From 593ac4ae9d7fbc9e28ac4de1c7460f17cd131005 Mon Sep 17 00:00:00 2001 From: "j.w.jonkers" Date: Mon, 8 Jun 2026 09:25:20 +0200 Subject: [PATCH] =?UTF-8?q?ci:=20increase=20flux-local=20helm=20retry=20bu?= =?UTF-8?q?dget=20to=205=C3=9730s?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit flux-local passes --repository-cache /tmp/ to every helm template invocation, bypassing HELM_CACHE_HOME and actions/cache entirely — chart tarballs are always downloaded fresh. GitHub release CDN 504s (the observed failure mode: external-dns-1.21.1.tgz) typically clear within 60-90 seconds, but the previous retry 3×10s gave only 20 s of retry window, which was not enough. 5 attempts × 30 s delay gives ~2 minutes of retry budget, covering the observed failure mode without burning excessive CI minutes on a real template/schema error. --- platform/scripts/validate/render-flux.sh | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) diff --git a/platform/scripts/validate/render-flux.sh b/platform/scripts/validate/render-flux.sh index 46004ff6..9a1e0db1 100755 --- a/platform/scripts/validate/render-flux.sh +++ b/platform/scripts/validate/render-flux.sh @@ -67,7 +67,14 @@ echo "==> flux-local expand HelmReleases against remote charts" # controller does post-merge. Remote chart downloads occasionally return # transient GitHub/registry 5xx responses; retry only this network-heavy # step so real template/schema errors still fail deterministically. -retry 3 10 flux-local build all \ +# +# flux-local passes --repository-cache /tmp/ to every helm +# invocation, so GitHub Actions cache and HELM_CACHE_HOME are both +# bypassed — chart tarballs are always fetched fresh. The only knob +# available is the retry budget. GitHub release CDN 504s typically +# clear within 60-90 s; 5 attempts × 30 s = ~2 min retry window +# covers the observed failure mode without burning excessive CI time. +retry 5 30 flux-local build all \ --enable-helm \ "${flux_root}" \ >> "${render_output}"