Repository navigation
Update dependencies to fix Dependabot security alerts (#885)
- Update dependencies to fix Dependabot security alerts
Bump skaffold to v2.24.0 and raise all vulnerable modules to their
patched versions (or above):
- github.com/moby/buildkit v0.31.1
- github.com/go-git/go-git/v5 v5.19.2, go-billy/v5 v5.9.0
- google.golang.org/grpc v1.82.1
- golang.org/x/crypto v0.53.0, x/net v0.56.0, x/oauth2 v0.36.0
- github.com/sigstore/{cosign/v2 v2.6.3, rekor v1.5.2, sigstore v1.10.8}
- github.com/containerd/containerd v1.7.33
- github.com/docker/docker v28.5.2, docker/cli v29.6.1
- go.opentelemetry.io/otel/sdk v1.44.0
- go-jose/v4 v4.1.4, circl v1.6.3, mapstructure/v2 v2.5.0,
selinux v1.15.1, mongo-driver v1.17.7 - postcss 8.5.26 and picomatch 2.3.2 in both doc-gen lockfiles
Mirror skaffold's k8s.io/{api,apimachinery,client-go}=v0.33.4,
go-yit and kube-openapi replace directives, which do not apply
transitively.
Code fixes required by the bumps:
- drop invalid
divevalidator tag on the Workload struct field
(validator v10.30 panics on non-slice/map dive) - fix non-constant format string vet errors surfaced by Go 1.26
- bump CircleCI Go executor to 1.26 (go.mod now requires go >= 1.26.4)
Not fixable here: docker/docker alerts requiring v29.3.1 - the v29
line moved off the github.com/docker/docker module path; needs
upstream skaffold migration.
- Surface skaffold dev loop startup errors
RunSkaffoldDev ran as a goroutine with its error return discarded, so
any failure after render (config validation, run context setup, the dev
loop itself) left tako silently watching files while nothing deployed.
Report the error through the UI with error styling instead.
- Strip tags from generated skaffold build artifacts
Skaffold rejects tagged artifact image names ("no tag should be
specified. Use taggers instead"), but compose files routinely reference
images as name:tag. Strip any tag or digest when deriving build
artifacts from compose services or analysis images - the configured
tagger owns the tag.
- Fix rendered PVC naming and service account token mounting
Two kubernetes converter fixes, both surfaced by running the skaffold
dev loop end to end:
-
PVCs generated for volumes without a name (e.g. bind mounts) were
rendered with an empty metadata.name, which kubectl apply rejects
("resource name may not be empty"). Name the PVC after the generated
PVCName - the same name the pod's volume source references. -
Workloads configured with a custom serviceAccountName never received
an API token: the rendered ServiceAccount hardcodes
automountServiceAccountToken: false and the pod spec set no override,
so the SA-level false always won. Set the pod-level automount to true
for workloads that explicitly opt into an identity, matching the
SA-disabled/pod-enabled pattern.