lab/e2e.sh: the whole lab from nothing (up.sh, the chart with the images release/suite.yaml names, seeded), then mail, the webmail across replicas and the console, timing each part; the summary goes to the GitHub job summary too.
lab/tools.sh: kind v0.33.0, kubectl v1.37.2, Helm v4.3.2 for amd64/arm64, each checked against its published SHA-256.
.github/workflows/e2e.yml: both on ubuntu-24.04 and ubuntu-24.04-arm, on the GitHub replica only (vars.BUILD_ON). Chart from inbuxa-charts main, or the branch in vars.CHART_REF: set to fix/hook-network-policies until inbuxa/inbuxa-charts#34 merges, then removed.
up.sh now does what a fresh machine needs: Helm repositories, MetalLB (pool computed from kind's network), KIND_NODE_IMAGE, and waits for two cold-start races (Kyverno's webhook, Garage's status) that the reused bench cluster never hit.
README: the suite's header, badges and notes.
Bench, from scratch: 179 s in all (cluster and data layer 149 s, chart install 27 s, the three checks seconds each), with inbuxa/inbuxa-charts#34. The first from-scratch runs found the two races and the chart bug fixed in #34.
S7: the GitHub runs on this branch answer it (time and memory on both architectures); I'll put the numbers here and on #7.
- `lab/e2e.sh`: the whole lab from nothing (up.sh, the chart with the images `release/suite.yaml` names, seeded), then mail, the webmail across replicas and the console, timing each part; the summary goes to the GitHub job summary too.
- `lab/tools.sh`: kind v0.33.0, kubectl v1.37.2, Helm v4.3.2 for amd64/arm64, each checked against its published SHA-256.
- `.github/workflows/e2e.yml`: both on `ubuntu-24.04` and `ubuntu-24.04-arm`, on the GitHub replica only (`vars.BUILD_ON`). Chart from inbuxa-charts `main`, or the branch in `vars.CHART_REF`: set to `fix/hook-network-policies` until inbuxa/inbuxa-charts#34 merges, then removed.
- `up.sh` now does what a fresh machine needs: Helm repositories, MetalLB (pool computed from kind's network), `KIND_NODE_IMAGE`, and waits for two cold-start races (Kyverno's webhook, Garage's status) that the reused bench cluster never hit.
- README: the suite's header, badges and notes.
**Bench, from scratch:** 179 s in all (cluster and data layer 149 s, chart install 27 s, the three checks seconds each), with inbuxa/inbuxa-charts#34. The first from-scratch runs found the two races and the chart bug fixed in #34.
**S7:** the GitHub runs on this branch answer it (time and memory on both architectures); I'll put the numbers here and on #7.
lab/e2e.sh: up.sh, then the chart with the images release/suite.yaml
names, seeded, then mail, the webmail across replicas and the console,
timing each part. lab/tools.sh installs kind, kubectl and Helm at pinned
versions against their published checksums. .github/workflows/e2e.yml
runs both on ubuntu-24.04 and ubuntu-24.04-arm (GitHub replica only).
up.sh now does what a fresh machine needs: adds the Helm repositories,
installs MetalLB with a pool computed from kind's network, takes
KIND_NODE_IMAGE, and waits out two cold-start races (Kyverno's webhook,
Garage's status) that a reused bench cluster never showed.
The README gets the suite's header, badges and notes.
GitHub, this branch: both green, e2e 220 s on amd64 and 217 s on arm64 (jobs 3 min 52 s / 3 min 50 s), ~4 GiB of 15 used. Details on #7, which this answers.
GitHub, this branch: both green, e2e 220 s on amd64 and 217 s on arm64 (jobs 3 min 52 s / 3 min 50 s), ~4 GiB of 15 used. Details on #7, which this answers.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
lab/e2e.sh: the whole lab from nothing (up.sh, the chart with the imagesrelease/suite.yamlnames, seeded), then mail, the webmail across replicas and the console, timing each part; the summary goes to the GitHub job summary too.lab/tools.sh: kind v0.33.0, kubectl v1.37.2, Helm v4.3.2 for amd64/arm64, each checked against its published SHA-256..github/workflows/e2e.yml: both onubuntu-24.04andubuntu-24.04-arm, on the GitHub replica only (vars.BUILD_ON). Chart from inbuxa-chartsmain, or the branch invars.CHART_REF: set tofix/hook-network-policiesuntil inbuxa/inbuxa-charts#34 merges, then removed.up.shnow does what a fresh machine needs: Helm repositories, MetalLB (pool computed from kind's network),KIND_NODE_IMAGE, and waits for two cold-start races (Kyverno's webhook, Garage's status) that the reused bench cluster never hit.Bench, from scratch: 179 s in all (cluster and data layer 149 s, chart install 27 s, the three checks seconds each), with inbuxa/inbuxa-charts#34. The first from-scratch runs found the two races and the chart bug fixed in #34.
S7: the GitHub runs on this branch answer it (time and memory on both architectures); I'll put the numbers here and on #7.
GitHub, this branch: both green, e2e 220 s on amd64 and 217 s on arm64 (jobs 3 min 52 s / 3 min 50 s), ~4 GiB of 15 used. Details on #7, which this answers.