conjurer: pin the model Ollama actually has, and raise the AI timeout

Probing the real server (192.168.1.72:11434) after it was opened to the LAN:
/v1/models returns exactly one model, gemma4:e2b. Without pinning it the
bot would request the built-in default llama3.1:8b and every reply would
fail with "model not found", so set it explicitly on both bots.

Also raise CONJURER_AI_TIMEOUT_SECONDS to 240. Self-hosted generation is far
slower than a hosted API, particularly the first request after the model is
evicted from VRAM. It applies to every backend, so it is deliberately not
set higher than needed.

Caveat recorded honestly: at the time of writing, generation on that server
does not complete. /v1/models answers instantly, but both /v1/chat/
completions (180s) and native /api/generate with num_predict=5 (60s) return
nothing, and /api/ps shows no model ever becomes resident - so the model
never finishes loading. Ollama is 0.32.14 and gemma4:e2b is 5.1B Q4_K_M
(~3.5GB) despite the "e2b" name. That is a server-side problem, not a
configuration one; these values are correct and take effect once it loads.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

astrololo/README.md

@@ -29,6 +29,49 @@ Runbook: **[README-pliki.md](README-pliki.md)**. Ekran „Pliki" wymaga zapisu d
 udziału z bazami, więc znosi gwarancję z DAN-25, że baz nie da się zmienić przez
 NFS. Co zostaje z zabezpieczeń i jak to wdrożyć — w runbooku.
 
+## ⚠️ Wymóg węzłów: klient NFS
+
+Wszystkie warstwy montują udziały z NAS-a, więc **każdy węzeł, na którym może
+wylądować pod, musi mieć klienta NFS**. Bez niego kubelet nie zamontuje wolumenu:
+
+```
+mount: ... bad option; for several filesystems (e.g. nfs, cifs) you might need
+a /sbin/mount.<type> helper program
+```
+
+Ten komunikat **nie oznacza problemu z udziałem ani z uprawnieniami** — serwer
+w ogóle nie został zapytany. Jądro nie znalazło programu pomocniczego
+`/sbin/mount.nfs`.
+
+### Sprawdzenie
+
+```bash
+for N in 192.168.1.73 192.168.1.80 192.168.1.81; do
+  printf "%-15s " "$N"
+  ssh hammer@$N 'test -x /sbin/mount.nfs && echo MA-KLIENTA || echo BRAK-KLIENTA'
+done
+```
+
+### Instalacja
+
+```bash
+ssh hammer@<węzeł> 'sudo apt-get update && sudo apt-get install -y nfs-common'
+```
+
+> **Na WSZYSTKICH węzłach, nie tylko na tym, gdzie pod stoi teraz.** Braku na
+> pozostałych nie widać, dopóki scheduler tam czegoś nie przeniesie — a wtedy
+> awaria wygląda na nagłą, choć przyczyna leżała od dawna. Dokładnie tak wyszło
+> za pierwszym razem: pody działały miesiącami na jednym węźle, aż zejście do
+> zera replik przy innej naprawie przeplanowało je gdzie indziej.
+
+### Skąd wiadomo, że to TO
+
+| komunikat | co znaczy |
+|---|---|
+| `bad option ... mount.<type> helper program` | **brak `nfs-common` na węźle** — serwer niepytany |
+| `access denied by server while mounting` | serwer odmawia: eksport nieprzeładowany, węzła nie ma na liście `hosts`, albo udział wyłączony |
+| `Permission denied` przy zapisie | montowanie działa, brakuje praw — patrz `mapall_user` i właściciel katalogu |
+
 ## ⚠️ Sekret `astrololo-auth` — utwórz PRZED wdrożeniem
 
 Aplikacja wystawia treść **oryginalnych baz interpretacyjnych**, dlatego wymaga

astrololo/kustomization.yaml

@@ -11,10 +11,10 @@ resources:
   - ingress.yaml # wejście po https + przekierowanie z http
 images:
   - name: gitea.czernobog.pl/gitea/astrololo-data
-    newTag: ee3c515d
+    newTag: f0d07ee8
   - name: gitea.czernobog.pl/gitea/astrololo-logic
-    newTag: ee3c515d
+    newTag: f0d07ee8
   - name: gitea.czernobog.pl/gitea/astrololo-render
     newTag: latest
   - name: gitea.czernobog.pl/gitea/astrololo-presentation
-    newTag: ee3c515d
+    newTag: f0d07ee8

conjurer/bot.yaml

@@ -31,6 +31,19 @@ spec:
             # Where the librarian sends THIS bot's results/pongs back to (its own
             # NodePort). Lets one librarian serve both bots - see deploy-bot.yaml.
             - { name: CONJURER_SELF_CALLBACK,     value: "http://192.168.1.73:32442" }
+            # Self-hosted models (Ollama). No API key - the endpoint IS the
+            # configuration, and the backend stays unselectable while unset.
+            # Pick a model at runtime with: $gadaj_teraz ollama <model>
+            # ($modele_ai lists what the server actually has pulled).
+            - { name: CONJURER_OLLAMA_URL,        value: "http://192.168.1.72:11434" }
+            # The server currently has exactly one model pulled (verified via
+            # /v1/models): gemma4:e2b. Without this the built-in default
+            # (llama3.1:8b) would be requested and every reply would fail.
+            - { name: CONJURER_OLLAMA_MODEL,      value: "gemma4:e2b" }
+            # Self-hosted generation is far slower than a hosted API, especially
+            # the first request after the model is evicted from VRAM. Applies to
+            # every backend, so keep it only as high as you actually need.
+            - { name: CONJURER_AI_TIMEOUT_SECONDS, value: "240" }
           volumeMounts:
             - { name: data,  mountPath: /data }
             - { name: netrc, mountPath: /secrets, readOnly: true }

conjurer/deploy-bot.yaml

@@ -51,6 +51,19 @@ spec:
             # query and answers results/pongs HERE - so it serves this bot AND the
             # test bot from one instance, no CONJURER_MAIN_BOT repointing needed.
             - { name: CONJURER_SELF_CALLBACK,     value: "http://192.168.1.73:32443" }
+            # Self-hosted models (Ollama). No API key - the endpoint IS the
+            # configuration, and the backend stays unselectable while unset.
+            # Pick a model at runtime with: $gadaj_teraz ollama <model>
+            # ($modele_ai lists what the server actually has pulled).
+            - { name: CONJURER_OLLAMA_URL,        value: "http://192.168.1.72:11434" }
+            # The server currently has exactly one model pulled (verified via
+            # /v1/models): gemma4:e2b. Without this the built-in default
+            # (llama3.1:8b) would be requested and every reply would fail.
+            - { name: CONJURER_OLLAMA_MODEL,      value: "gemma4:e2b" }
+            # Self-hosted generation is far slower than a hosted API, especially
+            # the first request after the model is evicted from VRAM. Applies to
+            # every backend, so keep it only as high as you actually need.
+            - { name: CONJURER_AI_TIMEOUT_SECONDS, value: "240" }
           volumeMounts:
             - { name: data,  mountPath: /data }
             - { name: netrc, mountPath: /secrets, readOnly: true }