Implement pageserver drain via storage-controller API

[bittermandel]

Summary

Currently, the pageserver drain functionality in the operator uses a placeholder implementation with annotations. This needs to be replaced with actual calls to the storage-controller API to properly drain pageservers before deletion.

Current Implementation

The current implementation in crates/neon_cluster/src/controllers/pageserver/pod.rs has two TODO functions:

• is_pageserver_drained() - Currently checks for a neon.io/drained annotation
• trigger_pageserver_drain() - Currently only logs the drain request

Required Implementation

Based on the storage controller API (found in storcon_cli), the drain operation should:

1. Trigger drain: Make a PUT request to control/v1/node/{node_id}/drain
2. Check drain status: Poll the node status to verify drain completion

API Details

// Trigger drain
storcon_client.dispatch::<(), ()>(
    Method::PUT,
    format\!("control/v1/node/{node_id}/drain"),
    None,
).await?;

Implementation Tasks

• Add drain methods to StorageControllerClient in crates/neon_cluster/src/storage_controller/client.rs
• Update trigger_pageserver_drain() to call storage controller API
• Update is_pageserver_drained() to check actual drain status from storage controller
• Handle error cases and retries appropriately
• Add tests for the new functionality

Files to Modify

• crates/neon_cluster/src/controllers/pageserver/pod.rs - Update drain functions
• crates/neon_cluster/src/storage_controller/client.rs - Add drain API methods

Additional Context

The pageserver ID is available in the pod labels as neon.io/pageserver-id.

[linear]

NEON-19 Implement pageserver drain via storage-controller API

Summary

Currently, the pageserver drain functionality in the operator uses a placeholder implementation with annotations. This needs to be replaced with actual calls to the storage-controller API to properly drain pageservers before deletion.

Current Implementation

The current implementation in crates/neon_cluster/src/controllers/pageserver/pod.rs has two TODO functions:

• is_pageserver_drained() - Currently checks for a neon.io/drained annotation
• trigger_pageserver_drain() - Currently only logs the drain request

Required Implementation

Based on the storage controller API (found in storcon_cli), the drain operation should:

1. Trigger drain: Make a PUT request to control/v1/node/{node_id}/drain
2. Check drain status: Poll the node status to verify drain completion

API Details

// Trigger drain
storcon_client.dispatch::<(), ()>(
    Method::PUT,
    format\!("control/v1/node/{node_id}/drain"),
    None,
).await?;

Implementation Tasks

• Add drain methods to StorageControllerClient in crates/neon_cluster/src/storage_controller/client.rs
• Update trigger_pageserver_drain() to call storage controller API
• Update is_pageserver_drained() to check actual drain status from storage controller
• Handle error cases and retries appropriately
• Add tests for the new functionality

Files to Modify

• crates/neon_cluster/src/controllers/pageserver/pod.rs - Update drain functions
• crates/neon_cluster/src/storage_controller/client.rs - Add drain API methods

Additional Context

The pageserver ID is available in the pod labels as neon.io/pageserver-id.