow runs in each window (
gh run list --workflow test.yaml): the number of runs and of AWS jobs, split by event (pull_request, schedule, dispatch, release). Report the cost per run as well as the totals.#834 closes after this measurement.
laughingman7743 commentedon Oct 10, 2026
Measurement results (Cost Explorer, daily granularity, test account).
The test account started incurring test costs on 2026-09-18 (2026-09-02 to 2026-09-17 are close to $0 per day), so the 2026-09-01 to 2026-09-14 comparison does not apply. The "before" window is 2026-09-18 to 2026-09-28, and the "after" window is the same length, 2026-09-29 to 2026-10-09 (measured on 2026-10-10 instead of waiting for 2026-10-12).
Per day by service (before → after):
The reduction is clear, so this closes the measurement.