Asked whether Vulkan on both devices unlocks the split. It does not: the
per-tensor design aborts byte-identically, because the gate is not the driver
mix but `--override-tensor` — `has_tensor_overrides()` is a required term of
the scheduler's `pipeline_parallel` flag (llama-context.cpp:433, the only place
in the tree that reads it), and that flag sets `n_copies` in
`ggml_backend_sched_new`. With `n_copies == 1` a pre-allocated weight on a
non-primary device has no backend and the graph aborts.
Chasing it exposed a harness artifact that invalidated the earlier conclusion:
llama-bench splits a device LIST with `/` and SWEEP values with `,`, so
`-dev CUDA0,Vulkan1` was two single-device runs all along (and `-ts 0.85,0.15`
two sweep values, i.e. device 0 takes everything — which is where the bogus OOM
came from). With the correct syntax two-device splits work in both flavours:
pure Vulkan -dev Vulkan1/Vulkan0 -ts 0.85/0.15 -> 27.55 tok/s @64k (12.8 GiB on the 3090)
cross-vendor -dev CUDA0/Vulkan1 -ts 0.15/0.85 -> 27.23 tok/s @64k, pp512 268.5 (15.0 GiB)
iGPU alone 25.77 tok/s @64k, pp512 231.8
+7 % decode, +16 % prefill — real but not a regime change, and it costs the
3090 ~15 GiB, which this fleet spends on its only vision endpoint. The design's
premise still fails: whole layers carry their experts (1.62 GiB each), the 12
attention layers are spread every 4th, and the 32.8 GiB PLE must stay on the
main device, so KV cannot be consolidated on the 3090.
Also records the evening's incident (a dead rootless podman pause process took
all container management down; `podman system migrate` panics in 5.7.0; reboot
repaired it) and the netavark host-address lesson from the hermes 502.
hermes-webui is Network=host, so 8787 is a bare host-netns bind, and netavark
DNATs only *published* ports. From Caddy's bridge netns 192.168.0.143:8787 and
10.89.1.1:8787 are both refused, while host.containers.internal:8787
(169.254.1.2) answers 200 — measured live 2026-09-16 from inside the Caddy
container. The dashboard's published 9119 answers on that same LAN IP, which is
what made the old upstream look sound.
The vhost comment asserted "192.168.0.143:8787 is open" from that bridge; it is
not, so the claim is replaced with the measurement, and the hermes-webui quadlet
comment carries the same correction. The seam guarding this invariant accepted
any numeric upstream, so the LAN IP passed it; it now pins the measured-working
form and fails on the LAN IP (teeth verified by substitution).
The design we sketched (trunk + KV on the 3090, MoE experts on the iGPU) is not
expressible in mainline llama.cpp. Per-tensor cross-backend placement aborts in
the scheduler -- "pre-allocated tensor (blk.0.ffn_down_exps.weight) in a buffer
(Vulkan1) that cannot run the operation (NONE)" -- and both per-layer
cross-backend forms fail at load. CUDA + Vulkan cannot serve one model; the CPU
is the only fallback side that works, which is exactly why -ot exps=CPU runs.
What the same build does instead is the surprise: iGPU-only (Vulkan/RADV, PLE in
page cache) gives 32.64 tok/s shallow, 25.77 tok/s at 64k depth and 456.7 tok/s
prefill, with the 3090 idle at 0 %. That matches halogen's deployed ~25 tok/s --
but commits 116 of the host's 121 GiB against halogen's 95.4 held with 14.7
spare, so it is not deployable beside the co-tenants. Throughput parity, memory
bankruptcy: the fleet's two-endpoint split stands, now on evidence rather than
preference.
New page LLM/x4-pair-experiment.md carries the tensor inventory, the results
table, the two scheduler limits, reproduction commands, and the traps found on
the way. Both index pages also stop calling halogen retired -- the ik swap's
stale claims were still sitting in them.
The fleet now runs draft-dflash,ngram-mod, so the rule needs to be written down
where the next agent will look: draftless types (ngram family) are added to the
spec configuration unconditionally, while draft-mtp/dflash/dspark/eagle3 all
require a draft CONTEXT -- therefore one context-backed drafter plus any number
of ngram types is legal, and two context-backed ones never are ('draft-mtp,draft-dflash'
dies at load with 'model doesn't contain MTP layers'). Plus the operational
trap: ngram-mod prints no startup banner, so a working chain looks unengaged in
the journal.
The port gate was written for the cutover, when the vLLM proxy held 7770. Once
frank-medium itself owns the port, a re-deploy waits 90s for a socket that only
its own predecessor is holding and fails: 'FAIL: port 7770 is still held after
90s', with the old llama-server still serving scrapes. Stop the unit first, then
sweep a stray manual llama-server (the shape a benchmark harness leaves behind);
pkill -x, never -f, so the deploying shell cannot match itself.
ngram-mod is draftless (added to the spec config vector unconditionally, while
every model-backed drafter needs ctx_dft), so unlike MTP it can share the
process with DFlash2. Chained at the fleet's 48/64 window, over three shapes:
prose 1k/512 dflash 43.77 | ngram 17.72 (0 drafts) | chain 43.65
echo 1.5k/400 dflash 76.82 | ngram 102.50 | chain 200.58
structured/400 dflash 91.08 | ngram 17.80 (0 drafts) | chain 91.22
identity dflash 70.37 | ngram 17.77 | chain 70.40
On prose the chain is byte-identical to the drafter alone (same 119 drafts/49
accepts) -- that is what makes it a safe default -- and on prompt-echo traffic
it is 2.6x. VRAM is unchanged: the matcher is a 64 MiB host-RAM hash table.
The window sweep says keep the long one: 48/64 beats 16/32 (161.0) and 8/16
(113.8) on echo, and shortening it costs prose (43.9 -> 36.0) because the
matcher then proposes on ordinary fragments and the rejections still have to be
verified. That is the 'ngram hurts' result this fleet hit twice before.
Two documentation traps recorded with it: ngram-mod prints NO startup banner in
this build (its drafts are only visible behaviourally), and the harness's
'structured' shape is a negative control because its lines differ.
Two real defects from patching a rewritten file by line number: the CUT that
removed a duplicate parse_configs also took write_env with it (NameError after
the first configuration, with the unit already stopped), and the restore step
lived after the loop, so a crash left frank-medium DOWN. The endpoint is the
reason the harness exists; it now restores itself in a finally and shouts if the
service does not come back healthy.
ngram-mod's match window (n_min/n_max) decides which traffic it can see at all
-- the fleet's 48/64 preset is tuned for long verbatim echoes and finds nothing
in a repetitive-but-short output. Config entries now take |KEY=VALUE overrides,
so the window can be swept in the same run as the configurations it applies to.
The first run asked for a verbatim repeat with thinking ON, so the model copied
the passage inside reasoning_content, content came back empty for all three
configurations, and the cross-configuration identity check compared '' to '' --
a pass that proves nothing. echo and structured now disable thinking via
chat_template_kwargs, the echo row carries a copy_ratio (share of answer lines
that really are passage lines), and an empty answer in any shape is a loud
warning rather than a silent pass.
'structured' is new and answers the case the fleet actually runs on: repetitive
OUTPUT with no prompt overlap (record lists, tables, config rewrites), which is
where an n-gram matcher should pay and an echo-shaped test cannot tell.
A relative serve-script path plus a cd into its own directory double-counts the
prefix, so every configuration failed to start ('scripts/serve-frank-medium.sh:
No such file or directory'). The safety net did its job -- the unit was restored
-- but the run measured nothing.
Unlike MTP, ngram-mod needs no draft context: common/speculative.cpp adds the
ngram types unconditionally while the model-backed drafters are gated on
ctx_dft != nullptr, so 'draft-dflash,ngram-mod' is structurally allowed (and the
config vector's fixed order -- ngram first, then the drafters -- makes CLI order
irrelevant).
The harness grows the two things this question needs: --configs, and an ECHO
shape. ngram-mod can only ever match text that is present in the prompt, so
measuring it on prose calls it useless and measuring DFlash on echo calls it
wasteful; both shapes now run, and the greedy text of each is compared across
configurations because speculation may not change answers.
Four configurations of the same binary, identical prompts at temperature 0, the
unit stopped between runs, on the deployed payload:
none 17.7 tok/s card1 26.07 GB
draft-mtp n=3 32.4 tok/s 58 % accepted card1 29.35 GB
draft-mtp n=7 26.7 tok/s 33 % accepted card1 29.65 GB
draft-dflash 7 43.5 tok/s 41 % accepted card1 29.42 GB
DFlash2 is worth 2.46x over autoregressive and 34 % over the best MTP, and all
four produced IDENTICAL greedy text, so the 'lossless verification' claim in the
runbook is now a check rather than an assumption. MTP is not the cheaper option
either: its head loads ~3.3 GB against the drafter's 2.06 GB file, and deeper
drafting loses (n=7 accepts a third of its proposals, n=3 more than half).
The two CANNOT be chained, and the failure is structural rather than tunable:
'spec-type draft-mtp,draft-dflash' dies at load with 'context type MTP requested
but model doesn't contain MTP layers', because a chain has ONE draft context and
MTP builds against whatever model that context holds -- here the DFlash2 file.
The serve script therefore passes --spec-draft-model in the dflash branch ONLY,
and the seam now asserts exactly that (the wiring whose violation is this
error), plus that the deployment stays on the measured winner.
scripts/bench-spec-ab.py makes the comparison reproducible: the benchmarking
policy wants the runtime's own harness for this, and the numbers above were
otherwise unreproducible one-offs.
The loader prints no per-device ledger at this verbosity, so the claim is
measured rather than read: +131072 tokens of context added 4.83 GB to one card
and 4.84 GB to the other (both carry their own attention layers' KV), and
neither card's total usage could hold the 31.46 GB file. Worth writing down
because the retired shape WAS two full copies — that is what a replica is, and
it is why the 29.8 GB FP8 checkpoint could never run there.
The LLM index still opened with vllm-radiance as frank-medium's runtime and
drew the replica pair in its topology diagram; the cooperation page's thesis
(collectives are impossible, so the answer is hardware) now has to name the
path the fleet actually took: llama.cpp's layer split, which launches no RCCL
kernel on either card. That reframes the M.2 riser as a means to TENSOR
parallelism rather than a means to using both cards at all.
Two families of ~30k-token turns ran against the deployed endpoint: 24.8 s and
51.4 s for the first turn (the second waits for the slot), 30.7 s and 8.0 s for
the second. Against the prefill table that is ~25 s of cold prefill plus the
wait, so the honest reading is that ONE slot is now the limiting resource for
concurrent agent loops — the trade this deployment makes for giving a single
conversation the whole 262144-token window. --parallel splits the window; it
does not add throughput.
The harness header still described the retired two-replica proxy ('the deployed
routing pins such a conversation to one replica', vLLM's prefix-cache key). It
now says what is true: one endpoint, one slot, concurrent families queue, and
the old routing numbers are history in the runbook.
Context: 262144 was measured before it was deployed -- 24.72 GB on GPU 0 and
29.14 GB on GPU 1 of 34.21 GB each, ~5 GB of headroom on the busier card, and a
140,069-token prompt prefilled through the single slot in 134 s to prove the
window is real. The 114688 wall was the vLLM predecessor's arithmetic, not this
runtime's.
Throughput, measured through the service: 1.0-1.4k tok/s prefill from 4k to
140k, 43.8 tok/s decode on a 512-token generation, 41-71 % DFlash2 acceptance,
and the prefix cache re-prefilling 4 of 998 tokens on a repeated prompt.
The signals move with the runtime, so the fleet's rules and dashboard are
re-cut rather than renamed: vllm: series do not exist any more, and two of the
old rules watched things llama.cpp does not export at all (KV-pool fill, p95
latency). Those are retired with the reason written down, and LlmSpecDecodeOff
takes one of the slots -- a lost drafter is the failure no client can see. The
observability seam now fails if any rule names a series the endpoint does not
export, which is what would have caught the stale vllm: expressions.
omp models.yml (outside the repo) is updated in the same pass: contextWindow
262144, and the full reasoning-effort map, because Qwen3.8's template answers
500 on 'high' -- verified against the live endpoint, low/medium/xhigh answer
200. Subagents run at high.
The runbook for the new deployment, plus the sweep of everything that still
described the vllm-radiance replica pair: the vault index and its ACTIVE
marker, the superseded banner on the radiance page, the pointer from the
ROCmFPX page, the homepage card, the Grafana dashboard description and the
observability page's scrape table.
The new page carries the two facts a future agent needs and the flags do not
say: llama.cpp's layer split is what lets BOTH cards serve ONE model without
any collective (vLLM's PP handoff is an RCCL send/recv, which is the primitive
this host cannot dispatch), and the rocBLAS Tensile-path trap that makes a
healthy-looking build fail on the first request.
'cp -r SRC DST/' with a missing DST creates DST as a copy of SRC and drops a
directory level, so rocblas/library/* landed as lib/rocblas/* and hipblaslt's
arch subdirectories landed one level high. Nothing complains at load time: the
engine reads 29 GiB of weights, answers /health, and dies on the first request
with
rocBLAS error: Cannot read .../lib/rocblas/library/TensileLibrary.dat:
No such file or directory for GPU arch : gfx1201
Declare the destination and copy contents, both trees.
The vllm-radiance replica pair could not use both cards for one model: TP=2
needs a cross-card collective and the second R9700 cannot dispatch one. The
same wall blocks vLLM's pipeline parallelism, whose stage handoff is an RCCL
send/recv of the hidden state (distributed/parallel_state.py) -- a fact worth
writing down, because the obvious reading of the ask ('use PP to dodge the
atomics problem') is exactly backwards.
llama.cpp's layer split needs no collective at all: layers are distributed and
the boundary activation is a per-backend copy, once per forward pass. One
process, both cards, one model. The runtime is mainline llama.cpp at fb27a525
(the ref is pinned by DFlash2 AND qwen35 MTP; the ROCmFPX fork this host had
before has neither), built for gfx1201 against the private ROCm 7.14 toolchain
the earlier build already used, and bundled next to the binary.
Server: ajgazin Swift-Qwen3.8-27B-Uncensored UD-Q8_K_XL with the model's MTP
head in the file, plus the z-lab DFlash2 Q8_0 block-diffusion drafter at depth
7 (its block is 8, so 7 is the ceiling llama.cpp allows). Text only: vision
stays on frank-small.
Retired into the trash rather than deleted: the three vLLM quadlets, the
radiance entrypoint/healthcheck, the routing proxy and its tests, and the two
bench scripts that drove 'vllm bench serve' inside the retired image. The
deploy handler now stops them, waits for 7770 to be released, installs the
ROOT unit and refuses to start anything when the runtime or payload is
missing. The planner maps the scripts/ trio to frank-medium only, and the
scoping seam asserts the retired service dir is gone rather than merely
unmapped.
halogen-flash-server is ACTIVE again with its 2026-09-16 measured ledger (budget
33.5 GiB against a need of 27.2 — no clamp — and 14.7 GiB left for everything
else), ik_llama.cpp is retired with its measurements kept as the record of why,
and the new frank-small runbook covers the -mtp file, the 64 KiB/token KV
arithmetic, the MTP contract, and the reasoning_effort trap that cost me every
subagent on the endpoint.
AGENTS.md: topology, Active Services, runbook index, and three new llama.cpp
lessons (MTP is a property of the file; --jinja is what `tools` needs; a model's
template can reject the harness's thinking level). Three pre-existing defects
found while editing it: a truncated sentence mid-file, a harness path that does
not exist on either server, and the benchmarking stop/mask recipe still calling
frank-large a root unit.
Truthfulness notes worth keeping: the co-tenant stall NOTE prints again, so the
claim that it "no longer prints" is gone and the three observed block counts are
recorded instead of a guessed threshold; and the Observability page says out
loud why frank-large has no scrape job rather than leaving a silent absence.
The runtime for the next frank-medium shape is mainline llama.cpp at
fb27a525, built for gfx1201 against the ROCm 7.14 nightly the ROCmFPX build
already used on this host, with the runtime bundled next to the binary.
Two things that pin the ref: DFlash2 (ggml-org/llama.cpp#27342) and qwen35
MTP. Neither is in the ROCmFPX fork's tip (c49ebdb, 2026-08-22), and the
DFlash2 drafter cannot run without the first.
The fetch script pulls the target at UD-Q8_K_XL plus the DFlash2 draft at
Q8_0, resumable and byte-verified -- a truncated 31 GB GGUF is otherwise a
load-time crash.
A redeploy of a running frank-large holds 7770 itself, so waiting on the port
before stopping the unit would have made every subsequent deploy fail. Stop the
user unit first (a no-op on a first cutover), then wait, then install and start.
The 2026-09-16 halogen restore failed exactly here. The ik root unit stopped,
systemd logged "Deactivated successfully", pasta tried to bind 7770 about a
second later, reported "Address already in use", and the quadlet exited 126.
Only Restart=on-failure recovered it -- minutes later, which is not a deploy.
The handler now waits for every socket on the port to clear (listener, accepted
and TIME_WAIT) before installing and starting the quadlet, and fails loudly with
the socket table attached if it never does. The retirement also reports when the
unit file was already gone, because this host's `disable` deletes it.
frank-large has no /metrics to scrape any more, and scraping it would be red by
construction: the halogen engine publishes /health, /cache and /v1/* only. That
job is removed with the reason written down (the remedy is the blackbox
exporter this fleet already wants for the comfyui/services probes), and
frank-small's llama.cpp endpoint takes the slot at 192.168.0.143:7771.
Homepage: frank-large's card still described the ik runtime, and the new
endpoint had no card at all. Both fixed, with a loopback probe -- the checker
runs on Dankclaw, which is also what makes the card green-able.
Dashboard: the endpoint picker was sourced from a vLLM-only metric, so a
llama.cpp job could never be selected, and the prose named frank-large as the
llama.cpp endpoint. The picker now matches both runtimes' series and the
llama.cpp targets carry the job filter every other panel already used.
The seams assert the new contract including the ABSENCE of a 7770 target, so a
well-meaning edit cannot re-add a permanently-red probe.
Dankclaw's endpoints are re-split. frank-large is the halogen-flash-server
quadlet again -- the 2026-09-13 unit file was never touched by the two-day
ik_llama.cpp interlude, so the runtime was still in the repo and its image is
still in local storage -- retuned to the shape asked for: HALOGEN_CTX 200192
with a 400,384-position pool and two slots, exactly two 200k conversations.
frank-small is new: mainline llama.cpp on the RTX 3090 serving
ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF at IQ3_S, using the -mtp build (it carries
blk.64.nextn.*, which is what lets --spec-type draft-mtp run on the main
model's own weights instead of a second 12 GiB copy) plus the BF16 vision
projector, as one 131,072-token slot with f16 KV on port 7771.
The host is shared now, and that is what the halogen retune pays for:
HALOGEN_MAX_TOK 24576 (arena 20.6 -> 16.4 GiB, ~4-5% prefill) and the vision
tower dropped (1.97 GiB) hand ~6 GiB back to the host, and the reserve rises to
20 to keep the fit check honest. Vision moved wholesale to frank-small.
Retired to scripts/.AGENTTRASH/: the ik trio (env, service, serve script), the
Qwen-Sharp template only that runtime read, and four benchmark harnesses bound
to retired runtimes. Because retiring a file makes any test that greps it pass
by absence, the seams assert the archive exists AND the live path does not.
New page: LLM/Dual-R9700-Cooperation.md — registered in the vault index, with
the measured evidence rather than the theory.
**The answer.** Card #2 sits behind the B650 chipset, whose bridges advertise
`AtomicOpsCap: Routing-` (read-only), so ROCr refuses to fill the hostcall
buffer RCCL's kernels declare and no RCCL op can dispatch on it. Two paths fix
it: an M.2→PCIe x4 riser into the free, CPU-attached M.2_2 slot (~$30; per
ASUS's spec page M.2_1 and M.2_2 are CPU-attached, PCIEX16_2 is chipset in x1
mode, and there is no x8/x8 bifurcation on this board), or a board whose CPU x16
bifurcates to x8/x8. Bandwidth is not the constraint: ≈10 MB per token-step ≈
1.4 ms at Gen4 x4.
**New measured results.** A bare 2-rank torch.distributed run: rank 0 (CPU-lane
card) passes broadcast/all_reduce/barrier; rank 1 (chipset card) fails all
three, and `send`/`recv` fails too — which kills pipeline parallelism as an
escape, since vLLM's PP handoff is exactly that primitive.
**Correction.** I previously said no software path exists. That overshot. One is
open and unverified: recompiling RCCL so its kernels stop declaring a hostcall
buffer (`-mprintf-kind=buffered` is the claim). Counter-evidence recorded: RCCL's
device source contains no live printf/assert/__assertfail, so the trigger is
unexplained, and the claim traces to search summaries, not a primary source. The
page carries the two-command test that would settle it. Also corrected: 0.9.3 is
radiance's version, not vLLM's — the base is vLLM 0.27.1 / torch 2.11.0 / ROCm
7.14, and its shipped RCCL exposes no `ncclDevComm*`, so torch's symmetric-memory
path cannot run.
**Rule added to the benchmarking policy**: it covers ANY GPU experiment, not just
benchmarks — including a bare 2-rank probe. Running one against the live service
on 2026-09-16 preceded the host going hard down (no ARP, no ICMP, power cycle
required); a hung AQL dispatch wedging the driver is the plausible mechanism.
The policy also now says how to tell off from wedged: `ip neigh` from a peer.
Retiring dankclaw/compose.yaml and adding alert rules invalidated claims across
the repo, including in the rule file agents read first. Clean cutover includes
the prose.
**Tests that had gone hollow.** Four assertions grepped dankclaw/compose.yaml
for a service name. With the file gone they passed by *absence* — a gate that
can no longer fail. Each now asserts the real invariant ("no dankclaw compose
file competes with the quadlet units") and keeps a live check for a re-added
entry, so it still fails when one appears: verified by recreating a compose file
with a comfyui service and watching three tests go red, then removing it.
**AGENTS.md, which was wrong about ComfyUI.** It said ComfyUI runs CPU-only
("no ROCm wheels for gfx1151, CUDA images wrong"). That was true in the
compose era and is false now: the restored unit runs with /dev/kfd + /dev/dri and
must NOT pass --cpu. An agent trusting the rule file would have re-broken it.
Also corrected: the observability paragraph ("RUNNING BUT UNCONFIGURED, no
datasource, no rules" — all three have been configured since 2026-09-16), the
Active runbook row that still named halogen for frank-large, halogen's absence
from the Retired table, the benchmark policy naming halogen's harness, and the
seam list.
**Unit and page claims.** prometheus.container said "no alerting rules anywhere
in this stack"; node-exporter.container said Dankserver runs from its compose
stack; Prometheus.md documented three jobs as working (they scrape HTML: they
can never report up) and said no dashboard existed; index.md and the three
authentik service pages pointed at the retired file; Authentik.md's upgrade
instruction told you to bump an image tag in it.
One bad edit of my own on the way (a stale line number clobbered a See Also
entry in Prometheus.md) — caught by reading back, fixed, and the vault link
checker now proves the cross-references hold.
The vault is the project's source of truth, so a link to a page that does not
exist is a promise the repo cannot keep. Three were dangling:
- Dankclaw/Gitea.md pointed twice at dankclaw/compose.yaml — **which this
session retired**. The clean cutover obligation from my own change: the page
now describes the unit that publishes 3000 and the canonical Caddy unit, and
records why the compose file is gone.
- Dankclaw/Hermes-Dashboard.md linked [[LAN DNS]]; the page is [[LAN-DNS]].
- LLM/frank-medium-radiance-mxfp4.md linked [[Observability on Dankclaw]]; the
page is [[Observability]].
scripts/check-vault-links.py + a seam keep it that way. The checker models
Obsidian's rules, which took three corrections to get right — and each was a
false positive that would have sent me editing healthy docs:
1. Wikilinks are case-insensitive. A case-sensitive matcher reported 45
breaks; 39 were `[[qBittorrent]]` vs qbittorrent.md.
2. `[[folder/Note]]` resolves by path suffix, not by stem.
3. `%20` in a markdown link is URL-encoding, not a filename.
Validating the linter against the resolver it models is the prerequisite for
trusting any of its output — the same discipline that caught a bad assertion of
my own earlier tonight.
The board showed mariadb, valkey and Postgres as "down". None of them is down:
the checker runs on Dankclaw and can reach that host's loopback, its container
names, and published LAN addresses — while all three cards probe
container-internal database ports (192.168.0.235:3306, 127.0.0.1:6379/:5432).
They were red by construction, which is how a status board stops being read.
Publishing a database port on the LAN to make a badge green is not a trade this
fleet makes, so they are wip with the reason in their description, and the apps
that front them carry the status. Gluetun (WireGuard, UDP) was the pre-existing
example of the same call.
The rule is now written down in the Homepage doc: a card must be probeable from
the monitor's host or be wip. That is the same lesson as the Prometheus jobs
that scrape HTML pages and can never report up — an always-red signal teaches
the operator to ignore the surface it lives on.
Two gaps behind the same near-miss.
1. **No spec covered hermes.franksdank.cloud** — the one guarded vhost whose
upstream is a host address (`reverse_proxy 192.168.0.143:8787`) rather than a
container name, because its service is host-networked. That is how a
loopback-only bind could take the vhost down with a 502 and leave every gate
green. tests/e2e/specs/17-hermes-vhost.spec.ts asserts what a dead upstream
violates: no 5xx through the guard chain, no Caddy error page, and no app
bytes to an unauthenticated raw client. It does not assert SSO-silence, since
this app has its own password login.
2. **Specs are matched by number prefix, not globbed.** A spec file whose prefix
is in no project silently never runs — which is exactly what happened to the
new one until it was registered. The prefix is registered now, the comment in
playwright.config.ts says why, and tests/e2e_scaffold.bats fails if any
spec file matches no project.
The hermes-webui seam assertion is corrected too: it asserted the loopback bind
I wrongly imposed, and now asserts the real contract — the unit's bind must serve
the address the Caddyfile fronts it on.
I tightened HERMES_WEBUI_HOST to 127.0.0.1 on the reasoning that the compose
service had published loopback and the unit had widened it. That reasoning
missed the consumer: this unit is host-networked (it shares the CDP daemon's
loopback), so Caddy cannot reach it by container name and fronts the guarded
vhost hermes.franksdank.cloud with `reverse_proxy 192.168.0.143:8787` — recorded
in dankclaw/Caddyfile with the same explanation. With the app on loopback that
upstream is dead, so the vhost broke.
The wildcard bind is what serves BOTH consumers: Caddy on the LAN address and
the phone's SSH tunnel on the host's localhost. Controls are the auth chain in
front of the vhost plus the webui password. The unit comment now states this
instead of the incorrect 'as before' claim, and the near-miss is worth the
paragraph: the e2e gate stayed green because no spec covers that vhost.
Two things, found while checking whether the hermes trio still needed migrating.
1. **The phone-facing webui was exposed to the LAN.** Its unit runs with
Network=host, so there is no Podman port forwarder and the app's own bind is
the exposure. The compose service it replaced published
`127.0.0.1:8787:8787` (loopback); the unit bound `HERMES_WEBUI_HOST=0.0.0.0`,
and the LAN address answered 200 — password-gated, but on every interface,
which is not the documented "SSH tunnel is the sole ingress" posture. It now
binds 127.0.0.1, which serves the tunnel just as well (sshd dials
localhost:8787 from the host). The unit comment claiming "LAN-facing as
before" was wrong about "before", and is corrected.
2. **Dankclaw's compose.yaml is retired.** Auditing ownership showed every
running container on that host reports a PODMAN_SYSTEMD_UNIT label — the
hermes trio's units have owned their containers since 2026-09-12 — so the
three service blocks left in compose.yaml were dead weight that could only
create a duplicate container fighting a unit for a name. deploy_hermes was
already fully unit-based. The file is staged in .AGENTTRASH/; the planner
plans nothing for it, an unknown service on that target now errors instead of
silently no-oping through `podman-compose`, and DANKCLAW_COMPOSE_SERVICES is
empty.
The hermes-webui seam now reads the unit — it read compose, which is exactly how
the bind drifted while the test kept passing against a file nobody ran.
ComfyUI's page described a compose service and a systemd unit that shelled out
to `podman compose up`. It now documents the Quadlet, the keep-groups + udev
mechanism (verified inside the running container: torch sees Radeon 8060S), the
9,146-restart incident that forced the move, and the trap that caused it — a
stale container outlives the compose edit that should have replaced it.
The observability page gains the audit result: four jobs read "down" and three
of them always will, because they scrape HTML pages or a /metrics that does not
exist. The page names each error, says plainly that none is an outage, and
refuses the tempting fix (deleting the job makes an absent scrape look like a
healthy one) in favour of the remedy it already recommended: a blackbox exporter
in probe mode, and Gitea's metrics enabled or the job repointed.
ComfyUI had failed 9,146 times and had been down ~7 hours when this was found:
its container was created 2026-09-15 14:12, BEFORE the GPU declarations were
added to the compose service, and `restart: unless-stopped` kept restarting
that stale container — no /dev/kfd, no /dev/dri, no keep-groups — so it died
every time on `RuntimeError: No HIP GPUs are available`. Its "unit" was a
hand-written ~/.config/systemd/user/comfyui.service that shelled out to
`podman compose up -d`, which is not repo-tracked state and collides by name
with the Quadlet the generator produces.
- dankclaw/comfyui/quadlet/comfyui.container: image comfyui-rocm:gfx1151
(built from Containerfile.comfyui), /dev/kfd + /dev/dri, keep-groups,
the four comfyui-data volumes, loopback-only publish (an unguarded service
is internal-only here; the homepage card probes 127.0.0.1:8188), Restart=
always, WantedBy=default.target.
- the compose service is deleted — one owner per container name.
- deploy_comfyui builds the image, installs the unit, retires the compose-era
unit into ~/.AGENTTRASH, removes the stale container so the unit's own can
take the name, and waits on :8188.
- the healthcheck is a file, because the podman 5.x v1 generator splits
HealthCmd on spaces and an inline `python3 -c "..."` becomes a quoting
puzzle that fails silently.
The comfyui seam now reads the unit instead of compose for devices, command and
groups (and resolves the Containerfile from the deployer's declaration, not
from a directory listing).
The page said "all five are rootless Quadlet units on Dankclaw", which stopped
being true when Dankserver's node-exporter was restored to a unit. It now names
both hosts and both deployments, records why the second one happened (the same
compose-only reboot argument, a day and a half of missing host metrics, and a
wildcard bind), and states the exposure rule for a two-listener fleet with the
`ss` evidence that the wildcard is gone.
Dankserver's node-exporter has been dead since a reboot: `podman ps -a` showed
it `Exited (143)` and Prometheus' `node-dankserver` job has been scraping
nothing. It ran only from `dankserver/compose.yaml`, which is the failure mode
this repository's boot-persistence rule exists for (rootless Podman does not
restart bare containers after the stop that shutdown performs), and the compose
form bound the wildcard address — the unauthenticated-metrics-API-on-every-
interface posture already removed on Dankclaw.
- dankserver/observability/quadlet/node-exporter.container — mirrors the
Dankclaw unit, listening on exactly 127.0.0.1:9100 (host-local probes) and
192.168.0.235:9100 (Prometheus on Dankclaw), with Restart=always and
WantedBy=default.target.
- the compose service is deleted: one owner per container name.
- deploy_observability now installs from "$TARGET/observability/quadlet"; it
hardcoded the Dankclaw path, so a Dankserver unit could not be installed by
the shared dispatcher at all.
- planner: "node-exporter" joins the Dankserver service set, with a path rule
for dankserver/observability/**; the dispatcher's allowlist gains
dankserver:node-exporter.
Seams cover the silent halves: both listen addresses present and the wildcard
absent, the compose service gone, the allowlist entry present, and the planner
mapping the path to the service.
Every measurement behind the frank-medium runbook was taken with scripts that
lived in /tmp on two hosts — reproducible only until the next reboot, and
invisible to anyone reading the repository. This moves the useful four into
scripts/, where this repo already keeps its bench harnesses:
bench-vllm-corpus.py code-shaped corpus from the repo's own sources, because
synthetic random prompts misreport draft acceptance
badly (2.9-3.6 tok/update against ~5.8 on code)
bench-vllm-serve.sh concurrency sweep; encodes the two traps that produced
silent garbage — the tokenizer argument, without which
the client dies resolving "frank-medium" on HF and
reports no numbers at all, and the --ignore-eos
semantics that must not be compared across
bench-turn-latency.py per-turn TTFT for an agent-loop shape; the instrument
that decided the routing policy, and the one that shows
cold-vs-warm prefill directly
bench-llm-soak.py sustained multi-family load; what found the capacity
wall (three 30k-token families, 75% cold re-prefills,
zero failures)
Each header carries its traps and the interpretation; the runbook's Benchmarks
section now points at them instead of describing methodology with no tooling.
Also adds the missing half of the script seam: scripts/*.py are compiled, not
just scripts/*.sh syntax-checked — a harness with a syntax error could otherwise
ship unnoticed until someone needed it during an incident.
Both python harnesses were smoke-tested against the live fleet before commit:
the corpus builder wrote 15 prompts, and the turn harness measured 12.5 s cold
against 1.4 s warm on the same prompt — the cache behaviour the runbook claims.
dankclaw/comfyui/Containerfile is referenced by nothing — compose builds from
dankclaw/Containerfile.comfyui, deploy-remote.sh copies that one, and the
ComfyUI doc names it — and its content is an older iteration (torch
2.9.1+rocm7.12.0rc1 against the deployed gfx1151 base). It also caused real
damage: the comfyui test's directory-listing selector picked non-comfyui files
and reported a red suite for sessions.
Staged in .AGENTTRASH/ rather than deleted, so it stays recoverable and visible
until Frank decides it can go for good. This is also why the earlier test fix
had to resolve the Containerfile from compose rather than from the filesystem:
the directory is not the contract.
This test failed on a clean tree for many sessions, and the working belief
became "test 73 is a known failure, ignore it" — which is exactly what a
permanently red suite teaches. The cause was a selector bug: `find
dankclaw -name 'Containerfile*' | head -1` returned whichever Containerfile the
directory listing happened to put first (dankclaw/alloy/Containerfile), so the
assertion read a file that never claimed to be ComfyUI's. It would equally have
passed on a wrong file.
The test now resolves the Containerfile the way the build does — build.context
joined with build.dockerfile, both read from dankclaw/compose.yaml — so it
checks the file the deploy actually builds (dankclaw/Containerfile.comfyui, the
one compose, deploy-remote.sh and the ComfyUI doc all name) and follows a future
rename instead of guessing.
Suite: ALL TESTS PASSED (310), no exceptions to explain away.
The card probing 192.168.0.235:7770 — frank-medium's endpoint since the
radiance cutover — was still called "frank-small", a service retired on
2026-09-15. The probe target was right; only the identity was wrong, which is
the worse half: a green card named after a service that no longer exists teaches
the wrong thing about the fleet. It is now frank-medium, described as what it
actually is (Qwen3.8-27B Uncensored MXFP4, vllm-radiance, two R9700s behind one
endpoint), and frank-large's description stops crediting the retired halogen
runtime for work ik_llama.cpp does.
The seam added to homepage_config.bats pins the cross-file contract that let
this drift: a card's probe target and its name must agree with the fleet, and a
retired service must not keep a card.
Both pages claimed there was no alerting; there are six rules now, so they say
which rules exist, that promtool and /api/v1/rules both confirm them, and — in
the same breath — that no rule reaches a person, because this stack has no
Alertmanager and no contact point. The dashboard's ALERTS panel is named as the
only place an alert becomes visible without opening Prometheus.
The scraped metrics have no rules over them, so every graph was read after the
fact by a human. This adds shared/prometheus/rules/llm-fleet.yml, loaded via
rule_files and installed by the same copier that ships prometheus.yml (a rule
file nothing loads evaluates nothing, and Prometheus reports no error for it):
LlmKvPoolSaturated kv pool > 90% for 10m — the 122,636-token wall
LlmRequestsQueueing waiting > 0 for 10m — MAXSEQS=4 reached
LlmPrefixCacheCollapse reuse < 20% for 15m with traffic — the thrash signature
LlmEndpointDown scrape target down 2m — both replicas gone
LlmLatencyHigh p95 end-to-end > 60s for 10m — what an operator feels
job:llm_prefix_cache_reuse:rate5m — recording rule shared by the alert and the
dashboard, so the two cannot drift
Every threshold is tied to something measured on this fleet, not to a guess,
and the file says plainly at the top that **nothing is notified**: there is no
Alertmanager here, so a firing alert reaches Prometheus' /alerts page, the
ALERTS metric and the dashboard panel added alongside it — and nowhere else.
Choosing how Frank actually gets paged is his call, not a missing config line.
Seam covers the silent failure mode: rules exist, the config references the
directory, the deployer installs it, and the dashboard graphs ALERTS.
The page said Grafana has zero datasources and zero dashboards; that is now
false, so it says what exists (Prometheus datasource, LLM fleet dashboard), how
it was verified (the API — a provisioned dashboard does not show up in the
legacy dashboard table on this Grafana version, which is a trap worth recording)
and what is still missing (nothing alerted on, Loki unread, correlation manual).
The section heading drops "UNCONFIGURED" because it no longer is, entirely.
Grafana has been running with zero datasources and zero dashboards, so the
metrics scraped last commit were only reachable through the Prometheus API.
This wires the plumbing and one dashboard, both declared in
shared/grafana/provisioning/ and installed to /opt/grafana/provisioning by the
deployer (the unit already mounted that directory read-only, so nothing about
the container changes):
- datasources/prometheus.yml — Prometheus at http://prometheus:9090, the
container name, not the loopback publish, with a fixed uid so dashboards
reference it without a lookup.
- dashboards/llm-fleet.json — KV pool fill, queue depth, prefix-cache reuse
rate, e2e latency p95, and decode/prefill throughput, with an endpoint
variable covering both frank-medium and frank-large. Every series name was
checked against the live /metrics endpoint: this fleet already shipped one
comment naming a metric that does not exist, and a dashboard full of those
would be worse.
Dashboards stay editable in the UI (allowUiUpdates) because the panels are a
starting point, but the file is the source of truth and the deployer
overwrites Grafana's copy on every deploy.
Planner and seams follow the existing pattern: shared/grafana/** selects the
grafana service, and the tests pin the container-name address plus the
provisioning contract so a rename cannot silently empty the dashboard.
The capacity work in this runbook was done with ad-hoc probes; the same signals
now have history in Prometheus, so the page says which series to watch and
admits the honest gap — this stack has no alerting, so a saturated pool shows up
on a graph after the fact rather than paging anyone.
vllm:prefix_cache_hit_rate does not exist in this build; the exported pair is
vllm:prefix_cache_hits_total / vllm:prefix_cache_queries_total, whose rates give
the hit rate. A comment naming a series nobody can query is worse than no
comment.
Neither LLM endpoint was scraped, and frank-medium exports 418 vllm: series —
including the ones this fleet argues about under load: kv_cache_usage_perc (how
full the per-card pool is), prefix_cache_hit_rate (whether pinned conversations
are being reused), num_requests_waiting (queue depth) and the request-latency
histograms. Everything diagnosed tonight by hand — 122,636 KV tokens per card,
thrash when three large contexts overlap, cold-vs-warm prefix reuse — is a
graph now, on the LAN addresses rather than the guarded vhosts.
frank-large comes along in the same change: llama.cpp exposes throughput
counters under llamacpp:, and watching one LLM endpoint without the other was
arbitrary.
The seam added to observability_quadlet.bats pins the cross-file contract that
breaks silently: if a scrape target drifts from the :7770 the units publish,
the graphs vanish with no error anywhere.
Two more hypotheses, both measured on the same three-family workload (one
~80k-token family plus two ~50k-token families, each turn re-sending its
document):
1. Long-prefix caching is fine. One card, idle engine, a ~50k-token prompt
sent three times: 22.3 s cold, then 1.4 s, 1.4 s — with a varying tail too
(22.4 / 1.41 / 1.41). So the engine's lifetime "Prefix cache hit rate: 29%"
is an average across every prompt that replica ever saw, not a verdict on
any one request, and the pinning policy is safe.
2. Serialising large requests does not fix the overload. Admission control
(at most one >16k-token request per card, small requests untouched) gave no
clear win on small samples — heavy 25.4 s -> 48.2 s median, mediums
34.7/45.2 s -> 22.7/22.6 s, cache-hit rates 14-33 % both ways — so it was
reverted like the placement change before it.
The arithmetic now closes: with three families on two cards, one card holds two
families and must carry 2 x 50k of cached prefix plus the 50k of the sequence
generating — 150k against 122,636 tokens. Those co-tenants evict each other
regardless of concurrency limits. The wall is capacity, and the runbook says so
with the numbers.
Frank