Context
ASF Infra disabled GitHub Actions on apache/hudi on 2026-08-04 over CI runner-time consumption (discussion #19515). #19523 and #19514 are the interim fix — reset required_status_checks and trim the matrix — to get merging again within the ASF GitHub Actions policy (concurrency MUST ≤20, SHOULD ≤15; 4,200 runner-hours/week cap). The trim buys headroom by deleting coverage ([CI-TRIM] lanes, commented in place, not restored). This epic tracks getting to CI that stays within budget, is fast, and covers what we ship.
Streams
Unblock (in flight) — merge the interim fix. #19525, #19526.
Runner budget & concurrency — get under the SHOULD target and stop paying for what a change doesn't touch. Possible directions: cap concurrency (#19527), cache deps/build output across jobs (#19529), path-filter scoped changes (#19530).
Wall clock — shrink time to first signal and to full green. Possible directions: rebalance jobs by measured duration (#19528), fix/quarantine flaky tests so retries don't double-pay (#19534).
Test coverage — restore and broaden coverage lost in the trim. Possible directions: per-version Spark coverage (#19531), compile-only for parked Flink 1.18–2.0 (#19532), bundle validation for Flink/Kafka Connect/metaserver (#19533) and Presto (#19468).
Cleanup — keep the CI config itself maintainable. Possible directions: drop validate-ci-baseline once no pre-trim PR is open (#19535), keep per-job contexts out of required_status_checks, restore [CI-TRIM] lanes opportunistically (cheap compile/smoke over full suites for older versions).
Tracking
Watch weekly runner-hour usage against the 4,200 cap so a regression is visible to us before Infra.
Related
Context
ASF Infra disabled GitHub Actions on apache/hudi on 2026-08-04 over CI runner-time consumption (discussion #19515). #19523 and #19514 are the interim fix — reset
required_status_checksand trim the matrix — to get merging again within the ASF GitHub Actions policy (concurrency MUST ≤20, SHOULD ≤15; 4,200 runner-hours/week cap). The trim buys headroom by deleting coverage ([CI-TRIM]lanes, commented in place, not restored). This epic tracks getting to CI that stays within budget, is fast, and covers what we ship.Streams
Unblock (in flight) — merge the interim fix. #19525, #19526.
Runner budget & concurrency — get under the SHOULD target and stop paying for what a change doesn't touch. Possible directions: cap concurrency (#19527), cache deps/build output across jobs (#19529), path-filter scoped changes (#19530).
Wall clock — shrink time to first signal and to full green. Possible directions: rebalance jobs by measured duration (#19528), fix/quarantine flaky tests so retries don't double-pay (#19534).
Test coverage — restore and broaden coverage lost in the trim. Possible directions: per-version Spark coverage (#19531), compile-only for parked Flink 1.18–2.0 (#19532), bundle validation for Flink/Kafka Connect/metaserver (#19533) and Presto (#19468).
Cleanup — keep the CI config itself maintainable. Possible directions: drop
validate-ci-baselineonce no pre-trim PR is open (#19535), keep per-job contexts out ofrequired_status_checks, restore[CI-TRIM]lanes opportunistically (cheap compile/smoke over full suites for older versions).Tracking
Watch weekly runner-hour usage against the 4,200 cap so a regression is visible to us before Infra.
Related