S-04c · Run the ACL integration test on the CI runner (compose-network) #55
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Follow-up split from #46 (CLAUDE.md §13). S-04a delivered the ACL ↔ real-OpenZaak integration test; it passes via
make integrationlocally / on a host-executing runner, but it cannot run on Gitea's hostedubuntu-latestrunner.Why: the runner starts the OpenZaak stack as sibling containers via the host daemon, so published ports bind to the host — not to the runner's
localhost. A process on the runner (the seed, anddotnet test) getsConnection refusedonlocalhost:8000. Bind mounts have the same reach problem (see gitea-actions-gotchas.md §1). This is whycompose-smokeonly polls health viadocker inspectand never talks to a service port.Outcome: run the integration test (and the publishing seed) inside the compose network so they reach
http://openzaak:8000by service name, with sources delivered via a built image (not bind mounts). Then re-add a gatingintegrationjob to CI.Options to weigh in this issue:
Dockerfile.integration(multi-stage: buildAcl.IntegrationTests, rundotnet test --filter Category=Integrationat container start) + a containerised seed step, joined to the OpenZaak network (OZ_BASE=http://openzaak:8000).services:for OpenZaak + Postgres + Redis (awkward multi-step init/seed ordering).Out of scope: changing the test assertions or the gateway behaviour — those are done in #46.
Refs #46. Documented as a known gap in
docs/runbooks/gitea-actions-gotchas.md.not referenced this issue2026-06-29 10:05:11 +00:00
not referenced this issue2026-06-29 10:29:08 +00:00