[jira] [Closed] (FLINK-39356) FlinkStateSnapshot cleanup fails with NPE when status is null, permanently blocking CR and namespace deletion

["Gyula Fora (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/FLINK-39356?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Gyula Fora closed FLINK-39356.
------------------------------
    Fix Version/s: kubernetes-operator-1.16.0
         Assignee: Nihar Rao
       Resolution: Fixed

> FlinkStateSnapshot cleanup fails with NPE when status is null, permanently blocking CR and namespace deletion
> -------------------------------------------------------------------------------------------------------------
>
>                 Key: FLINK-39356
>                 URL: https://issues.apache.org/jira/browse/FLINK-39356
>             Project: Flink
>          Issue Type: Bug
>          Components: Kubernetes Operator
>    Affects Versions: 1.14.0
>            Reporter: Yanis Djeridi
>            Assignee: Nihar Rao
>            Priority: Major
>              Labels: pull-request-available
>             Fix For: kubernetes-operator-1.16.0
>
>
> h3. Problem
> When a FlinkStateSnapshot CR is deleted before the FlinkStateSnapshotController has reconciled it, the cleanup() method throws a NullPointerException on getStatus().getState(). The exception is caught in a way that prevents the finalizer from being removed, causing the CR to be permanently stuck in a terminating state. This blocks namespace deletion. 
> h3.  Root Cause
> FlinkStateSnapshotController.cleanup() and FlinkResourceContextFactory.getFlinkStateSnapshotContext() assume the status is non-null, but FlinkStateSnapshot CRs are created without a status (the status is only populated when reconcile() runs). If the CR receives a deletion timestamp before reconcile() runs, JOSDK calls cleanup() directly and the only code path that initializes null status is in reconcile(), not cleanup().
> This causes an infinite retry loop where every attempt crashes the same way.
> reconcile() already handles this case correctly:
> {code:java}
> flinkStateSnapshot.setStatus(
>         Objects.requireNonNullElseGet(
>                 flinkStateSnapshot.getStatus(), FlinkStateSnapshotStatus::new));
> {code}
> cleanup() and updateErrorStatus() are missing this guard.
> h3. Impact
>  - FlinkStateSnapshot CRs with null status accumulate as zombie resources with finalizers that can never be cleared.
>  - Namespace deletion is blocked indefinitely
> h3. Fix 
> Add the same null-status initialization to cleanup() and updateErrorStatus() that already exists in reconcile(). A null-status snapshot was never triggered against Flink, so no data exists on storage and cleanup can safely proceed



--
This message was sent by Atlassian Jira
(v8.20.10#820010)