materio_ai: cached search vectors, 1 s delay, pause and cron warm-up (H8)

- QueryEmbedder: vector of a search phrase cached 30 days (cache bin
  materio_ai), model given search_timeout seconds (1 by default, new
  setting, update 11001), left alone 60 s after a failure
  (ModelPausedException, not logged).
- NeighbourSearch honours the enabled setting: the search called the model
  even when materio_ai was switched off.
- materio_ai_cron() loads the model with the normal timeout: after a
  restart, loading (2.9 s) exceeds the search delay and Ollama drops it,
  so searches alone would never load it. Ultimate cron job exported.
- On any model failure the search answers with the keyword results only.
- QueryEmbedderTest: cache, delay, pause, cache during the pause, warm-up.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ShDspHaZGwHoRZCsPBJzsM
This commit is contained in:
2026-11-17 18:26:19 +01:00
co-authored by Claude Opus 5.5
parent a656427aec
commit 6d8c64c94f
15 changed files with 379 additions and 13 deletions
+1
View File
@@ -5,4 +5,5 @@ url: 'http://ollama:11434'
model: 'bge-m3:567m'
dimension: 1024
timeout: 10
search_timeout: 1.0
num_thread: 4
@@ -0,0 +1,17 @@
uuid: 77f99dd9-eeab-40af-9925-95a9f7cd6cee
langcode: en
status: true
dependencies:
module:
- materio_ai
title: 'Default cron handler'
id: materio_ai_cron
weight: 0
module: materio_ai
callback: 'materio_ai#cron'
scheduler:
id: simple
launcher:
id: serial
logger:
id: database