ops: Signal disk-tier attach pending vs ready #33

New issue

Closed

opened 2026-09-04 18:24:47 +00:00 by eva · 0 comments

eva commented

2026-09-04 18:24:47 +00:00

Owner

Problem

Large disk caches can look "memory-only / broken" for seconds–minutes at startup while the disk tier attaches. That is intentional but opaque in ops — Justin shouldn't need to read source to know the barrier is still warming.

Why

False "broken" reads burn operator time and hide a healthy warm-up path.

Scope

One clear signal when disk attach is pending vs ready: log line + /metrics flag and/or heartbeat detail
No new daemon / second metrics stack

Acceptance

Operator can tell from logs and existing HTTP endpoints whether disk tier is attaching or ready
Behavior documented in README next to Quick check / startup notes
Tests cover the signal where practical

Out of scope

#31 HTTP HEAD
Redesigning tiered storage

From Forge Meeting (Linus ops list; Justin asked these filed).

## Problem Large disk caches can look "memory-only / broken" for seconds–minutes at startup while the disk tier attaches. That is intentional but opaque in ops — Justin shouldn't need to read source to know the barrier is still warming. ## Why False "broken" reads burn operator time and hide a healthy warm-up path. ## Scope - One clear signal when disk attach is pending vs ready: log line + `/metrics` flag and/or heartbeat detail - No new daemon / second metrics stack ## Acceptance - Operator can tell from logs and existing HTTP endpoints whether disk tier is attaching or ready - Behavior documented in README next to Quick check / startup notes - Tests cover the signal where practical ## Out of scope - #31 HTTP HEAD - Redesigning tiered storage From Forge Meeting (Linus ops list; Justin asked these filed).