Issues

Runtime promotion stalls while Tom plays: the offload room check reads gex45's load average, so farm CI alone makes it look full

tompassarelli/north#407 · opened

on Oct 31, 2026

3

Runtime promotion of the admission fix waits on the host it repairs

tompassarelli/north#406 · opened

on Oct 31, 2026

Fleet launch still places sessions on a host whose admission queue is over its limit

tompassarelli/north#405 · opened

on Oct 31, 2026

1

An agent's edit under node_modules leaks into every checkout through Bun's hard links

tompassarelli/north#404 · opened

on Oct 31, 2026

The overload watchdog pages when every busy tree is inside its lease or quota and it has nothing to cap

tompassarelli/north#403 · opened

on Oct 31, 2026

Split the north crate so a leaf edit rebuilds in under 7 s

tompassarelli/north#402 · opened

on Oct 31, 2026

2

Admission settles 30 s after each start, so builds stop piling in before pressure shows

tompassarelli/north#400 · opened

on Oct 31, 2026

1

The main-checkout write guard misses in-place perl edits

tompassarelli/north#399 · opened

on Oct 31, 2026

safe-push reports NOT landed (lane commits missing) after it pushed the commit to main

tompassarelli/north#397 · opened

on Oct 31, 2026

A headless Chrome run on the laptop stalls on a keyring unlock prompt

tompassarelli/north#396 · opened

on Oct 31, 2026

main is red

tompassarelli/north#395 · opened

on Oct 31, 2026

A session's pkill -f pattern can kill other sessions' processes (it cancelled other lanes' safe-push waiters)

tompassarelli/north#393 · opened

on Oct 31, 2026

1

wt-rescue fails on a binary change in main, which leaves main dirty

tompassarelli/north#392 · opened

on Oct 31, 2026

Build offload to gex45: say why it declined, and measure six builds at once when gex45 has room

tompassarelli/north#391 · opened

on Oct 31, 2026

6

One shared mutant-test lock serialises every north landing, so landings wait 30+ minutes behind unrelated lanes

tompassarelli/north#390 · opened

on Oct 31, 2026

1

Builds through other nix forms and interactive dev shells still skip the capacity lease

tompassarelli/north#387 · opened

on Oct 30, 2026

Heavy work waits up to 29 minutes on a 5-minute load average after the CPU has gone idle

tompassarelli/north#380 · opened

on Oct 30, 2026

3

Issue filing guard lets a plain Done when line through, so dispatch reports the issue as having no box

tompassarelli/north#373 · opened

on Oct 30, 2026

The farm-only guard refuses a filtered Lua test run because it narrows by flags only, not by environment

tompassarelli/north#362 · opened

on Oct 29, 2026

North landing flow: proven stall alerts, no cancelled gex45 land checks, first commit to main under 10 minutes

tompassarelli/north#360 · opened

on Oct 29, 2026

3

Keep the GPU host's 5-minute load under its core count while builds queue: admission must count load that leases don't bound

tompassarelli/north#359 · opened

on Oct 29, 2026

6

The GPU host's disk cannot fill by construction: shared client base, evidence budget, reclaim before DEFER_DISK_FREE

tompassarelli/north#357 · opened

on Oct 29, 2026

6

Agent configuration check failing on gex45

tompassarelli/north#356 · opened

on Oct 29, 2026

shared_target lock test fails under the full parallel suite

tompassarelli/north#354 · opened

on Oct 29, 2026

Budget gex45's shared disk across the cargo target, wisp's map store and retained smashcraft data

tompassarelli/north#353 · opened

on Oct 29, 2026

4

Navigation Menu

Repository navigation

Issues

Runtime promotion stalls while Tom plays: the offload room check reads gex45's load average, so farm CI alone makes it look full

Runtime promotion of the admission fix waits on the host it repairs

Fleet launch still places sessions on a host whose admission queue is over its limit

An agent's edit under node_modules leaks into every checkout through Bun's hard links

The overload watchdog pages when every busy tree is inside its lease or quota and it has nothing to cap

Split the north crate so a leaf edit rebuilds in under 7 s

Admission settles 30 s after each start, so builds stop piling in before pressure shows

The main-checkout write guard misses in-place perl edits

safe-push reports NOT landed (lane commits missing) after it pushed the commit to main

A headless Chrome run on the laptop stalls on a keyring unlock prompt

main is red

A session's pkill -f pattern can kill other sessions' processes (it cancelled other lanes' safe-push waiters)

wt-rescue fails on a binary change in main, which leaves main dirty

Build offload to gex45: say why it declined, and measure six builds at once when gex45 has room

One shared mutant-test lock serialises every north landing, so landings wait 30+ minutes behind unrelated lanes

Builds through other nix forms and interactive dev shells still skip the capacity lease

Heavy work waits up to 29 minutes on a 5-minute load average after the CPU has gone idle

Issue filing guard lets a plain Done when line through, so dispatch reports the issue as having no box

The farm-only guard refuses a filtered Lua test run because it narrows by flags only, not by environment

North landing flow: proven stall alerts, no cancelled gex45 land checks, first commit to main under 10 minutes

Keep the GPU host's 5-minute load under its core count while builds queue: admission must count load that leases don't bound

The GPU host's disk cannot fill by construction: shared client base, evidence budget, reclaim before DEFER_DISK_FREE

Agent configuration check failing on gex45

shared_target lock test fails under the full parallel suite

Budget gex45's shared disk across the cargo target, wisp's map store and retained smashcraft data