fallback.sh starts covered containers with docker start and never creates them, so a coverage list
the partner has never been sent is a promise nothing can keep — all twelve were missing. Adds the
push and remove paths, a readiness card that checks rather than infers, and the fallback state the
assistant needs to answer for it.
The card described the machine and said nothing about the thing whose dashboard
it is. Cache size matters most: VV_CACHE_ROOT is under /tmp, which on Unraid is
RAM, so it is shown against the rootfs percentage it counts against. Jobs are
locks with a live process — four lock files were sitting there and one job was
actually running.
A cross-scope identifier and a fetch chain ending in an empty catch both pass
php -l and node --check, and together they turned a ReferenceError into what
looked like a slow load for hours. --self-test asserts both detectors still
find a known fault, because a checker that silently stops working reports a
confident zero.
Every turn runs on the AI owner now, so the ledger's per-host column had
collapsed to a single name and the fleet breakdown said nothing. The asking
node travels with the request and is validated before it is written, which
also retires ai_token_sync.sh — a partner ledger is empty by construction.
Curated state copied to every node is state that can disagree, so the index,
the model and the shared memory stay on the owner and each node reaches them
over the SSH trust onboarding already builds. Chats stay on the node that had
them; memory and bug reports stay the owner's to write.
The old card totalled rsync's own logs, so it reported 'no data moved' across a link that had carried hundreds of gigabytes over SSH, the arr APIs, conf pushes and the Unraid API.
Comments asserting a partner is down, being rebuilt or not yet installed are true on the day they are written and wrong afterwards, and nothing prompts anyone to revisit them. Behaviour is described generically; current state belongs in operator memory, not in tracked files.
Step 12 filed all thirteen deployed containers under <Owner>-Fallback, which says whose they are and nothing about what they do — and they are not failover coverage, they run there continuously.
The .plg hardcoded the flash path and runs every boot, so an appdata node served a stale copy that never receives pulls — fixes appeared to do nothing, indefinitely.
Both checks already answered their question on demand and both needed
somebody to press a button on the right row. One host here has returned
nothing but 5xx for months.
Filed as findings, which is the existing answer to a condition that
persists while nobody is looking. Grouped by cause rather than by
hostname: a default policy of bypass produced twenty-two findings that
were one sentence repeated, and they have one fix between them.
A low uptime figure, a refused login and a certificate that stopped
renewing all looked the same from the row: a number, with the reason
split across NPM, an Authelia config and the directory.
The why-check goes and looks — TCP to the forward target, HTTP through
the proxy, a second handshake with verification off to tell a broken
certificate from a broken service. Forward hosts are docker names that
only resolve on NPM's network, so an unresolvable one is redirected to
the container address and the substitution is reported; a check that
could not be made must never read as a check that failed.
The access simulator walks the rules the way Authelia does and shows the
ones it stepped over, reading whichever instance the chosen host points
at rather than the one conf names — there are two here.
Cert triage counts runs rather than log lines and orders by rotation
suffix rather than mtime, both of which change the answer.
NPM's access log only describes hosts somebody visited; the host most likely to be quietly broken
is the one nobody does. Probes carry a User-Agent npm_access_stats.sh drops — unmarked, this
monitor would be fifty thousand requests a day in the very logs it reports on.
NPM writes an access log per host and counts nothing, so 475 MB of logs held the only answer to
"is anything using this". Aggregated on a schedule and read from a few kB of JSON; the row also
now says whether an auth_request block is in front of the site, which nothing showed before.
NPM knows what a certificate is today and nothing about what it was, so ten of them could fail
renewal for months — 1001 certbot runs, zero successes — without anything on any page saying so.
Counts start at zero and are only ever observed; only first_seen is seeded, from NPM's own date.
vv_is_ai_host() was `=== 'host1'`, which made a physical fact — that is where the
GPU is — look like a rule. AI_OWNER_HOST declares it, so the card can move to a
rebuilt host3 or a friend's spare.
The gate was also doing two jobs. Assistant docks and findings strips now ask
whether a model is reachable, so a node without a GPU gets them by borrowing;
the AI tab asks whether this is the owner, because that page carries the bug
reports, the index and the model configuration — the surface where the
vocabulary assumes you built the mesh.
Resolution is local, then owner, then anyone else declaring a model, pinned once
it answers. Pinned rather than re-derived per call: a mesh that re-decides every
request eventually decides differently mid-conversation, and a chat whose second
turn lands on another machine has no history there. Cleared only on a transport
failure, and only when there is somewhere else to go — a single-node mesh whose
model is down should say so, not report AI as unconfigured.
A partner's Emby URL is http://localhost:8096 — true there, meaningless here —
so nobody queries a partner's media server directly. Each host reports its own
over the SSH call remote_arr_cache_writer.sh already makes, which keeps its API
keys on it and costs no extra connection. Local is the default: opening the tab
to check your own server should not first make you read past a partner's.
A partner with no cache is stated as not collected rather than omitted, because
an empty mesh view and a partner that is down look identical otherwise.
The tab collected over SSH on every 30s poll — 8.3s a load with HOST2 down.
vv_parse_conf_scalar() captured to end of line, so a commented toggle parsed as
"true # HOST2 back online": every threshold read right because (int) stops at
the first non-digit, and 38 booleans read wrong. The Fallback tab has been
showing failover disabled while it was on.
The detector answers no to "why is my array so slow", so claiming that meant a
question about the wider world would have answered it from general web pages.
A paragraph beside the rule did not shift it; the ready-made refusal was simply
the nearest thing to hand, so the correction goes next to the question instead.
Chat was told it could not look anything up, which is true of this machine and
false the moment a search succeeds, so it deflected while holding six sources.
The map documents 447 settings, so without this every question about where a
control lives earns a proposal that costs a dismissal and teaches nothing.
The comments above a setting already state its choices, units and bounds, so the
form can read them instead of asking for the file to be annotated first.
Gmer4Lfe