release: advance to v26.2.2 and operator v1.0.0-rc.3#657
Conversation
| `cockroachdb.crdbCluster.postInitSQL` sets `spec.postInitSQL` on the `CrdbCluster` | ||
| resource. The operator runs these SQL sources once after cluster initialization. | ||
| TLS must be enabled, and the `CrdbCluster` reconciliation mode must be empty/default or | ||
| `MutableOnly`. |
There was a problem hiding this comment.
Nit: I think the empty/default mode is MutableOnly(IIRC, we handle this in helpers ). So we could just say MutableOnly?
|
|
||
| ## Under-Replicated Ranges Check | ||
|
|
||
| Before starting a rolling restart or scale-down, the operator checks cluster-wide replication |
There was a problem hiding this comment.
Nit: rolling upgrades/cluster upgrades might suit better here?
| a new pod eviction or node decommission is pending. | ||
|
|
||
| The operator executes SQL from a running CockroachDB pod against `crdb_internal.kv_store_status` | ||
| and sums the `ranges.underreplicated` metric across stores. If the result is non-zero, or if the |
There was a problem hiding this comment.
Nit: We can omit these details here.
|
|
||
| Before starting a rolling restart or scale-down, the operator checks cluster-wide replication | ||
| health so it does not make an already degraded cluster less available. The check runs only when | ||
| a new pod eviction or node decommission is pending. |
There was a problem hiding this comment.
I think "when node decommission is pending" might sound differently - If the node is already selected for scale down or pending decom they are not affected. This check only runs before scale down is initiated.
"The check runs before the operator starts a new node decommission during scale-down".
For rolling updates, it runs right before starting a new node disruption/eviction. It intentionally does not block an already in-flight pod update.
Nit: new pod eviction or node decommission - we can mention pod or node in both places.
|
|
||
| ## [1.0.0-rc.3] — 2026-07-20 | ||
| ### Added | ||
| - Added operator support for `spec.postInitSQL` on `CrdbCluster` to run raw SQL once after cluster |
There was a problem hiding this comment.
Nit: Since this changelog is specifically for operator, we can just say Added support?
| return "", fmt.Errorf("no pods found with label %s", operator.LabelSelector) | ||
| } | ||
|
|
||
| output, err := k8s.RunKubectlAndGetOutputE(t, kubectlOptions, |
There was a problem hiding this comment.
nit: I think there must be some helper which does similar stuff by execing into pod and executes sql command. Just check if we could reuse and eliminate this block here.
If changes turns out to have moe effort, we can continue as is.
557af69 to
8412197
Compare
Render and validate postInitSQL in the v2 CockroachDB chart, preserve it in v1beta1 values export, and document it in the CockroachDB chart README. Consume the operator v1.0.0-rc.3 image and update the operator chart changelog and docs for RC3 features, including the under-replicated ranges gate. Expose CrdbCluster feature gates through the CockroachDB chart and set the under-replicated-ranges recovery override in rollout-focused e2e upgrades so RC3's gate does not block unrelated pod-template assertions. Add upgrade compatibility coverage for postInitSQL and reuse the shared SQL exec helper in operator e2e tests.
8412197 to
efd9037
Compare
VishalJaishankar
changed the title
2 participants
This release advances CockroachDB to v26.2.2 and the CockroachDB operator to v1.0.0-rc.3.
Highlights:
spec.postInitSQLsupportskip-under-replicated-ranges-checkbatch/jobsupdate RBAC for operator-managed jobsSee CHANGELOG.md for full details