Pass structured cluster resource data and taints upstream via ping path (aligned with interLink#516)

[Copilot]

The interlink core needs real-time SLURM cluster resource occupancy to update the virtual node's advertised capacity. Previously the /status ping path returned a plain-text sinfo -s dump with no machine-readable structure.

Changes

• pkg/slurm/types.go — new PingResponse, ResourcesResponse, AcceleratorResponse, and TaintConfig types mirroring the schema from interlink-hq/interLink#516; minimal slurmNodeList/slurmJSONNode types for sinfo --json parsing. A TODO marks the local types for replacement with upstream commonIL.* imports once interLink#516 is merged. SlurmConfig gains two new fields: Taints []TaintConfig for declaring static node taints and ResourceScriptPath string for plugging in a custom resource-gathering script.

• pkg/slurm/Status.go — when /status is called with an empty pod list (the interlink-api ping path), returns Content-Type: application/json with a PingResponse body. Resource data is fetched via a three-tier resolution:

  1. Custom script (ResourceScriptPath) — when set, the script is executed and its JSON stdout is parsed as a PingResponse; on failure the sinfo chain is tried as a fallback.
  2. sinfo --json (SLURM ≥ 20.11) — accurate per-node alloc_cpus and alloc_memory; reports available resources (total − allocated) as Kubernetes quantity strings.
  3. sinfo --noheader --format=%c|%m|%e — reports total installed CPUs and sum of free memory as Kubernetes quantity strings (fallback when JSON unavailable).

  Static SlurmConfig.Taints are always merged into the response (overriding any taints returned by the custom script). When no taints are configured the field is omitted (backward compatible).

• pkg/slurm/resources_test.go — parsing logic extracted into package-level parseClusterResourcesFromJSON / parseClusterResourcesFromText functions; tests cover the PingResponse format with Kubernetes quantity strings, taint serialization, and the custom script path (happy path, non-zero exit, invalid JSON, missing status field).

Example response on ping

{
  "status": "ok",
  "resources": {
    "cpu": "24",
    "memory": "96000Mi"
  },
  "taints": [
    { "key": "vendor.io/maintenance", "value": "true", "effect": "NoSchedule" }
  ]
}

Example SlurmConfig.yaml configuration

# Optional: plug in a custom resource-gathering script
ResourceScriptPath: /etc/interlink/get_resources.sh

# Optional: static taints to advertise on the virtual node
Taints:
  - Key: vendor.io/maintenance
    Value: "true"
    Effect: NoSchedule
  - Key: vendor.io/drain
    Effect: NoExecute

Example custom script output

{"status":"ok","resources":{"cpu":"128","memory":"512Gi","pods":"1000"}}

The VK's nodeUpdate() (interLink#516) parses this body as PingResponse and:

• when resources is non-nil, calls updateNodeResources() to update both node.Status.Capacity and node.Status.Allocatable
• when taints is non-nil, calls updateNodeTaints() to replace the node's non-system taints with the provided list

Both updates happen on every 30-second heartbeat, keeping the virtual node's advertised capacity and taint set in sync with the SLURM cluster state. Omitted fields retain their currently configured values.

[dciangot]

align this changes to work with interlink-hq/interLink#516

[dciangot]

align this changes to work with interlink-hq/interLink#516

@copilot

[Copilot]

align this changes to work with interlink-hq/interLink#516

@copilot

Done in d69348e. The ping-path response now returns a PingResponse that mirrors the schema from interLink#516 exactly — {"status":"ok","resources":{"cpu":"24","memory":"96000Mi"}} — using Kubernetes quantity strings so the VK's updateNodeResources() can call resource.ParseQuantity() directly. Local type definitions (PingResponse, ResourcesResponse, AcceleratorResponse) carry a TODO to be replaced with the upstream commonIL.* imports once interLink#516 is merged and the dependency is bumped.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• https://api.github.com/repos/interlink-hq/interLink/pulls/516
  • Triggering command: /usr/bin/curl curl -s -H Accept: application/vnd.github.v3+json REDACTED (http block)
• https://api.github.com/repos/interlink-hq/interLink/pulls/516/files
  • Triggering command: /usr/bin/curl curl -s REDACTED (http block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)

[dciangot]

@copilot include also taints to be passed from the config file via the api interlink-hq/interLink#516 (comment)

[Copilot]

@copilot include also taints to be passed from the config file via the api interlink-hq/interLink#516 (comment)

Done in db90940. Operators can now configure taints in SlurmConfig.yaml:

Taints:
  - Key: vendor.io/maintenance
    Value: "true"
    Effect: NoSchedule
  - Key: vendor.io/drain
    Effect: NoExecute

The ping response will include them as:

{
  "status": "ok",
  "resources": { "cpu": "24", "memory": "96000Mi" },
  "taints": [
    { "key": "vendor.io/maintenance", "value": "true", "effect": "NoSchedule" },
    { "key": "vendor.io/drain", "effect": "NoExecute" }
  ]
}

The VK's updateNodeTaints() (interLink#516) will replace the node's non-system taints with this list on every heartbeat. When no taints are configured the field is omitted entirely (backward compatible).

[dciangot]

@copilot make get cluster resource script configurable. Imagine I want to pass my own logic from the config file of the plugin.

[Copilot]

Done in 819db1f. Add ResourceScriptPath to SlurmConfig.yaml to plug in your own resource-gathering logic:

ResourceScriptPath: /etc/interlink/get_resources.sh

The script must write a PingResponse-shaped JSON to stdout:

{"status":"ok","resources":{"cpu":"128","memory":"512Gi","pods":"1000"}}

When configured, the script runs instead of sinfo. A non-zero exit code, empty output, or unparseable JSON logs a warning and falls back to the built-in sinfo chain. Static Taints from the config always override any taints returned by the script.