Per the F9 storage episode — live validation of a user-facing feature
must exercise the real flow end-to-end, not shortcut via direct
API/agent/CLI calls. Low-level mechanism tests are exempt.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A1: AutoDiscoverStoragePaths no longer bails on a non-empty registry;
registers only deployed-app paths missing from the registry. Never
mutates/removes existing entries, never re-adds or reactivates a path
present in ANY state (incl. Decommissioned), never flips IsDefault.
A2: InferStorageLabel maps base==felhom-data namespace dir to
'Belső SSD (rendszer)' to disambiguate the internal system volume.
Table-driven tests incl. a companion that fails without the
skip-by-presence guard.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The agent (v0.31.0) now returns these on /disks, but the controller dropped them when
re-marshalling into its agentapi.DiskInfo mirror. Added both fields (additive) so they reach
/api/disks + the dashboard: wipe_durable_id (gate scheme, for the wipe-confirm) and guest_attached
(drive bound into THIS guest vs merely host-present). Controller behaviour otherwise unchanged.
ListDumpFiles ran ValidateDump (line-by-line scan) for every dump on every ~5-min
RefreshCache cycle — wasted I/O+CPU on large customer dumps. ListDumpFiles now takes
an optional cached(name,size,mod) lookup; on a (size+modtime) match it reuses the
prior result and skips ValidateDump. settings.DBValidationCache gains Size+ModTime;
listAllDumpFiles builds the lookup from the persisted cache and writes back only fresh
validations (cache miss), so an unchanged dump triggers neither a re-validation nor a
settings.json write each cycle. nil cached = legacy validate-always (back-compat).
Tests: cache-hit skips validate (sentinel), cache-miss validates, nil validates.
deriveStackName pure-suffix-stripped on '-' (postgres/db/mariadb/.../cache), so a
stack whose slug ENDS in a role token (e.g. 'my-cache') was misattributed (stripped
to 'my') — filing its DB dump under the wrong/nonexistent stack. Now threads the set
of deployed stack names (m.knownStackNames() <- ListDeployedStacks) into
DiscoverDatabases and cross-references: candidate suffix-strip if known, else the
container name if it IS a known stack, else longest known stack that is a prefix
(handles <stack>_postgres / <stack>-1), else legacy strip. nil/empty known = legacy
behaviour (appexport passes nil). Table test incl. the my-cache case (fails pre-fix).
All shippable work → main directly; report-only artifacts → felhom.eu/documentation/
(audits/backlog); risky/supervised fixes implemented on main during the supervised
session, not prepared on a branch; unattended escape hatch = revert+report, never park
on a branch. Supersedes the old 'prepared on fix/... branch, pending review' pattern.
The live-drive findings + fixspec now live in felhom.eu/documentation/audits/
(with the other audit records); they no longer belong loose at the controller
repo root. Transient SESSION-*.md working logs dropped (their lasting content is
in CHANGELOG/CONTEXT and the audit records). Trunk-based no-branches cleanup.
The cgroup-only approach was a no-op on the demo: the controller container's OWN
cgroup is unlimited (the 2GB cap is on the LXC ancestor, hidden), and /proc has no
lxcfs, so it kept reporting the host's 16GB. The Docker daemon runs IN the LXC, so
'docker info' MemTotal reports the guest's real cap (2048MB) — now the authoritative
source (cgroup limit preferred when present, e.g. non-nested). The deploy memory
guard now uses the controller's committed-app memory (sum of running mem requests)
for 'used' — accurate and cheap — instead of host /proc RSS (unobservable per guest;
would make the guard never/always fire). /api/system/info reports the guest cap as
total and committed memory as used. Tests: cgroup-limit path, docker-info fallback
(nested case), GuestMemTotalMB fallback (dockerMemTotalFn stub).
The restore paths (RestoreFromRecoveryUnit + the RestoreApp fallback) repopulated
Docker volume tars but NEVER replayed the captured <stack>-<dbtype>.sql dump, so
DB-resident data (e.g. rows in a DB whose data dir is a bind mount) did not come
back — the romm marker round-trip in the audit lost the row.
New appbackup.ImportDump (read-side counterpart to DumpOne) replays a .sql/.sql.gz
into the running DB using the live container's OWN discovered credentials (no env
threading; reuses DiscoveredDB + getMariaDBPassword). backup.reimportDBDumps
orchestrates it AFTER volume restore + stack bring-up, so the logical dump WINS
over any volume-tar copy of the DB (operator-chosen precedence). pg_dump
--clean --if-exists and mariadb-dump (default --add-drop-table) make replay
idempotent; psql ON_ERROR_STOP=1 surfaces real import errors.
Also: volume-restore per-volume failures and DB-import failures now SURFACE (the
restore returns an error) instead of a swallowed WARN, so a failed data restore
cannot read as success.
Tests (restore_db_test.go, injectable discover/import seams): imports when dump+DB
present, failure surfaces, no-dump skips discovery, dump-but-no-matching-DB is a
non-fatal skip. Live DB round-trip to be validated post-deploy.
Traefik only publishes a route to a healthy container, so an unhealthy deployed app
returns 404 at its URL though the container runs — previously shown only as 'Nem
egészséges' with no hint the URL is dead. New routeUnpublished() funcmap helper +
a distinct indicator on the dashboard and stacks cards (gated on .Deployed). Tests:
routeUnpublished across all states, real templateFS parses with the funcmap, and the
card guard renders the indicator only for deployed+unhealthy.
The Hub config-apply handler wrote controller.yaml 0644; it holds cf_api_token,
cf_tunnel_token and hub api_key in plaintext. New writeConfig0600 helper writes
0600 atomically (tmp+rename, bind-mount fallback) and chmods to enforce 0600 even
when the file pre-existed 0644 (os.WriteFile doesn't chmod existing files).
Test asserts mode 0600 (Linux; skipped on Windows). Setup path already used 0600.
The dashboard stacks list served the in-memory map refreshed by a 30s ticker, so
container state lagged Docker health by up to ~30s after a deploy. RefreshStatus
is a cheap docker-ps refresh; 10s (matching health-probes) cuts the lag without
loading Docker.
F4: GET /api/stacks/rescan fell through to GET /stacks/{name} → misleading
'stack not found: rescan'. Now returns 405 + Allow: POST.
F6: the deploy POST returns before compose/health complete (async; UI polls).
Message changed 'Stack X deployed' → 'Telepítés elindítva…' and status 200→202
Accepted, so API/script consumers aren't told a deploy finished when it hasn't.
UI checks data.ok (not HTTP status), so 202 is safe.
A failed agent format (e.g. 502 'device is mounted', ok:false, data:null) fell
through FormatDisk's trailing 'return out, nil', so the web layer reported a
zero-value FormatResult as ok:true — a failed DESTRUCTIVE format read as success.
postWithStatus now returns the full envelope; FormatDisk returns a non-nil error
on any non-2xx/ok:false that is not a recognized refusal (403/needs-confirmation).
Test TestFormat_MountedFailureSurfacesError (502 → non-nil err) fails on old code.
/api/system/info reported the Proxmox host's 16GB (the controller container
reads host /proc/meminfo with no lxcfs), defeating the deploy memory-headroom
hard-block (deploy.go uses GetMemoryMB). readMemInfo now prefers the cgroup
memory limit (v2 memory.max / v1 memory.limit_in_bytes; sentinels = unlimited)
when finite and below the host total; used = memory.current/usage_in_bytes.
Test info_cgroup_test.go (cgroup v2 cap wins, v2 max sentinel, v1 unlimited,
v1 finite) — fails on pre-fix code.
CLAUDE.md line 31 wrongly said 'bulk strip has NOT happened' — slice 8C executed
the de-privileging (storage/restic/watchdog deleted; disk ops via agentapi).
Corrected to current state + pointer to central docs. CONTEXT banner -> v0.60.0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
NewServer launches the SyncFileBrowserMounts goroutine (reads integrationMgr)
from the constructor, BEFORE main.go's SetIntegrationManager write — so the
init-only happens-before that covers the other Set* fields does NOT hold here,
making it a genuine data race (handlers.go:358/360/1433 reads vs server.go:162
write). Converted the field to atomic.Pointer[integrations.Manager]; setter
Stores, all 3 readers Load(). Regression test reproduces the concurrent access
(clean under -race; flags on the pre-fix plain-pointer field).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- controller/README.md: authoritative banner (v0.59.0, docs moved to
felhom.eu/documentation/controller/, bootstrap-deploy note); legacy body retained.
- CONTEXT.md: banner refreshed to v0.59.0 state.
- Removed controller/mnt/user-data/outputs/... — an accidental Claude-sandbox
output dir (stale duplicate README) committed in the initial import.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The single write (SetStackProvider, main.go:225) was mutex-guarded while all 11
reads were unlocked — the lock implied a runtime concurrency the reads don't
honour. It is called once during single-threaded startup before any goroutine,
so the write happens-before every read and no race exists. Removed the
misleading lock and documented the init-only contract.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
deploy.go, one slice (both edit SaveAppConfig / the deploy goroutine):
CTRL-T2-1 (ghost-deployed on crash): DeployStack wrote app.yaml Deployed:true to
disk BEFORE the async 'docker compose up -d'; a crash during the image-pull
window left a ghost-deployed stack (Deployed:true, no containers) that DeployStack
then refused to redeploy. Now the env is persisted with Deployed:false
(transitional), and Deployed:true is written by runComposeDeploy ONLY after up -d
succeeds. In-memory Deployed stays true during the pull to preserve the
no-stale-Telepítés-button UX. On a post-success save failure, revert so the stack
is redeployable.
H10 (plaintext secret on encrypt failure): SaveAppConfig logged a WARN then fell
through to persist the secret in PLAINTEXT. Now fail-closed: return an error on
crypto.Encrypt failure, never write plaintext. Callers already propagate it.
Regression tests: H10 fail-closed (+ good-key encrypts) and the CTRL-T2-1
transitional durable-state contract (transitional reads not-deployed).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
manifest.AppName / HDDSubdirs / VolumeNames are attacker-controlled JSON inside
an imported .fab and reach filepath.Join+MkdirAll/extractTar with a trusted base
(restore.go:339/606/678). UnmarshalManifest did zero validation, so '../..' in
any of them escaped the stacks / HDD destination dir.
- New appexport.ValidateSegment + validateManifestPaths; UnmarshalManifest now
fails the parse on a traversal segment (the chokepoint).
- Defence-in-depth ValidateSegment guards at the HDD-subdir and volume-name join
loops in restore.go.
- ConfigFiles deliberately NOT validated (holds dotfiles like .felhom.yml; never
used in a restore join).
- Permanent regression test (was the deep-sweep failing audit test) now asserts
rejection of traversal + acceptance of legit names.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Reserved-buffer headroom guard on the Docker-data volume (system/dockervol.go,
max(5GB,10%)); deploy-time hard gate refuses (HTTP 507) when below the buffer
(api/router.go); deploy page warns + disables the button (deploy.html); runtime
disk monitor confirmed to watch the Docker volume above the buffer. Log rotation
baked into the golden (agent side). Phase 1 = felhom-agent v0.29.0.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Part A of the UI-fixes/storage-spike spec.
A1: enrichHostStorageTargets sorts /api/host-metrics storage_targets
server-side and attaches friendly Hungarian labels + purpose, fixing the
#host-storage-bars reorder-on-poll bug. Display labels only — PVE storage
ids are never renamed.
A2: new GET/POST /stacks/{name}/backup Tier-2 config panel; the "2. mentés"
Beállítás button is repointed there from the dead-end deploy page. Customer
can pin a target drive or disable Tier 2; preference is preserved across the
runner's status writes. Always visible (single-SSD + non-HDD apps included).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
4A: scope FileBrowser bind to <drive>/appdata (recovery units + Tier 2 copies under
backups/ are no longer mounted into FileBrowser — customer can't browse/delete the
thing that restores them). 4B: deploy storage-selection step states the chosen drive
holds files while the DB runs on the fast internal SSD + is backed up with the app.
4C: buildStorageBars stable sort + purpose description on the monitoring storage list.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
admin