viktor/infra
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

homelab: v0.3.1 — fix k8s db PG target (resolve CNPG primary pod, not the Service)
Some checks are pending
Build infra CLI / build (push) Waiting to run
Details
ci/woodpecker/push/default Pipeline was successful
Details

Browse source 
`k8s db <app>` (Postgres path) execed `pg-cluster-rw`, which is the CNPG
read-write SERVICE, not a pod — so kubectl exec failed with
`pods "pg-cluster-rw" not found`. The unit test only checked the plan; the verb
was never fired at live state (the gap flagged in v0.2), so it shipped broken.

Fix: the PG plan now carries a label selector (cnpg.io/instanceRole=primary)
instead of a pod name, and k8s db resolves the actual primary POD via
`kubectl get pod -l <selector>` before exec. MySQL path (real pod
mysql-standalone-0) unchanged. Live-verified both paths (psql + mysql).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Viktor Barzin 2026-06-19 09:09:34 +00:00
parent 90c944a265
commit 787ce4edfa
4 changed files with 25 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

4
cli/k8s_test.go
View file

@ -34,7 +34,9 @@ func TestK8sTargetObjectRef(t *testing.T) {
func TestPlanDBExecPostgresDefault(t *testing.T) {
p := planDBExec("fire-planner", "", "SELECT 1", false)
if p.ns != "dbaas" || p.pod != "pg-cluster-rw" || p.container != "postgres" {
// pg-cluster-rw is a Service, so the PG plan resolves the primary POD by
// label rather than naming an (un-exec-able) Service.
if p.ns != "dbaas" || p.pod != "" || p.selector != "cnpg.io/instanceRole=primary" || p.container != "postgres" {
t.Fatalf("unexpected pg target: %+v", p)
}
// db name defaults to the app; SQL passed via -tAc