test-case: multi-process flock contention #44

New Issue

2026-05-13T08:21:40Z

claude-noether commented

2026-05-13 08:21:40 +00:00

Steps

Open two terminals on the same host with the same history.dir.
In terminal A: luajit main.lua (aish boots, takes the flock).
In terminal B: luajit main.lua (second boot attempt).

Expected

Terminal B startup banner includes: [aish] memory disabled: memory.jsonl held by another aish process (Resource temporarily unavailable).
Terminal B continues to function — no memory injection, but all other Phase 0/1/2/3 features work.
:memory list in terminal B prints (no memory items) (or a similar status; the load reads via io.open which doesn't need the flock).
Hmm — :memory list actually CAN read memory.jsonl without holding the flock. So terminal B can READ but not WRITE. :remember X in terminal B should fail with memory unavailable.
Close terminal A's aish (:quit). Now :memory inject in terminal B... still won't enable writes (terminal B's memory handle was never created). Closing + re-launching terminal B would now succeed.

What this exercises

R-B1 single-writer enforcement via flock(LOCK_EX | LOCK_NB)
Graceful degradation: aish runs without memory if the lock is held
Read-only access via io.open(load_memory) is unaffected by the lock

Likely failure modes

Terminal B crashes instead of status-logging → check the if m then ... else renderer.status(...) end in repl.lua startup.
Terminal B silently functions with no memory hint → check that renderer.status("memory disabled: ...") actually fires.
After terminal A closes, terminal B's :remember still fails forever (no auto-reacquire) → that's by design v1; document if surprising.

## Steps 1. Open two terminals on the same host with the same `history.dir`. 2. In terminal A: `luajit main.lua` (aish boots, takes the flock). 3. In terminal B: `luajit main.lua` (second boot attempt). ## Expected - Terminal B startup banner includes: `[aish] memory disabled: memory.jsonl held by another aish process (Resource temporarily unavailable)`. - Terminal B continues to function — no memory injection, but all other Phase 0/1/2/3 features work. - `:memory list` in terminal B prints `(no memory items)` (or a similar status; the load reads via io.open which doesn't need the flock). - Hmm — `:memory list` actually CAN read memory.jsonl without holding the flock. So terminal B can READ but not WRITE. `:remember X` in terminal B should fail with `memory unavailable`. - Close terminal A's aish (`:quit`). Now `:memory inject` in terminal B... still won't enable writes (terminal B's memory handle was never created). Closing + re-launching terminal B would now succeed. ## What this exercises - R-B1 single-writer enforcement via flock(LOCK_EX | LOCK_NB) - Graceful degradation: aish runs without memory if the lock is held - Read-only access via io.open(load_memory) is unaffected by the lock ## Likely failure modes - Terminal B crashes instead of status-logging → check the `if m then ... else renderer.status(...) end` in repl.lua startup. - Terminal B silently functions with no memory hint → check that `renderer.status("memory disabled: ...")` actually fires. - After terminal A closes, terminal B's `:remember` still fails forever (no auto-reacquire) → that's by design v1; document if surprising.

claude-noether added the test-case label 2026-05-13 08:21:40 +00:00

claude-noether commented

2026-05-13 11:50:32 +00:00

PASS (autonomous run, 2026-05-13). Process A held a memory.jsonl flock via a fifo-driven background luajit. Process B's startup status: [aish] memory disabled: memory.jsonl held by another aish process (Resource temporarily unavailable). Process B continued functioning (:memory list returned cleanly, no crash). R-B1 single-writer enforcement works as designed. Closing.

**PASS** (autonomous run, 2026-05-13). Process A held a memory.jsonl flock via a fifo-driven background luajit. Process B's startup status: `[aish] memory disabled: memory.jsonl held by another aish process (Resource temporarily unavailable)`. Process B continued functioning (`:memory list` returned cleanly, no crash). R-B1 single-writer enforcement works as designed. Closing.

claude-noether closed this issue

2026-05-13 11:50:32 +00:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: marfrit/aish#44