Enhance CLI options and idle-killer functionality

- Added a `--update` option to the `up` command in the CLI for forced updates during warm ACTIVE states.
- Improved the idle-killer logic with new functions to check for reusable credentials and refresh settings on existing remote credentials.
- Enhanced the `seed_civitai` function to implement caching for model jobs, optimizing the provisioning process.
- Updated the installation script for Ollama to prevent unnecessary restarts when the service configuration has not changed.
- Added diagnostics for model job caching and improved error handling in various functions.

src/gpu_rent/provision.py

@@ -38,6 +38,18 @@ DATA = "/mnt/swarm_data"
 _OLLAMA_INSTALL_ENV = ("OLLAMA_VERSION", "OLLAMA_SHA256")
 
 
+def _models_manifest_fp(cfg: Config) -> str:
+    """Stable fingerprint of models.yaml so warm seed can skip API+fetch."""
+    import hashlib
+
+    path = Path(cfg.models_manifest)
+    try:
+        raw = path.read_bytes()
+    except OSError:
+        raw = b""
+    return hashlib.sha256(raw).hexdigest()[:16]
+
+
 def _remote_llm_env(cfg: Config, *keys: str) -> dict[str, str]:
     import os
 
@@ -408,6 +420,41 @@ def seed_civitai(cfg: Config, host: str, log: Log) -> None:
         log("model-seed пропущен: манифест пуст — дефолт SwarmUI")
         return
 
+    # Warm short-circuit: last successful jobs list still present on disk.
+    cache_remote = f"{DATA}/.gpu-rent-models-jobs.json"
+    try:
+        cached_raw = run_ssh(
+            cfg,
+            host,
+            f"test -f {cache_remote} && cat {cache_remote} || true",
+            check=False,
+            timeout=20,
+        ).strip()
+    except Exception:
+        cached_raw = ""
+    if cached_raw.startswith("{"):
+        try:
+            cached = json.loads(cached_raw)
+            dests = [str(d) for d in (cached.get("dests") or []) if d]
+            fp = str(cached.get("fp") or "")
+            want_fp = _models_manifest_fp(cfg)
+            if dests and fp == want_fp:
+                check = " && ".join(f"test -f {shlex.quote(d)}" for d in dests)
+                ok = run_ssh(
+                    cfg,
+                    host,
+                    f"if {check}; then echo ALL_OK; else echo MISSING; fi",
+                    check=False,
+                    timeout=30,
+                ).strip()
+                if "ALL_OK" in ok:
+                    log(
+                        f"model-seed: skip — {len(dests)} файл(ов) уже на диске (cache)"
+                    )
+                    return
+        except (json.JSONDecodeError, TypeError, CloudError):
+            pass
+
     jobs: list[dict] = []
     for entry in entries:
         url = (entry.url or "").strip()
@@ -514,6 +561,31 @@ def seed_civitai(cfg: Config, host: str, log: Log) -> None:
 
     if not any(e.kind == "checkpoint" for e in entries):
         log("в манифесте нет checkpoint — генерация может не стартовать")
+
+    dests = [str(j.get("dest") or "") for j in jobs if j.get("dest")]
+    if dests:
+        check = " && ".join(f"test -f {shlex.quote(d)}" for d in dests)
+        present = run_ssh(
+            cfg,
+            host,
+            f"if {check}; then echo ALL_OK; else echo MISSING; fi",
+            check=False,
+            timeout=30,
+        ).strip()
+        if "ALL_OK" in present:
+            put_text(
+                cfg,
+                host,
+                cache_remote,
+                json.dumps({"fp": _models_manifest_fp(cfg), "dests": dests}, indent=2)
+                + "\n",
+            )
+            log(
+                f"model-seed: skip fetch — {len(dests)} файл(ов) уже на диске "
+                f"(civitai={civ_n}, huggingface={hf_n})"
+            )
+            return
+
     put_text(cfg, host, "/tmp/gpu-rent-civitai-jobs.json", json.dumps(jobs, indent=2))
     if cfg.civitai_api_token:
         put_text(
@@ -533,6 +605,12 @@ def seed_civitai(cfg: Config, host: str, log: Log) -> None:
         timeout=7200,
         log=log,
     )
+    put_text(
+        cfg,
+        host,
+        cache_remote,
+        json.dumps({"fp": _models_manifest_fp(cfg), "dests": dests}, indent=2) + "\n",
+    )
 
 
 def ensure_swarmui_running(cfg: Config, host: str, log: Log, restart: bool) -> None:
@@ -594,16 +672,39 @@ def provision_llm(cfg: Config, host: str, log: Log) -> None:
         if not names:
             log("ollama-models.yaml пуст — pull skip")
         else:
-            put_text(cfg, host, "/tmp/gpu-rent-ollama-models.json", json.dumps(names, indent=2))
-            log(f"Ollama: pull {len(names)} из манифеста")
-            run_python(
+            # Fast path: all tags already present — skip upload/pull script.
+            listed = run_ssh(
                 cfg,
                 host,
-                _pkg_text("ollama_pull.py"),
-                remote_path="/tmp/gpu-rent-ollama_pull.py",
-                timeout=7200,
-                log=log,
+                "ollama list 2>/dev/null | awk 'NR>1 {print $1}' || true",
+                check=False,
+                timeout=30,
             )
+            have = {ln.strip() for ln in listed.splitlines() if ln.strip()}
+            missing = []
+            for name in names:
+                if name in have or (
+                    ":" not in name and f"{name}:latest" in have
+                ) or (
+                    name.endswith(":latest") and name.rsplit(":", 1)[0] in have
+                ):
+                    continue
+                missing.append(name)
+            if not missing:
+                log(f"ollama pull: skip — все {len(names)} уже есть")
+            else:
+                put_text(
+                    cfg, host, "/tmp/gpu-rent-ollama-models.json", json.dumps(missing, indent=2)
+                )
+                log(f"Ollama: pull {len(missing)} из манифеста (нет: {len(missing)})")
+                run_python(
+                    cfg,
+                    host,
+                    _pkg_text("ollama_pull.py"),
+                    remote_path="/tmp/gpu-rent-ollama_pull.py",
+                    timeout=7200,
+                    log=log,
+                )
     else:
         raise CloudError(f"неизвестный LLM_RUNTIME={runtime!r}")
     st = load_state()