K8s CRD backend: tcp_port_list not persisted in LayerDrbdResources

[kvaps]

Bug Description

When using the K8s CRD database backend, the tcp_port_list field in LayerDrbdResources CRD entries is never persisted. This causes TCP port re-allocation on every controller restart, leading to port mismatches between controller and satellites after toggle-disk operations.

Root Cause

Migration 28 (Migration_28_v1_31_1_MoveTcpPortsToNodes) adds the tcp_port_list field to GenCrdV1_31_1.LayerDrbdResourcesSpec and upserts all entries with port values. However, examining a production cluster shows:

• CRD schema has tcp_port_list: string (correctly)
• CRD storage version is v1-31-1 (correctly)
• All 1100+ CRD entries have tcp_port_list missing — both entries created before AND after the migration

The field is defined in GenCrdV1_31_1.LayerDrbdResourcesSpec (line 2190) and dataToCrd at line 624 correctly calls the tcpPortSetter. The GenCrdV1_15_0.LayerDrbdResourcesSpec does NOT have this field.

Possible causes:

1. The migration upserts through v1-31-1 API, but K8s stored entries in v1-15-0 format (race between schema update and data write)
2. A subsequent controller operation re-writes CRD entries through a code path that drops the field
3. K8s trivial conversion between CRD versions doesn't preserve unknown fields

Impact

Every controller restart re-allocates TCP ports via initPorts() using preferred ports from peers. Since the pool state is empty at startup, ports may be allocated differently than the previous run. This causes:

• Controller allocates port X for a resource on a node
• Satellite's DRBD kernel is still bound to old port Y from the previous allocation
• Satellite's .res file gets regenerated with port X, but drbdadm adjust fails because port X conflicts with another resource that got reassigned
• Resources stuck in StandAlone/Connecting state after controller restarts

Reproduction

1. Use K8s CRD backend
2. Create DRBD resources
3. Check CRD entries: kubectl get layerdrbdresources.internal.linstor.linbit.com -o json | python3 -c "import json,sys; data=json.load(sys.stdin); print(sum(1 for i in data['items'] if 'tcp_port_list' not in i.get('spec',{})))"
4. All entries will show missing tcp_port_list
5. Restart the controller
6. Ports are re-allocated, potentially different from what DRBD is using

Environment

• LINSTOR Server: v1.32.3
• K8s CRD API version: v1-15-0 stored, v1-31-1 served
• Database version: 28 (migration completed)

Suggested Fix

After loading DrbdRscData from DB, if tcp_port_list was null/empty, persist the allocated ports back to the CRD via an upsert. This ensures ports are persisted on the first startup after upgrade.

[kvaps]

Closing this issue — the premise was incorrect.

The tcp_port_list field IS properly persisted in the K8s CRD entries. The confusion arose because kubectl queries through the v1-15-0 API version (which lacks the field), while the LINSTOR Java client reads through v1-31-1 (which includes it). The K8s CRD conversion strategy is None, so kubectl shows the raw v1-15-0 format without the field, but the actual stored data in the v1-31-1 storage version contains tcp_port_list correctly.

The actual root cause of the TCP port mismatches after toggle-disk operations is addressed in #476.