The disk probe (internal/infrastructure/readiness/probes.go, NewDiskProbe/DefaultDiskThreshold 0.90) is constructed with a single path. On carina that is work_dir, which lives on /home.
Measured 2026-09-17: /home was 95% used and correctly reported degraded, while / was 59% and /var 47%. The inverse case is the dangerous one and it is invisible: a runaway cache under / would fill a 30G root with 12G free and nothing in /health/detailed would say so until writes started failing.
This is not hypothetical for the Go toolchain as configured: GOCACHE and GOMODCACHE now point at /mnt/app_caches (a ZFS dataset). Go does NOT error when a cache directory is missing -- it creates it. So a failed zpool import at boot silently redirects every build cache write onto /, which the probe would not report.
Two parts:
Probe / (and ideally /var) as well as work_dir, reporting per-filesystem results rather than one path.
Consider a mount guard for /mnt/app_caches in the build path, since Go's fail-open behaviour turns a missing mount into writes on the wrong disk.
Confirmed the fail-open behaviour directly: GOCACHE=/tmp/does-not-exist-xyz/build on 'go build' exited 0 and created the directory.
Imported from beads archie-core-yijm · GitHub #835 · type bug · priority P2 · created 2026-09-17 · closed 2026-09-17
The disk probe (internal/infrastructure/readiness/probes.go, NewDiskProbe/DefaultDiskThreshold 0.90) is constructed with a single path. On carina that is work_dir, which lives on /home.
Measured 2026-09-17: /home was 95% used and correctly reported degraded, while / was 59% and /var 47%. The inverse case is the dangerous one and it is invisible: a runaway cache under / would fill a 30G root with 12G free and nothing in /health/detailed would say so until writes started failing.
This is not hypothetical for the Go toolchain as configured: GOCACHE and GOMODCACHE now point at /mnt/app_caches (a ZFS dataset). Go does NOT error when a cache directory is missing -- it creates it. So a failed zpool import at boot silently redirects every build cache write onto /, which the probe would not report.
Two parts:
1. Probe / (and ideally /var) as well as work_dir, reporting per-filesystem results rather than one path.
2. Consider a mount guard for /mnt/app_caches in the build path, since Go's fail-open behaviour turns a missing mount into writes on the wrong disk.
Confirmed the fail-open behaviour directly: GOCACHE=/tmp/does-not-exist-xyz/build on 'go build' exited 0 and created the directory.
### Close reason
Landed in 8378c92; GitHub #835 already closed.
---
_Imported from beads `archie-core-yijm` · GitHub #835 · type bug · priority P2 · created 2026-09-17 · closed 2026-09-17_
sam
added the bugmedium labels 2026-10-27 13:41:21 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
The disk probe (internal/infrastructure/readiness/probes.go, NewDiskProbe/DefaultDiskThreshold 0.90) is constructed with a single path. On carina that is work_dir, which lives on /home.
Measured 2026-09-17: /home was 95% used and correctly reported degraded, while / was 59% and /var 47%. The inverse case is the dangerous one and it is invisible: a runaway cache under / would fill a 30G root with 12G free and nothing in /health/detailed would say so until writes started failing.
This is not hypothetical for the Go toolchain as configured: GOCACHE and GOMODCACHE now point at /mnt/app_caches (a ZFS dataset). Go does NOT error when a cache directory is missing -- it creates it. So a failed zpool import at boot silently redirects every build cache write onto /, which the probe would not report.
Two parts:
Confirmed the fail-open behaviour directly: GOCACHE=/tmp/does-not-exist-xyz/build on 'go build' exited 0 and created the directory.
Close reason
Landed in 8378c92; GitHub #835 already closed.
Imported from beads
archie-core-yijm· GitHub #835 · type bug · priority P2 · created 2026-09-17 · closed 2026-09-17