feat(inspur): keep DIMM state on the DIMM, parse maintenance.log installs

component.log memory status produced synthetic "Memory module … not
detected (capacity 0GB)" events that duplicated IDL/SEL "Memory Device
Disabled … POR Mapout" Asserts already in the journal, with a wrong text
(the modules are present).

- Drop synthetic memory events. The snapshot status stays authoritative
  on the DIMM; the journal Assert/Deassert supplies status_history, the
  fault description and status_changed_at (latest active Assert).
- Drop a component.log fan state event when the journal already holds an
  active Assert for that fan.
- Parse "Component Replace <dev> SN changed from X to Y" from
  maintenance.log and rotated maintenance.log.N.gz into Info events,
  mapping DIMM<cpu><ch><dimm> to the inventory slot (DIMM021 -> CPU0_C2D1).

ADL-071.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Mikhail ChusavitinandClaude Opus 5.5 committed 2026-09-28 13:01:30 +03:00
1 parent f2ad0aca96
commit 693906ceba
7 files changed
+421 -57

No files matched your search

+30
View File
@@ -2071,3 +2071,33 @@ collection may use the current time, because it is the collection time.
- `time.Now()` stays legitimate only for job bookkeeping, export metadata
(`exported_at`, `generated_at`), live-collection timing and the Redfish
live log-entry window.
---
## ADL-071 — Snapshot component state lives on the component, not in the event log
**Date:** 2026-09-28
**Context:** Inspur `component.log` RESTful memory status produced synthetic
"Memory module … not detected (capacity 0GB)" events. On 24C319575 all three
duplicated IDL/SEL `Memory Device Disabled … POR Mapout` Asserts already in
the journal, and the text was wrong (modules present, 64 GB each). The same
dump showed the three DIMMs were installed that day: the only record of it is
`Component Replace … SN changed from NULL to …` in `maintenance.log`, which was
not parsed.
**Decision:** Component state read from a snapshot is not journal data and is
not duplicated into the event log. DIMM state stays on the DIMM: the snapshot
status is authoritative, and the IDL/SEL Assert/Deassert journal supplies
`status_history`, the fault description and `status_changed_at` (latest
active Assert). A component.log fan state event is kept only when the journal
has no active Assert for that fan. Inspur `maintenance.log` (and rotated
`maintenance.log.N.gz`) is parsed for `Component Replace <dev> SN changed from
X to Y` records only, emitted as Info events of type `Component Replace` with
`DIMM<cpu><ch><dimm>` mapped to the inventory slot (`DIMM021` → `CPU0_C2D1`).
**Consequences:**
- Inspur no longer emits "Memory module … not detected" events; a mapped-out
DIMM shows its journal fault on the DIMM record instead.
- Installation is not written to `status_history`: that field carries health
statuses only, and the BMC logs the SN change after POST, i.e. after a POR
mapout of the same module.