A model whose profiles run engine comfyui is placed on a node's card like
any other, and makes pictures there. ComfyUI asks for no key, so the agent
starts it behind llmgw comfyui-gate: ComfyUI on the loopback, the gate on
the replica's port demanding the replica's key and letting through only
what the router asks — queueing, waiting, fetching, cancelling. The gate
answers the agent's /health and /v1/models, shows ComfyUI the fetched
weights by their own names, and gives each replica its own base directory.
ComfyUI runs as the gate's child, so its card memory is the replica's.
A node offers the engine when agent.yaml names its ComfyUI
(engines.comfyui.dir). The model's comfyui: block names the workflow and,
for a source that is a set of files, the checkpoint.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
An external model with api: comfyui is a ComfyUI server. A request for
pictures goes through the same key, allowlist, limit, admission and guard
checks as any other; at the replica it becomes a workflow — a built-in
(flux-schnell, flux-dev) or a JSON file in API format, with {{prompt}},
{{width}}, {{seed}} and the like filled in — queued, waited on, and its
pictures sent back as b64_json. A client that leaves takes its workflow
off the queue or interrupts it. ComfyUI is never scraped for metrics.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The console and ctl on a standby showed its own empty view: nodes and
replicas live in the leader's memory. A standby now passes /api/ to the
leader over the agent port, with the cluster's certificates both ways; the
leader answers only a certificate naming a control plane the store lists,
and only while it leads. Control plane certificates are issued for client
use as well, and server add reissues older ones.
An agent's reconnect backoff starts again after a session that worked, so
each failover it lives through no longer makes the next one slower.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Conflicts with facility-aware placement (config, control, admin API routes,
store test tables) all keep both; the generated reference and OpenAPI
document regenerated.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
ctl federation add/list/remove/sync keeps peers in the store (migration
0029); their /v1/models are offered here as external models named
PEER/MODEL and kept in step, a model still named by a fallback kept
withdrawn. Peers are checked at /ready, timed, and taken out of routing
while down. Requests to a peer carry x-llmgw-hop and are refused (508) when
they come back to a cluster they passed through; residency and no_store
keep requests off peers; X-Request-Id and, with --send-tags, attribution
tags go along. ctl model fallbacks sets overflow to a peer. Console view,
metrics, admin API routes and docs.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Nodes say where they are with llmgw.io/site, room, row, rack and pdu labels,
from agent.yaml or `llmgw ctl label` (kept in the store, over the agent's).
Placement spreads a model's replicas across racks and PDUs, keeps a
multi-node replica within one domain when its profile says keep_within,
places nothing in a domain held with `llmgw ctl facility hold` (optionally
draining it), and refuses a shard that would take a domain past its power
cap from facility.budgets, counting declared, derived and measured draw.
Explain-placement names the domain, and keeps the rejections of a placed
replica; `ctl nodes` gains LOCATION and --domain; `ctl facility` lists the
domains; the console's Nodes view shows and sets location and holds.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The cluster page draws every card coloured by utilisation, memory or
temperature, and lists replicas whose every card is under 5% busy.
llmgw_gpu_assigned says which replica, model and owner hold each card.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A node that cannot run the CUDA 13 build installs the same version's
CUDA 12.9 build from vLLM's GitHub release, with PyTorch's cu129 torch;
the install computes on the GPU before it counts. The build can be
chosen per node — auto, cu13, cu129, none — on the cluster page, with
ctl engine-build, or over the admin API; the control plane keeps the
choice and tells the node at once.
Refs #2.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A catalog model with external {url, model, api_key_file} and no profiles
is published as an endpoint by the control plane, the key read from its
file each time; the router rewrites the request's model to the provider's
name and sends its key. It is never scraped nor stale, and nothing is
placed for it. Named in fallbacks, it serves when the cluster has no room.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A model has an owner (org, team or project). Quotas guarantee a scope
GPUs, cores and memory and may limit it; what a scope holds is summed from
its models' replicas and those beneath it. A replica within its owner's
guarantee is not taken for another scope; what is held beyond a guarantee
is borrowed and taken back for a scope within its own. A limit stops a
model before it starts, and headroom stops there too. ctl quota and ctl
model owner set them; metrics and a CapacityAtLimit alert report them.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A model's max never did anything: desiredFor returned the minimum, or one
for an on-demand model someone was asking for, whatever the queue. Six small
CPU nodes run out of slots quickly, and a model with room to grow stayed at
its minimum while requests waited.
The router already scrapes each engine's running and waiting requests. The
control plane now reads them per model on every tick, alongside when each
replica was last used, and the reconciler keeps how many replicas load has
added above a model's base. One or more requests waiting per ready replica,
held for a minute, adds one; ten minutes with nothing waiting removes one of
those again. It never goes above max or below the base, moves one step at a
time, and does not add while the last addition is still placing or starting
— on this fleet a replica is minutes of weights and gigabytes of memory, and
a spike should not leave a pile of replicas behind it. Decisions are logged
with the load behind them.
Nothing live changes until a model's max is raised above its min.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Every upgrade so far has been a script of mine copying a binary to each
machine over ssh, restarting it and polling until it came back — which is
the job, but not something the product could do, and which needed ssh into
every node.
`sudo llmgw ctl upgrade VERSION` on the control plane does it. This machine
first: the release for its platform, checked against the release's
checksums, the old binary kept beside it, llmgw-server restarted, and a wait
until every node that was ready has reconnected. Then each node in turn: the
control plane tells its agent the version over the stream it already holds;
the agent fetches and checks the release itself, swaps it in with a rename,
answers, and exits with status 75 so systemd starts the new binary. Engines
keep running, as for any agent restart, and are adopted again.
A node is done when it is back on the new version, ready, with as many ready
replicas as before; one that is not within the timeout stops the run before
the next node, so a bad release reaches one machine. Nodes already on the
version are skipped, so running it again resumes.
The fetch-and-verify is one package now, shared with onboarding's idea of
what a release version looks like.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
On the live cluster kev-0.5b — an architecture the nodes' llama.cpp did not
know — was started 155 times in eight hours, failing the same way each time,
and phi-4, which fitted on no node, logged "no placement" 250 times in two
days. Neither showed up anywhere an operator looked: the backoff never gave
up, no metric said a model was failing or unplaceable, and the one alert for
a model with nothing serving could not fire, because its series was published
only while a replica was ready.
After five failures in a row a model is parked: not placed again until its
definition changes — a new source or profile — or `llmgw ctl retry` is run.
Status, the model list and the admin API show it with its last error. The
placement warning is logged when its reason changes or every ten minutes.
A request for a parked or unplaceable model is answered at once with why,
rather than "loading; retry shortly", which a client retries forever. Each
request still clears the previous placement result and triggers a fresh
attempt, so a model that fits once room is made is not refused on a stale
answer; the router waits for that attempt, not for its full activation
window.
Metrics now cover model health (desired, ready at 0 for what should run,
consecutive start failures, parked, unplaceable), engine installs per node,
leadership, and certificate expiry read from disk at each scrape, with alerts
for each. `ctl status` counts a model against what it should have now, so an
on-demand model that was asked for and fitted nowhere no longer reads as
satisfied.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Engines listen on every address their node has, with no key. Checked on the
live cluster: from a neighbouring node, a request straight to an engine's
port was answered — past the gateway's authentication, quotas and
accounting, by anyone who could reach the node.
Each replica's engine is now started with a key and refuses requests without
it. The key is derived from the replica id with an HMAC over a secret kept
in the store, rather than stored per replica: the router has to send the
right key to replicas this process never started — adopted after a restart,
or placed by the other control plane — and a function of a shared secret
gives that without a table to keep in step. A key taken from one node opens
that replica only. The secret is made on first use and copied with the rest
of the store when moving to PostgreSQL; a new one would lock the router out
of every running engine at the moment of the move.
The key reaches llama.cpp and vLLM through their environment
(LLAMA_API_KEY, VLLM_API_KEY) rather than their command line, which every
user on the node can read; SGLang only takes it as an argument. The agent's
readiness check, the router's proxy and its metrics scrape send it, the
client's own Authorization still never reaches the engine, and `ctl bench`
fetches the key of the replica it is pointed at from the admin API.
Engines started before this keep running without a key until their replica
is next started. With no store there is nowhere to keep the secret, and
engines start without one, with a warning.
Also formats two files that went in unformatted.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
doctor checks a node alone, and every check can pass while two nodes still
cannot reach each other. ctl fabric-test A B asks B's agent for an
ib_write_bw server and A's for the client, adapter by adapter as NCCL pairs
them, one rail at a time, and grades each against its link's rate. It goes
through the agents' streams, so there is no SSH between nodes; the agent
installs perftest the first time, and names the RoCE v2 GID itself.
The result is read from perftest's table by its heading, not a column
number. Agents now list the requests they answer beyond the usual
(Register.features), and the control plane refuses a test on one too old to
answer instead of waiting minutes for it. The forwarding dispatcher carries
the new capability, and its test says so: a missing forward would have made
the command quietly unavailable in production.
Not yet run over real InfiniBand or RoCE; this cluster has neither.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Seeing what a node did meant ssh through a jump host, to a machine whose
address nobody remembers, to read a journal only root may open. When an
engine install failed on a new node today, the reason was one line — and it
was four steps away from the page that said the install had failed.
Each node's row on the cluster page now links to its logs: the agent's own,
and every engine running there, chosen from the list the node itself sends.
The page can follow, refreshing every five seconds, and is on both the admin
listener and /admin. `llmgw ctl logs NODE [--source SHARD] [-n N]` reads the
same.
They travel over the stream the agent already holds, so they need nothing
the control plane does not already have: a request goes down, the lines come
back, and a reply that does not come in ten seconds is an agent too old to
answer. The agent keeps its last 2000 lines in memory as well as writing them
wherever it did before; the engines' last few hundred lines were already kept
and sent only when one crashed.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The first thing HA needs that a shared store does not give. Two control planes
against one store both reconcile: both see a model wanting two replicas and
none running, and both place. The model ends up with twice what it asked for,
each replica holding GPUs the other's inventory believes are free — and nothing
crashes to say so.
A lease in a table rather than a lock held by a connection. PostgreSQL has
advisory locks and SQLite has nothing like them, and this store keeps one set
of queries for both; a row with an expiry works everywhere, and it survives a
leader being powered off rather than asked to stop, which is the case that
matters. Taking and renewing are the same statement, so the two cannot disagree
about what a held lease looks like, and it is one statement, so two processes
racing cannot both believe they won.
The owner id is random rather than the hostname. A hostname would let a
restarted process match its own stale lease and believe it had been leading all
along — including through the gap that made it restart, during which somebody
else may have taken over and started placing.
A store that cannot be reached is not evidence that somebody else is leading,
so an unreachable store keeps the claim. Two leaders is worse than one, but no
leader means nothing is placed and nothing is evicted, and a brief database
blip should not stop a cluster healing itself.
Unset means leader. Every deployment that existed before this, every
single-control-plane install and every test acts unconditionally, with nothing
to configure.
The same seven assertions run against both backends, which is how the test
harness was caught inheriting a lease between subtests — it passed on SQLite,
where each subtest gets a fresh file, and failed only on PostgreSQL.
What this does not yet do is move agents. A standby holding an agent's stream
is one the leader cannot command, so failover is degraded rather than clean
until agents know more than one address.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Two halves of the same problem: nothing stopped a model being placed on a node
that could not hold it, and nothing ever removed the weights of a model that
had been deleted from the catalog.
Placement now has a disk check, shaped exactly like the memory one it sits
beside, with one addition: it is skipped on a node that already holds the
weights. A node holding the file needs no room for it, and without that the
node best placed to run a model — because it already has it — is the one most
likely to be rejected, since holding the file is what used the space.
Reclaiming is the dangerous half, so the whole design is about what it will
not do. llmgw deletes only files it fetched itself, and that is a calculation
rather than a heuristic: the cache name is a hash of the source, so the set of
wanted files can be computed from the catalog exactly. Anything without that
name was put there by an operator — scp, a shared mount, a copy by hand — and
is none of this system's business however unreferenced it looks. A model that
took a week to copy must not be removable by a command someone ran to tidy up.
The node checks every path again before unlinking it, and that duplication is
the point. The control plane knows the catalog; the node knows its disk. A
control plane that is wrong, newer than the agent, or tampered with must not be
able to talk a node into deleting something, so the agent refuses anything
outside its own cache directory or without the cache name — including a path
that tries to traverse out of it.
Listing is the default and deleting takes --collect. The summary says "asked to
remove", because each node decides, and one that refuses a path it cannot prove
it fetched is this working rather than failing.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The router published twenty-two metric families about requests, endpoints,
admission and scoring — everything about traffic and nothing about the
machines carrying it. There was no way to graph GPU usage, watch a node go
unhealthy, see a fabric fill up, or answer "was the cluster full at the
time", which is the first question anyone asks after an incident. The
agents exported nothing at all.
Exported by the control plane rather than by each agent. It already holds
every number: hardware arrives at registration and state on each heartbeat,
so this is a map read away. One scrape target for a fleet of hundreds, no
Prometheus configuration per node, no firewall rule per machine, and
nothing new running on the nodes. On the same /metrics as the traffic
figures, because correlating "the cluster was full" with "requests were
queueing" should not need two systems.
A collector rather than gauges kept current by a ticker: the numbers are
derived from state that already exists, so computing them when asked is
simpler than maintaining a copy that can drift, and nothing runs between
scrapes.
GPU utilisation is new, and is the number people actually watch. Memory
used says a model is loaded; this says whether it is doing anything, and a
fleet at ninety percent occupancy and two percent utilisation is a fleet
holding models nobody is asking for.
Adding it caught two things. utilization.gpu goes last in the nvidia-smi
query and is optional, because a driver too old to know it returns a
shorter row — requiring it made an llmgw upgrade silently stop discovering
GPUs on an older node, which is far worse than not knowing how busy a card
is. And I gave the new protobuf field number 14, which ecc_errors already
had; buf lint is why that was a moment rather than a corrupted wire format.
Absent is kept distinct from zero throughout, for the same reason as
everywhere else here: an unreported utilisation shown as a flat zero reads
as an idle fleet rather than a missing measurement, and a replica nothing
has reported would be the most idle thing in the cluster.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Adding a model was an operator procedure with downtime: edit a file on the
control plane, restart it, wait out the settling window, watch replicas
churn while adoption catches up. I did that eight times in a day, and one
of the restarts cost two minutes of outage while stalled replicas timed
out. None of it was necessary. The reconciler already compares desired
against actual on every pass, so a new catalog is simply a different
desired.
A Catalog stays immutable — New is the only way to build one and it refuses
an invalid one — and what changes is which immutable catalog is current.
Readers take the pointer and finish their operation against it, so a swap
mid-request is invisible rather than a race. The router is told rather than
asked, the same shape as the endpoint list and for the same reason: a push
happens once per change, a pull would happen once per request.
The router still seeds from configuration as well as being pushed to. A
router with a static endpoint list and no control plane is a supported mode
and has nothing to push it a catalog; without seeding it would resolve no
model at all.
Which turned up a bug that was latent while catalogs were fixed.
reconcileModel only ever visits models that are in the catalog, so a
replica whose model had been removed was visited by nothing: it kept its
GPUs, kept counting, and kept running an engine for a name no client could
ask for. The endpoint publisher skipped such a replica with the comment
"the reconciler will tear it down", which was not true. It is now, and the
comment says what actually happens instead of what was assumed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Two things a cluster with many nodes and many GPUs needs that were missing,
and they are the same thing seen from two distances.
Idle tracking. The control plane knew what was placed and whether it was
healthy, never whether anyone wanted it — so on a fleet with more models
than cards there was no way to answer the first question anyone asks, which
is what can I stop. The router is the only thing that knows, so it now
reports it: a timestamp written in Acquire, where a request already meets an
endpoint under a lock already held, and collected once a tick as a map. That
shape is the point. Per-request notification would be a firehose carrying
one bit — the newest timestamp wins — and this costs the same on four
replicas as on four hundred.
It travels through an interface in the direction opposite EndpointSink, for
the same reason that one exists: the control plane decides what should run,
the router knows whether anyone wants it, and neither should need to be in
the other's process to say so.
Unknown is kept distinct from idle throughout. A router that has just
restarted knows nothing, and a zero timestamp would be the oldest idle
replica in every comparison — so one restart would nominate an entire fleet
for eviction. Absent from the map, "-" in the table, negative in the JSON,
and IdleFor returns a second value rather than a duration anything could
act on by mistake.
ctl status. A per-node listing is right for four nodes and wrong for two
hundred: how full am I, what is broken, what is holding cards nobody is
using are aggregates, and reading them off a long table is a chore, which
means it gets skipped. Capacity per GPU class with usage bars, free whole
nodes per fabric — the number that decides whether a multi-node replica can
land, since free cards on the wrong fabric are not capacity for a model that
spans nodes — replica and model counts, models that cannot be placed, and
what has gone unasked-for. Derived from the same state the tables show, so
it cannot disagree with them.
Unhealthy GPUs count as neither used nor free, because counting them free is
how a cluster looks like it has room it does not.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
llama.cpp is what makes a cluster with no cards useful: it serves the same
OpenAI API on CPU, so the whole path — placement, supervision, scraping,
scheduling, admission — can be exercised end to end on ordinary machines.
Three things had to give way for it.
A profile may now omit its GPU class, and that is how it says it runs on the
CPU. Only engines that can actually serve without a card may omit it; for
vLLM an omitted class stays an error, because otherwise a typo becomes a
replica placed on a machine with no hardware and diagnosed minutes later
during model loading. Placement spreads CPU replicas instead of bin packing
them, since unlike GPUs there is no allocation to stop two of them sharing a
machine's cores.
Metric mappings are now per engine rather than per cluster. llama.cpp calls
the same three numbers requests_processing, requests_deferred and
kv_cache_usage_ratio, and a single global mapping would leave half a mixed
fleet unmeasurable. An empty mapping still defaults to vLLM's names; one that
names its own Running metric is taken as a complete statement, so the fields
it leaves blank stay blank rather than being filled with names that engine
cannot have.
The adapter refuses multi-node and tensor-parallel profiles. llama.cpp splits
across local GPUs by a different mechanism than this adapter drives, and
accepting either would produce a replica that starts and then serves from one
machine while the catalog believes it has several.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Ready replicas become router endpoints with no static configuration involved.
Endpoints that survive an update keep their live state — inflight counts,
breaker state, drain rate — so a replica that was already serving does not start
from cold because another one moved. The sink is an interface, so when the
router and control plane split into separate processes it becomes a watch stream
and nothing above it changes.
Phase 3's exit criterion now holds end to end. A node hosting a live replica is
killed while four clients stream continuously; the control plane marks it
unreachable, then lost, rebuilds the replica on a survivor and republishes.
Across three runs ~370 requests completed with zero failures and no replica was
left on the dead node. This proves the control logic, not the transport: the
agent fleet is simulated and the real path still needs the stream.
The race detector caught a genuine bug while this went in. startReplica read the
replica's fields after publishing it to the map, and scaleUp walked replica
pointers outside the lock; both raced with agent status reports, which arrive
from another goroutine and mutate exactly those fields. Against the simulated
fleet the reports came back fast enough to hit the window on every run. State is
snapshotted under the lock now and those functions work from values.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
