docker: remove the four ADD fetches that are blocking every deploy - #131
Conversation
They are mine (3c76143) and they break the build. `AdaWorldAPI/ndarray`'s default branch is `master`, not `main`, so `.../ndarray/commits/main.atom` is a guaranteed 404 — and a failed `ADD` fails the build. Every deploy dies at step 11/16. Removed rather than corrected, because the whole block should not exist. It was added to bust the sibling-clone layer after a stale lance-graph clone took a deploy down — but what made a stale clone FATAL was `COUNT_FUSE`, a compile-time length assert that turned "this mirror is one concept behind OGAR" into E0080 at const-eval. lance-graph #954 removed that fuse: activation now checks the PLUGGED concepts against the mirror at runtime and names the drift. A slightly stale sibling compiles now; real drift surfaces where it can be read. So the fetches defended a failure class that had already been retired, and paid for it with four network calls in the build's critical path where any 404 or blip kills the deploy. The comment directly above them named that exact hazard and I shipped them anyway. Deletions only in effect — no build step changes, no clone step changes, no behaviour change beyond the layer being cacheable again. The comment that replaces them records why the block is absent, so it does not get re-added, and points at explicit SHA ARGs as the instrument to use if cache staleness needs solving: an ARG change busts the layer with zero network calls and fails legibly at `git checkout`.
|
Important Review available on request
Reviews should be triggered manually for repositories with fewer than 10 stars. Select Trigger review above or comment ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Bugbot couldn't run - usage limit reachedBugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit. A user or team admin can review and increase usage limits in the Cursor dashboard. (requestId: serverGenReqId_30ffefa1-2f74-433b-ad35-d8151c7755ac) |
Unblocks the deploy. Not a draft — every build on
mainis currently failing at step 11/16.One file, one commit. The four
ADD https://…/commits/main.atomlines and the comment block that justified them are deleted. No build step, clone step, or behaviour change otherwise — the layer is simply cacheable again, exactly as before3c76143.Why the build fails
AdaWorldAPI/ndarray's default branch ismaster, notmain, so that URL is a permanent 404.This is not a stale cache. BuildKit resolves an
ADD <url>by fetching the URL to compute the cache key, so it fails on a cold cache, a warm cache, and a fresh commit alike — disabling the build cache only guarantees the fetch is attempted. "Push a new commit" works in exactly one sense: a commit that removes the lines. This is that commit.Why removed, not repaired
Correcting
main→masterwould also have worked, and would still have left four network fetches in the build's critical path where any 404 or blip kills a deploy.More importantly, the block defends a failure class that no longer exists. It was added after a stale
lance-graphclone took a deploy down — but what made a stale clone fatal waslance-graph-ogar'sCOUNT_FUSE, a compile-time length assert that turned "this clone's codebook mirror is one concept behind OGAR" intoE0080at const-eval.lance-graph #954 removed that fuse. Activation now verifies the plugged concepts against the mirror at runtime and names the drift (
MirrorDrift { concept, authority_id, mirror_id }). A slightly stale sibling compiles; real drift surfaces where it can be read. The plug-and-play migration retired the hazard, and the cache-bust fetches outlived their reason.The comment that replaces them records why the block is absent so it does not get re-added, and points at explicit SHA
ARGs as the instrument to use if cache staleness needs solving again — anARGchange busts the layer with zero network calls and fails legibly atgit checkout.Provenance
Both the breakage and the fix are mine:
3c76143(q2 #130) added the lines, and the comment directly above them named this exact hazard — "a failedADDFAILS THE BUILD. That trades a stale cache for a new outage cause, which is the opposite of the point." — before shipping four of them anyway.🤖 Generated with Claude Code
https://claude.ai/code/session_01NMeiLmtDKhomJNSo2ecbJw
Generated by Claude Code