# 05 — Collectors

Collectors live in `internal/collector/`.

Core files:
- `registry.go` for protocol registration
- `redfish.go` for live collection
- `redfish_replay.go` for replay from raw payloads
- `ipmi_mock.go` for the placeholder IPMI implementation
- `types.go` for request/progress contracts

## Redfish collector

Status: active production path.

Request fields passed from the server:
- `host`
- `port`
- `username`
- `auth_type`
- credential field (`password` or token)
- `tls_mode`
- optional `power_on_if_host_off`

### Core rule

Live collection and replay must stay behaviorally aligned.
If the collector adds a fallback, probe, or normalization rule, replay must mirror it.

### Preflight and host power

- `Probe()` may be used before collection to verify API connectivity and current host `PowerState`
- if the host is off and the user chose power-on, the collector may issue `ComputerSystem.Reset`
  with `ResetType=On`
- power-on attempts are bounded and logged
- after a successful power-on, the collector waits an extra stabilization window, then checks
  `PowerState` again and only starts collection if the host is still on
- if the collector powered on the host itself for collection, it must attempt to power it back off
  after collection completes
- if the host was already on before collection, the collector must not power it off afterward
- if power-on fails, collection still continues against the powered-off host
- all power-control decisions and attempts must be visible in the collection log so they are
  preserved in raw-export bundles

### Discovery model

The collector does not rely on one fixed vendor tree.
It discovers and follows Redfish resources dynamically from root collections such as:
- `Systems`
- `Chassis`
- `Managers`

### Stored raw data

Important raw payloads:
- `raw_payloads.redfish_tree`
- `raw_payloads.redfish_fetch_errors`
- `raw_payloads.source_timezone` when available

### Snapshot crawler rules

- bounded by `LOGPILE_REDFISH_SNAPSHOT_MAX_DOCS`
- prioritized toward high-value inventory paths
- tolerant of expected vendor-specific failures
- normalizes `@odata.id` values before queueing

### Redfish implementation guidance

When changing collection logic:

1. Prefer alternate-path support over vendor hardcoding
2. Keep expensive probing bounded
3. Deduplicate by serial, then BDF, then location/model fallbacks
4. Preserve replay determinism from saved raw payloads
5. Add tests for both the motivating topology and a negative case

### Known vendor fallbacks

- empty standard drive collections may trigger bounded `Disk.Bay` probing
- `Storage.Links.Enclosures[*]` may be followed to recover physical drives
- `PowerSubsystem/PowerSupplies` is preferred over legacy `Power` when available

## IPMI collector

Status: mock scaffold only.

It remains registered for protocol completeness, but it is not a real collection path.