PodmanClient::pull_image had a single attempt with a hard 600s wall
clock — on a fresh node, any image that legitimately takes >10 min to
download got killed mid-pull and failed the install. Podman keeps the
completed layers, so the user's retry started half-done and succeeded:
the exact "first install fails, second works" pattern reported for
btcpay and indeedhub (and earlier for LND, whose fix in 72d7fa07 only
covered the legacy install path — this is the same cure ported to the
orchestrator path both fresh installs and upgrades use).
A pull is now only killed when nothing is observably happening — no
stderr output AND no staged-byte growth in TMPDIR for 180s — or at a
1800s absolute ceiling, with 3 attempts and backoff on top. Dead
registries still fail fast (no bytes ever land). Failures now surface
podman's actual stderr tail instead of a bare exit status, and the
pull stages via the user's containers tmp like every other pull path
(rootless /var/tmp is read-only).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Orchestrator-installed stacks (indeedhub, btcpay, netbird, immich,
mempool) sat at "Preparing… 5%" for the entire install — the
install_stack_via_orchestrator path never called set_install_phase /
set_install_progress after the initial Preparing, so a 7-image,
multi-gigabyte indeedhub pull gave the user zero feedback ("just says
preparing for ages"). The legacy installers already report phases and
the frontend already interpolates X-of-N across the PullingImage band;
the new path simply never fed it.
Now: PullingImage phase up front, an X-of-N counter plus a per-member
"Installing postgres (2 of 7)…" message as each member installs, and
the legacy installers' truthful PostInstall→Done settle at the end.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The full backup carried identity/lnd_aezeed.enc but NOT the per-node
wallet secret that encrypts it (secrets/lnd-wallet-password), so a
restored node could never decrypt its Lightning seed. The secrets dir
now rides the backup (archive v3; the .bak itself is
passphrase-encrypted). Restore gains a staging-presence guard so
restoring an older archive without a secrets dir can no longer displace
and delete the node's live secrets — covered by two new tests.
Backups can now also be downloaded through the browser: a session-gated
GET /api/blob/backup/<id> endpoint (under the existing nginx /api/blob
prefix, so no fleet nginx changes) plus a Download button next to
Verify / USB / Restore / Delete in Settings → Backup & Restore.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The health monitor opened a bare TCP connection to every published host
port each cycle. Against TLS listeners that connect+close logs noise on
the app side — LND printed 'http: TLS handshake error … EOF' endlessly.
Containers that define a podman healthcheck (LND uses lncli getinfo)
now rely on it alone; bare containers keep the port probe.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two confirmed fresh-node failure modes made 'install LND' fail on the
first try and succeed on the second:
- The dependency gate failed fast when Bitcoin was still mid-install
(image pulling, container not created yet): DepProbe.existing only
covered 'podman ps -a', so an in-flight dependency looked identical to
'not installed at all'. The probe now also reports packages in
Installing/Updating state, and the gate waits up to 30 minutes for an
in-flight dependency install to finish (surfacing 'Waiting for X to
finish installing…' on the card) before its normal 3-minute
started-but-not-running window.
- Image pulls had a hard 300s per-URL wall clock, which killed
slow-but-progressing pulls of large images (LND on fresh-node WiFi);
the retry then succeeded off the warm layer cache. Pulls are now
stall-aware: killed only after 3 minutes with no podman output AND no
byte movement in the staging dir, with a 30-minute absolute ceiling.
Dead registries still fail in ~3 minutes.
Adds dep-gate unit tests for the in-flight-install wait and the
neither-installed-nor-installing fail-fast.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
server.set-name ran hostnamectl but left Debian's 127.0.1.1 line on the
old name (breaking sudo's 'unable to resolve host' and hostname -f) and
waited on avahi to notice the change by itself. Sync /etc/hosts and
avahi-set-host-name (fallback: try-restart avahi-daemon) right after a
successful rename, so http(s)://<hostname>.local works immediately —
best-effort, never blocks the rename.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
'Add Service' only worked for the 19 apps in the hardcoded
known_service_port map — everything else failed with 'Unknown app'.
tor.create-service and tor.toggle-app now fall back to the app's live
launch address from the scanner state, so any manifest-driven app gets
a .onion without a per-app entry (the static map still wins for
protocol services like bitcoin, which must expose 8333, not a UI port).
toggle-app also no longer writes a broken port-0 HiddenServicePort for
unknown apps.
Every successful package.install now auto-creates the app's hidden
service (detached, best-effort, retries while the scanner derives the
port; skips protocol services and existing entries).
The demo mock gains stateful tor.create-service/tor.delete-service so
the demo round-trips instead of erroring.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
On a fresh ISO node, NetBird's own WireGuard tunnel (10.44.0.1) sorted
ahead of the real NIC in hostname -I, so the NetBird launch URL (and the
{{HOST_IP}} baked into its cert/config) pointed at the tunnel instead of
the LAN address. The main routing table's default route names the
physical uplink even while a VPN is active (NetBird/Tailscale steer
traffic via policy-routing rules, not the main table), so read src/dev
from 'ip -4 route show default' first, then fall back to a connected UDP
socket's source address, then to the old hostname -I scan.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
barkd (Ark wallet daemon, pinned 0.3.0, checksum-verified release binary)
packaged as an installable app; thin HTTP bridge in wallet/ark_client.rs
mirrors the fedimint_client pattern — the bark SDK stays out of the node
binary. wallet.ark-* RPCs cover status/balance/address/send/invoice/
board/offboard/history/configure; Ark movements merge into the unified
ecash history (kind="ark") and spendable Ark sats into total_sats.
Signet defaults (Second's public Ark server) until Ark matures.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The optimistic package_data entry painted at install-spawn survived
failure as a 'Stopped' tile for an app with zero footprint (netbird on
the fresh ISO). On INSTALL FAIL with no container/user-stop marker for
the app, the entry and the pre-saved dynamic app config are removed and
an error notification is pushed; failures that left a real container
keep today's behavior (visible exited state for retry/diagnosis).
Install suite 39/39.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
write_bitcoin_conf silently produced an unpruned (~810GB and growing)
config regardless of disk: the framework node got a full-archive chain
on a 205GB volume. Volumes under 1.2TB now get prune = 25% of the
volume, clamped to [550MB, 100GB]; bigger volumes keep full archive;
unknown df results stay unpruned (don't surprise big iron). Framework
node switched live to prune=50000.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
936b4cca made ANY loaded manifest with an absent container self-heal via
install_fresh — but the signed catalog and the ISO ship manifests for
EVERY app, installed or not, so 'manifest loaded' is not installation
evidence. On a fresh node this mass-installed the catalog: the framework
node grew portainer, vaultwarden, searxng, strfry, mempool, immich and
fedimint uninvited within an hour (fleet nodes showed the same loop as
reconcile-failed noise for unpullable images).
ExistingOnly (boot) reconcile now creates containers from nothing only
for required-baseline apps (filebrowser, fedimint-clientd); every other
absent app is Left('absent') and recreated ONLY via desired-state
recovery when it was running at the last snapshot — which still covers
the vanished-container regressions (indeedhub/immich) that motivated
936b4cca. Reconcile suite 16/16.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Plugging in a LoRa radio now pops a global two-step setup modal on any
page: step 1 shows the ACTUAL detected board (25 board SVGs vendored
from the Meshtastic web-flasher, GPL-3.0; matched by USB product string
on native-USB boards, vid:pid heuristics for bridge chips, animated
fallback otherwise); step 2 configures with Archipelago presets — the
LoRa region pre-selected from the node's location (new lat/lon→region
mapping), channel 'archipelago', node name, and an advanced firmware
pin. Options adapt per protocol: region+channel program Meshtastic;
MeshCore shows its community frequency plan (firmware owns RF); RNode
defers to the Reticulum daemon config.
Backend: mesh.configure now accepts lora_region (validated against the
driver's region table) and device_kind (probe pin); mesh.status returns
the persisted values plus per-port USB identity (sysfs vid/pid/product)
for board identification. The Mesh Device tab gains a full settings
form (region with duty-cycle/legality hints, firmware pin, channel,
name, identity broadcast) replacing the read-only panel; the old
Mesh-page-only onboarding modal is superseded by the global flow.
Mock backend: stateful mesh config so the dev UI demos the full
detect→configure round-trip (disable mesh → modal fires with Heltec V3).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fresh ISO installs had ZERO disk manifests: auto-install.sh copied
apps/ to /etc/archipelago/apps but the backend/orchestrator loads
/opt/archipelago/apps — so any app not overlaid by the signed catalog
(netbird: 'manifests not available on this node') could never install.
- installer now also populates /mnt/target/opt/archipelago/apps
- bootstrap gains run_apps_dir_repair(): when /opt/archipelago/apps is
empty and the installer copy exists, populate it at startup — heals
already-deployed ISO nodes via OTA without reinstalling
- bootstrap's bitcoin.conf repair now also ensures rpcbind=0.0.0.0
(same fix as install.rs; duplicate rpcbind with a manifest command
line is harmless — bitcoind binds the first and logs the rest)
Verified live on the framework node: /opt populated (56 dirs), backend
restart loaded '50 app manifest(s) (disk + registry catalog)', all
containers kept running (they live in user.slice, not the service
cgroup).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Root cause of the fresh-install triple failure (LND needing ~5 install
attempts, Bitcoin UI bitcoin-rpc 502, backend getblockchaininfo errors):
the legacy conf writer set rpcallowip but never rpcbind, so bitcoind
bound RPC to the container's loopback only — every dial over the
container network (LND, bitcoin-ui) and the rootless port-forward was
refused. LND exited instantly, the 60s post-start poll read that as
INSTALL CRASH, and the wallet (and its seed backup) never got created.
- write_bitcoin_conf + ensure_bitcoin_rpc_config now emit/ensure
rpcbind=0.0.0.0 (host exposure unchanged: publish stays 127.0.0.1)
- new install gate: bitcoin-dependent apps (lnd/electrs/electrumx/
mempool-electrs/btcpay) wait up to 3 min for an authenticated
getblockchaininfo to answer before their container starts, with
INSTALL WAIT progress lines; on timeout the install fails with an
actionable message instead of a crash-looping container
Verified live on the fresh-ISO framework node: after adding rpcbind and
restarting bitcoin-knots, LND connected, created its wallet, baked
macaroons, and settled into 'Waiting for chain backend to finish sync'.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
484.kitchen, Arch Presentation, Call the Operator, Next Web News
Network, Syntropy Institute, and T-0 were hardcoded web-only bookmark
apps (no manifests/containers). Removed across the frontend app grid /
curated / marketplace / session / launcher data, the backend
AppMetadata table, their icons, and the nginx /ext/ proxy blocks
(external-app-proxies.conf deleted — nothing installed it anymore).
botfights/nostrudel/gitea keep the shared external-app code paths.
Frontend: vue-tsc clean, 672 tests pass, production build verified
clean of the removed slugs. Backend: cargo check clean.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The tests landed twice (interrupted edit applied both paths), breaking
the test build with E0428. Federation suite: 33/33 green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Invites carried no trust level, and both accept_invite and the
peer-joined handler hardcoded TrustLevel::Trusted — so 'Invite a Peer'
(Observer) federated both sides as fully Trusted.
- invite codes now carry a 'trust' field (legacy codes parse as Trusted)
- federation.invite accepts trust_level: trusted|observer
- accept_invite assigns the invite's level on the acceptor side
- peer-joined resolves the granted level authoritatively by matching the
acceptor's echoed invite token against our stored outgoing invites;
the peer's own unsigned claim is honored only as a downgrade, so no
escalation is possible
- discovery/connection-request approvals mint Observer invites directly
(the post-hoc demotion in handshake.rs stays as a legacy safety net)
- asymmetry self-heal re-asserts at the locally-held level
Tests: observer threading + legacy default-trusted parse.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
telemetry.fleet-status only read the telemetry-fleet/ collector inbox —
an opt-in anonymous-telemetry pipeline that federated peers never write
to — so Fleet showed 'No nodes reporting' even with a healthy federation.
Build the fleet list from nodes.json instead: every TRUSTED federated
node is mapped from its last_state sync snapshot (cpu/mem/disk, uptime,
apps, peer count) into the FleetNode shape the UI expects, keyed by DID.
Observer ('peer') and untrusted nodes are excluded by design. Collector
reports are still merged in for non-federated telemetry nodes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Web5ConnectedNodes sent { request_id } but the backend only read
params.id, so accepting/rejecting a connection request always failed
with 'Missing required parameter: id'. Frontend now sends id; backend
accepts both spellings for older deployed UIs.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- aezeed words are now captured at BOTH wallet-init paths and stored
encrypted (Argon2 + ChaCha20-Poly1305, per-node wallet secret) at
identity/lnd_aezeed.enc; ack marker cleared when a wallet is recreated
- lnd.init-wallet-from-seed was posting seed_entropy to /v1/initwallet,
which is a GenSeed field — the wallet was never actually derived from
the master seed; now GenSeed(entropy) → InitWallet(words)
- new RPCs: lnd.seed-backup-status / lnd.seed-reveal (password + 2FA
gated, same as seed.reveal via shared verify_reveal_auth) /
lnd.seed-backup-ack
- LND app detail page: Lightning-seed card with first-launch backup
prompt, tap-to-reveal modal, 'I've backed it up' acknowledgment
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Kiosk (backgrounds + FIPS mismatch, one root cause each):
- kiosk service now also waits for the web-ui asset swap (probes a large
bg asset, bounded 60s) — it launched mid-rsync and CSS background 404s
are never refetched, hence blank backgrounds on first boot
- FipsNetworkCard polls every 15s instead of fetching once on mount
- PWA updates auto-apply on the kiosk (localStorage kiosk flag) — an
unattended kiosk could never click 'Update Now' and ran stale bundles
LND App Store:
- catalog icon pointed at lnd.svg which doesn't exist (lnd.png does);
store card also retries the sibling extension before the placeholder
- image pulls retry once after fast transient failures (<30s) — fresh
first-boot networks reset connections; slow failures aren't retried
so the UI never waits longer than before
Mesh detection:
- mesh.status now always reports device_present + detected_devices
(previously only when the service wasn't running, so the UI couldn't
tell 'no radio' from 'radio present, session connecting')
- Mesh page shows 'Mesh device detected' badge + 'Device detected —
connecting' status for present-but-not-connected radios
- the 99-mesh-radio.rules udev rule was never staged into the ISO tree
(installer searched for it and found nothing) — ship it at the media
root and install it from install-to-disk.sh too
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- image: create archipelago user with dialout so the backend can open
/dev/ttyUSB*/ttyACM* LoRa radios; doctor Fix 14 heals existing nodes
(verified live on the .65 RC install — device detected after the fix)
- Web5Federation.vue called nonexistent federation.list (Unknown method);
now uses federation.list-nodes + federation.list-pending-requests with
a last-seen-based online count
- sanitizer: let app-install dependency errors through — 'LND requires a
running Bitcoin node' was masked as 'Check server logs', so install
failures on fresh nodes looked random
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- new lnd.lightning-history RPC: settled invoices (incoming) + succeeded
payments (outgoing) normalized to the wallet-transaction shape; on-chain
history stays in lnd.gettransactions
- Home merges on-chain + lightning + ecash history into one sorted list;
every entry now carries kind (onchain/lightning/cashu/fedimint)
- modal renders a rail badge for non-onchain rows instead of a bogus
confirmation count, and only on-chain rows link into mempool
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Same kiosk.service conflict resolution as the previous merge (branch carried
rebased duplicates of the kiosk commits): CPUQuota=200% + main's memory tuning.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The tx1138 Gitea was retired as a release server 2026-06-13 and its registry
frontend is fully dead (500 on every /v2 manifest read, observed 2026-07-10).
Nothing may reference it anymore:
- registry.rs: no longer a default registry, no longer force-enabled on
load; saved configs are stripped of it on load (same one-time migration
treatment as the decommissioned Hetzner mirror), with a regression test.
- image_policy.rs: removed from TRUSTED_REGISTRIES — refs through the dead
host are now refused at the pull site (rejection test added).
- api/handler: dropped the legacy catalog-proxy fallback URL.
- .gitmodules: indeedhub submodule repointed to the OVH Gitea.
- scripts, image-recipe, app-catalog data, neode-ui strings, docs, and all
test fixtures repointed to 146.59.87.168:3000 (or neutral example hosts).
- image-versions.sh: ARCHY_REGISTRY_FALLBACK emptied (guarded consumers
skip it); reconcile-containers.sh candidate guard hardened.
The only remaining occurrences of the host string are the strip/reject
enforcement paths and their regression tests — the code that guarantees it
is never used again.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Materialise /var/lib/archipelago/secrets/btcpay-lnd-connection
(type=lnd-rest;server=https://lnd:8080/;macaroon=<hex>;certthumbprint=<sha256>)
in ensure_app_secrets before btcpay-server env resolution.
The macaroon travels inline as hex because LND's datadir is owned by its
container subuid (100999) — bind-mounting it into btcpay's userns EACCESes.
Reads the macaroon via the existing read_file_as_root sudo fallback; pins the
TLS cert by DER SHA256 thumbprint and rewrites on cert rotation. No-op while
LND is absent (btcpay's secret_env entry is optional), and generation errors
log-and-continue so btcpay never fails to start over this.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A secret_env entry with optional: true is skipped when its secret file is
missing, unreadable, or empty, instead of failing the whole resolution and
taking the app down. Needed for per-node integrations like btcpay's
internal-LND connection string: nodes without LND must still run btcpay.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
btcpay-server started and reported healthy while its declared
dependencies (archy-btcpay-db, archy-nbxplorer) failed every boot
reconcile (live-testing report 2026-07-10, Fix 3). After each
reconcile pass, flag every app the pass left up whose manifest
dependency landed in the failure list — a loud DEGRADED journal error
instead of silence. True start-gating on dependency health is a
larger design change (boot-ordering/deadlock risk) and stays open.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The strfry/vaultwarden 8082 collision shipped in a signed catalog
because nothing checks cross-app host ports. Fix the remaining latent
pairs from the cfb8d953 audit — did-wallet 8083→8088 (filebrowser
keeps 8083), morphos-server 8086→8089 (netbird-server keeps 8086),
lightning-stack 8087→8091/9737→9738 (netbird and fedimint-gateway
keep theirs) — and fix their healthchecks, which probed the HOST port
inside the container (same class as strfry 6fe62355; now 127.0.0.1 +
container port). load_manifests now flags collisions across the
merged disk+catalog set (bitcoin variants and the mempool umbrella
pair are exempt as mutually exclusive), and a repo-level test parses
apps/*/manifest.yml so a future latent pair fails CI instead of
shipping.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Channel open fee control, channels tab in wallet settings, mempool tx
link, and the channel-open async peer-connect race fix. Conflict in
LightningChannels.vue resolved to the branch's extracted
LightningChannelsPanel component, with main's Teleport-to-body modal
fixes and glass-button-warning class (8256fde1) ported into the panel
so they aren't lost. lnd tests 16/16, LightningChannels vitest green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
24fd97ed cascades on package.start/restart, but a reconcile pass that
recreates or starts a bitcoin backend (desired-state recovery, repair
recreate, boot InstallMissing) also moves the RPC address behind a
running lnd's back — §C 'restart lnd after ANY bitcoin recreate'.
After the pass, dependents from the shared address_caching_dependents
table (moved to app_ops as the single source of truth) that sat
untouched (NoOp) and aren't user-stopped are restarted under their op
lock; a dependent the pass itself (re)created already resolved the
fresh address and is left alone.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Fee selector on channel open: Standard (~6 blocks) / Medium (~3) /
Fast (next block) presets, or custom target confirmations / sat/vB;
backend validates and passes target_conf / sat_per_vbyte to LND
- Extract channel management into LightningChannelsPanel, reused by the
LND channels page and a new Channels tab in the Wallet Settings modal
- Funding tx link on each channel card opens the node's mempool app at
/tx/<txid>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
wait_for_manifest_host_ports polls up to 420s while the caller holds
the per-app lock, so a user's stop request queued behind it and the
scanner painted the app Restarting for the whole wait (the dd3afbba
transitional-state stick; gate take-2 iter 5 vaultwarden). The wait
now selects against a user-stop-marker watcher (2s poll on
user-stopped.json, by app id or container name) and aborts with a
recognizable error, and the host-port repair path skips its restart
when a stop was requested while it settled — restarting there would
fight the queued stop worker.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Connect to peer synchronously (perm=false, 30s timeout) and check the
result before opening the channel; previously the connect was fired
with perm=true (queued in background, returns immediately) and its
result ignored, so the open failed with 'peer is not online'
- Let LND channel/peer errors through the error sanitizer so the UI
shows the real cause instead of 'Check server logs'
- Add private (unannounced) channel option to backend and UI — some
peers (e.g. Olympus/ZEUS LSP) reject public channel opens
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The generated HealthCmd reported healthy (exit 0) whenever the image
shipped neither wget nor curl — btcpay-server sat 'healthy' in podman
ps while completely non-functional (live-testing report 2026-07-10).
Try a bash /dev/tcp connect to the health URL's host:port before
giving up; only images with none of wget/curl/bash still skip Podman
health (dropping exit 0 entirely would restart-loop those). Podman's
health timeout bounds the connect attempt.
Note: the rendered unit text changes, so every http-healthcheck app
gets a one-time unit rewrite + restart on first reconcile after this
binary lands.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Reconcile recreates route through install_fresh, which pulled
unconditionally: with git.tx1138.com dead, every boot reconcile of
archy-btcpay-db/archy-nbxplorer hard-failed on 'pulling ...' even
though both image:tags sat in local storage the whole time
(live-testing report 2026-07-10). Check image_exists first and only
pull when the ref is truly absent — the same semantics as
ensure_resolved_source_available and the generated quadlets'
Pull=never.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The health monitor restarted containers that were legitimately down
between a package.start/stop/restart worker's stop and start halves —
the same interleave class as the reconciler's mempool repair-recreate
(e275494a). Before restarting, probe the app_ops op-lock registry via
the container name, the derived app id, and a '_'->'-' normalization
(legacy stack containers are underscore-named while lock keys use
hyphenated app ids) and skip the cycle when a worker holds the lock.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Live verification on .228 showed catalog_changed:true on every
check-updates call: AppCatalog.apps is a HashMap, so the re-serialized
cache bytes flap with key order and the changed-detection never reports
"unchanged" — which would have reloaded the manifest overlay every
hourly poll. Cache the raw fetched body instead and compare against
that; as a bonus the cache now holds the exact signed preimage, so a
present release-root signature stays verifiable from disk.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
load_manifests() only ran at startup, so a manifest published via the
signed catalog sat dormant until the next service restart (found while
shipping the strfry manifest fixes: package.check-updates refreshed the
cache but the orchestrator kept rendering the old overlay).
- refresh_catalog now reports whether the cached bytes actually changed
(write_cache skips identical rewrites), so unchanged hourly polls
don't churn the manifest map.
- New ContainerOrchestrator::reload_manifests (default no-op); prod
delegates to load_manifests — the rebuild is atomic under the state
write lock, and the reconciler's durable user-stopped/uninstalled
marker filters make a mid-session reload equivalent to the restart
path that already runs on every boot.
- package.check-updates reloads on change and reports catalog_changed +
manifests_reloaded; the hourly update-scheduler tick (and its startup
refresh) do the same.
Tests: app_catalog 5/5 (new write_cache changed-detection test),
reconcile 16/16, knows_app 1/1.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
package.install routed to the manifest-driven orchestrator only for a
hardcoded per-app allowlist; every other app fell through to the legacy
flow, which ignores manifests entirely and creates a bare container with
no ports or volumes (strfry crash-looped this way on .228, 2026-07-09).
New ContainerOrchestrator::knows_app(app_id) (default false; prod checks
its loaded-manifest map, disk + signed-catalog overlay). The install gate
is now allowlist OR knows_app — no per-app Rust for manifest-driven apps,
matching the packaging invariant. Apps without a manifest keep the legacy
flow including the container-exists adopt block; the unknown-app_id →
legacy fallback inside the orchestrator branch is unchanged.
Tests: knows_app_reflects_loaded_manifests + install suite 37/37.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Formal gate take-2 went 4/5: iteration 5's vaultwarden stop-wait saw
'restarting' for the full 120s window. A reconcile pass had queued every
app into pending_boot_starts moments before the stop marker landed, and
the scanner's overlay painted the deliberately-stopped app Restarting
until the slow sequential pass reached it (legacy container, so no unit
state to contradict it).
Two filters: the scanner overlay skips user-stopped ids (the marker is
written before the stop runs, so it's reliably present), and the
reconciler no longer queues apps whose lifecycle op is in flight into the
pending overlay (it skips them via the same probe anyway).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
