Skip to content

deploy-test: retry stack creation once when FirstBootDone times out #112

Description

@defangdevs

Since PR 110, the first-boot gate turns the IMDS brick (issue 106, upstream NixOS/nixpkgs#542874) into CREATE_FAILED + rollback. First in-CI catch: master run 29571322617 — the ipv6-outputs leg failed honestly (previously it reported green over a dead box) while the ipv4-full leg passed the full smoke test in the same region at the same time, i.e. the race is intermittent, not a regional outage.

Until the upstream fetch-ec2-metadata fix lands, the race will keep flaking deploy-test runs. Proposal: in the Create stack step, on a CREATE_FAILED whose stack events show FirstBootDone ... WaitCondition timed out, delete the rolled-back stack and retry the create exactly once (a second consecutive brick should stay a failure — it likely isn't the race). Keep the retry OFF for dispatched runs when destroy=false (debugging wants the artifact, though rollback already tears it down).

Not a fix for end users clicking Launch — they see a clean CREATE_FAILED and relaunch — but it keeps master CI signal usable while upstream is broken.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions