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

[Copilot]

The interlink core needs real-time HTCondor cluster resource occupancy to update the virtual node's advertised capacity and taints. Previously the /status ping path returned a plain-text health check response with no machine-readable resource data.

Changes

• handles.py — new parse_cluster_resources_from_json(), parse_cluster_resources_from_text(), get_cluster_resources(), _run_cluster_resources_script(), and get_taints_from_config() functions; when /status is called with an empty pod list (the interlink-api ping call), returns Content-Type: application/json with a PingResponse body. Resource data is fetched via:

  • Custom script (optional): ClusterResourcesScript in SidecarConfig.yaml — when set, the specified shell command is executed and its JSON stdout is used as the ping response, allowing operators to supply their own resource-reporting logic without modifying the plugin code
  • Primary (built-in): condor_status --json — accurate per-slot Cpus and Memory; reports available resources (Unclaimed slots, excluding dynamic child-slots to avoid double-counting with partitionable-slot parents)
  • Fallback (built-in): condor_status -autoformat Cpus Memory — reports total installed CPUs and memory as Kubernetes quantity strings (allocation state not derivable from plain-text output)
  • Taints: read from SidecarConfig.yaml; when present (even as an empty list), the VK replaces the node's non-system taints with this list; when absent, existing taints are left unchanged

• SidecarConfig.yaml — documents the optional ClusterResourcesScript and Taints config keys with usage examples

• tests/test_resources.py — parsing logic exercised via parse_cluster_resources_from_json / parse_cluster_resources_from_text functions without a live HTCondor binary; integration tests verify the ping path returns the correct PingResponse JSON format including taints and custom script output

Example response on ping

{
  "status": "ok",
  "resources": {
    "cpu": "24",
    "memory": "96000Mi"
  },
  "taints": [
    {"key": "example.com/no-schedule", "effect": "NoSchedule"}
  ]
}

Example SidecarConfig.yaml

# Plug in a custom resource-reporting script instead of the built-in condor_status logic:
ClusterResourcesScript: "/usr/local/bin/my-resource-reporter.sh"

# Taints to advertise on the virtual node:
Taints:
  - key: "example.com/no-schedule"
    effect: "NoSchedule"

This mirrors the implementation in interlink-hq/interlink-slurm-plugin#135 adapted for HTCondor.

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 on every 30-second heartbeat
• when taints is non-nil, calls updateNodeTaints() to replace the node's non-system taints

TODO: Replace the locally-defined PingResponse dict structure with upstream commonIL.* imports once interlink-hq/interLink#516 is merged and the interLink dependency is updated.

[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 e2f2c82. Added get_taints_from_config() which reads the optional Taints list from SidecarConfig.yaml and merges it into the ping response. When present (even as an empty list), the VK replaces non-system taints; when absent, existing taints are left unchanged — matching the interLink#516 behaviour.

Example config:

Taints:
  - key: "example.com/no-schedule"
    effect: "NoSchedule"
  - key: "example.com/gpu"
    value: "true"
    effect: "NoSchedule"

Example ping response:

{
  "status": "ok",
  "resources": {"cpu": "24", "memory": "96000Mi"},
  "taints": [{"key": "example.com/no-schedule", "effect": "NoSchedule"}]
}

[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 582c899. Added a ClusterResourcesScript config key to SidecarConfig.yaml. When set, the specified shell command is executed instead of the built-in condor_status logic and its JSON stdout is used as the ping response. This lets operators supply any custom resource-reporting script without modifying the plugin code.

Config example:

ClusterResourcesScript: "/usr/local/bin/my-resource-reporter.sh"

The script must print a JSON object to stdout in the PingResponse format:

{"status": "ok", "resources": {"cpu": "48", "memory": "192000Mi"}}

When absent or empty, the built-in condor_status --json / condor_status -autoformat logic is used as before.