Ethanfel/ComfyUI-Tenaciousload
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Initial commit: in-process object_info cache + refresh UI

Browse Source 
ComfyUI-Tenaciousload speeds up ComfyUI page loads for large model/LoRA
collections by caching the slow /api/object_info response in memory and on
disk (survives restarts) and serving it gzipped in ~milliseconds, instead of
rebuilding it (and freezing the event loop) on every load.

Adds a "Refresh Models / LoRAs" menu button, a graph node, and a
POST /tenaciousload/refresh endpoint to rebuild the cache after adding or
removing models. No external dependencies; no nginx/docker required.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Ethanfel
2026-06-04 00:25:41 +02:00
commit 2c1777ae4e
7 changed files with 377 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.gitignore
+3
View File
@@ -0,0 +1,3 @@
cache/
__pycache__/
*.pyc
LICENSE
+21
View File
@@ -0,0 +1,21 @@
MIT License
Copyright (c) 2026 Ethanfel
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
README.md
+72
View File
@@ -0,0 +1,72 @@
# ComfyUI-Tenaciousload
Self-contained fix for slow / black-screen ComfyUI loading when you have a huge
model/LoRA collection (especially on a network mount). **Just install the pack
and restart ComfyUI — no nginx, no docker, no extra port.**
## The problem
ComfyUI's `/api/object_info` enumerates every node's inputs. With thousands of
LoRAs (worse on a network mount) it becomes tens of MB and takes **minutes to
build on every page load** — and the build **freezes ComfyUI's whole event
loop**, so you get a long black screen, worst over a remote network.
## How this pack fixes it
On load it injects an aiohttp **middleware** into ComfyUI that intercepts
`/object_info` and `/api/object_info` and:
- **caches the built response in memory _and_ on disk** (`./cache/`), so it is
built once instead of on every load — and the disk copy makes **restarts
instant** (no rebuild);
- **serves it gzipped in ~milliseconds** (≈85% smaller, independent of any CLI
flag);
- serves from cache **without running the build**, so the event-loop freeze is
gone for normal loads.
The only time a build runs is the first load after install, or when you
explicitly refresh (below).
## Refreshing after you add / remove models or LoRAs
The cache holds the old model lists until you refresh, so new files won't appear
until you do one of:
- **Menu:** `Extensions ▸ 🔄 Refresh Models / LoRAs` (also in the command palette).
- **Graph node:** `🔄 Refresh Models/LoRAs (Tenaciousload)` (for automated workflows).
- **HTTP:** `POST /tenaciousload/refresh`, then `GET /object_info?nocache=1`.
A refresh re-walks your model folders (slow over a network mount, ~minutes) — the
button shows a "refreshing…" toast meanwhile. Normal loads stay instant.
## Requirements
**None to install.** Only ComfyUI itself (tested on 0.23.0) and Python ≥ 3.8.
Everything used is Python stdlib or already bundled with ComfyUI (`aiohttp`,
`folder_paths`, `server`). The web button needs no npm packages.
## Install
Clone (or copy) this repo into your ComfyUI `custom_nodes/` folder and restart
ComfyUI:
```bash
cd ComfyUI/custom_nodes
git clone https://github.com/ethanfel/ComfyUI-Tenaciousload.git
# then restart ComfyUI
```
Nothing to `pip install`. ComfyUI-Manager can also install it from the registry.
## Verify it's working
After restart, load the page once (first time builds + caches), then:
```bash
curl -s -H 'Accept-Encoding: gzip' -o /dev/null \
-w '%{time_total}s | %{size_download} bytes | %header{x-tenaciousload-cache} | %header{content-encoding}\n' \
http://127.0.0.1:8188/api/object_info # use your ComfyUI port
# expect after the first load: ~0.00Xs | ~10 MB | HIT | gzip
```
ComfyUI's startup log should show `Tenaciousload: object_info cache middleware installed`.
## Notes
- The disk cache lives in `./cache/` (git-ignored). Delete it, or use the refresh
button, to force a rebuild.
- An nginx reverse proxy can cache `object_info` at the HTTP layer too, but this
pack does it in-process so no extra service, container, or port is needed.
## License
MIT — see [LICENSE](LICENSE).
__init__.py
+203
View File
@@ -0,0 +1,203 @@
"""
ComfyUI-Tenaciousload
=====================
Self-contained fix for slow / black-screen ComfyUI loading when you have a huge
model/LoRA collection (especially on a network mount).
The cause: /api/object_info is tens of MB and takes minutes to build on EVERY
page load (it re-walks the model folders), and the build blocks ComfyUI's event
loop the whole time. This pack installs an in-process caching layer that:
* caches the built object_info in memory AND on disk (so it survives restarts),
* serves it gzipped in ~milliseconds instead of rebuilding every load,
* exposes a one-click "Refresh Models / LoRAs" button + graph node to rebuild
the cache after you add/remove models.
No nginx, no docker, no extra port. Just install the pack and restart ComfyUI.
"""
import os
import gzip
import json
import logging
import threading
from aiohttp import web
import folder_paths
from server import PromptServer
log = logging.getLogger("Tenaciousload")
WEB_DIRECTORY = "./web"
# --- cache storage --------------------------------------------------------------
_CACHE_DIR = os.path.join(os.path.dirname(__file__), "cache")
_RAW_PATH = os.path.join(_CACHE_DIR, "object_info.json")
_GZ_PATH = os.path.join(_CACHE_DIR, "object_info.json.gz")
_OBJECT_INFO_PATHS = ("/object_info", "/api/object_info")
_GZIP_LEVEL = 5
_lock = threading.Lock()
_mem = {"raw": None, "gz": None} # bytes / bytes
_disk_loaded = False
def _load_from_disk():
"""Populate the in-memory cache from disk once (cheap, makes restarts instant)."""
global _disk_loaded
_disk_loaded = True
try:
if os.path.exists(_GZ_PATH):
with open(_GZ_PATH, "rb") as f:
_mem["gz"] = f.read()
if os.path.exists(_RAW_PATH):
with open(_RAW_PATH, "rb") as f:
_mem["raw"] = f.read()
if _mem["gz"] is not None and _mem["raw"] is None:
_mem["raw"] = gzip.decompress(_mem["gz"])
if _mem["raw"] is not None:
log.info("Tenaciousload: loaded object_info cache from disk (%d bytes raw)", len(_mem["raw"]))
except Exception as e: # pragma: no cover
log.warning("Tenaciousload: failed to load disk cache: %s", e)
def _store(raw_bytes):
"""Cache a freshly built object_info body (memory + disk)."""
with _lock:
gz = gzip.compress(raw_bytes, _GZIP_LEVEL)
_mem["raw"] = raw_bytes
_mem["gz"] = gz
try:
os.makedirs(_CACHE_DIR, exist_ok=True)
with open(_RAW_PATH, "wb") as f:
f.write(raw_bytes)
with open(_GZ_PATH, "wb") as f:
f.write(gz)
except Exception as e: # pragma: no cover
log.warning("Tenaciousload: failed to persist disk cache: %s", e)
log.info("Tenaciousload: cached object_info (%d bytes raw / %d gz)", len(raw_bytes), len(gz))
def invalidate_object_info_cache():
"""Drop the cached object_info so the next request rebuilds it."""
with _lock:
_mem["raw"] = None
_mem["gz"] = None
for p in (_RAW_PATH, _GZ_PATH):
try:
os.remove(p)
except FileNotFoundError:
pass
except Exception as e: # pragma: no cover
log.warning("Tenaciousload: could not delete %s: %s", p, e)
def clear_comfy_model_cache():
"""Force ComfyUI to re-scan every model/LoRA folder on the next build."""
try:
folder_paths.filename_list_cache.clear()
except Exception as e: # pragma: no cover
log.warning("Tenaciousload: could not clear filename_list_cache: %s", e)
try:
folder_paths.cache_helper.clear()
except Exception as e: # pragma: no cover
log.warning("Tenaciousload: could not clear cache_helper: %s", e)
def _serve_cached(request):
raw = _mem["raw"]
gz = _mem["gz"]
accepts_gzip = "gzip" in request.headers.get("Accept-Encoding", "")
if accepts_gzip and gz is not None:
return web.Response(
body=gz, status=200, content_type="application/json",
headers={"Content-Encoding": "gzip", "X-Tenaciousload-Cache": "HIT",
"Cache-Control": "no-store"},
)
if raw is None and gz is not None:
raw = gzip.decompress(gz)
return web.Response(
body=raw, status=200, content_type="application/json",
headers={"X-Tenaciousload-Cache": "HIT", "Cache-Control": "no-store"},
)
@web.middleware
async def _object_info_cache_mw(request, handler):
# Only touch the full object_info list (not /object_info/{node_class}).
if request.method != "GET" or request.path not in _OBJECT_INFO_PATHS:
return await handler(request)
global _disk_loaded
if not _disk_loaded:
_load_from_disk()
nocache = "nocache" in request.query
if not nocache and _mem["raw"] is not None:
return _serve_cached(request)
# MISS (or forced refresh): build via the real handler, then cache the body.
resp = await handler(request)
try:
body = getattr(resp, "body", None)
if resp.status == 200 and isinstance(body, (bytes, bytearray)) and len(body) > 0:
_store(bytes(body))
return _serve_cached(request)
except Exception as e: # pragma: no cover
log.warning("Tenaciousload: caching skipped: %s", e)
return resp
def _install_middleware():
try:
app = PromptServer.instance.app
app.middlewares.insert(0, _object_info_cache_mw)
log.info("Tenaciousload: object_info cache middleware installed")
except Exception as e:
log.error("Tenaciousload: could not install cache middleware (loads will be slow): %s", e)
_install_middleware()
# --- refresh API ----------------------------------------------------------------
@PromptServer.instance.routes.post("/tenaciousload/refresh")
async def _refresh(request):
clear_comfy_model_cache()
invalidate_object_info_cache()
log.info("Tenaciousload: caches cleared via refresh endpoint")
return web.json_response({"status": "ok"})
# --- optional graph node (for workflow automation) ------------------------------
class _AnyType(str):
def __ne__(self, other):
return False
ANY = _AnyType("*")
class TenaciousloadRefresh:
@classmethod
def INPUT_TYPES(cls):
return {"optional": {"trigger": (ANY, {})}}
RETURN_TYPES = (ANY,)
RETURN_NAMES = ("trigger",)
FUNCTION = "refresh"
CATEGORY = "Tenaciousload"
OUTPUT_NODE = True
DESCRIPTION = "Clear ComfyUI's model/LoRA cache and invalidate the object_info cache."
def refresh(self, trigger=None):
clear_comfy_model_cache()
invalidate_object_info_cache()
return (trigger,)
NODE_CLASS_MAPPINGS = {"TenaciousloadRefresh": TenaciousloadRefresh}
NODE_DISPLAY_NAME_MAPPINGS = {"TenaciousloadRefresh": "🔄 Refresh Models/LoRAs (Tenaciousload)"}
__all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS", "WEB_DIRECTORY"]
pyproject.toml
+14
View File
@@ -0,0 +1,14 @@
[project]
name = "comfyui-tenaciousload"
description = "Fast ComfyUI loading behind an nginx cache, with an in-UI button to refresh model/LoRA lists."
version = "1.0.0"
license = { text = "MIT" }
dependencies = []
[project.urls]
Repository = "https://github.com/ethanfel/ComfyUI-Tenaciousload"
[tool.comfy]
PublisherId = "ethanfel"
DisplayName = "Tenaciousload"
Icon = ""
requirements.txt
+9
View File
@@ -0,0 +1,9 @@
# ComfyUI-Tenaciousload has NO external dependencies.
# Everything it uses is either the Python standard library
# (os, gzip, json, logging, threading) or already provided by ComfyUI:
# - aiohttp (ComfyUI's web server, aiohttp>=3.11.8)
# - folder_paths, server (ComfyUI internal modules)
#
# Requirements to run:
# - ComfyUI (any recent version; tested on 0.23.0)
# - Python >= 3.8
web/tenaciousload.js
+55
View File
@@ -0,0 +1,55 @@
import { app } from "../../scripts/app.js";
import { api } from "../../scripts/api.js";
let busy = false;
function notify(severity, detail, life = 6000) {
const toast = app.extensionManager?.toast;
if (toast?.add) {
toast.add({ severity, summary: "Tenaciousload", detail, life });
} else if (severity === "error") {
alert("Tenaciousload: " + detail);
}
console[severity === "error" ? "error" : "log"]("[Tenaciousload]", detail);
}
async function doRefresh() {
if (busy) {
notify("warn", "A refresh is already running…");
return;
}
busy = true;
notify("info", "Refreshing models/LoRAs — scanning over the network, this can take a few minutes…", 10000);
try {
// 1) clear ComfyUI's folder cache + drop the object_info cache.
await api.fetchApi("/tenaciousload/refresh", { method: "POST" });
// 2) force a fresh object_info build and re-cache it (this is the slow step).
await api.fetchApi("/object_info?nocache=1").then((r) => r.arrayBuffer());
// 3) update open node dropdowns live if the frontend supports it.
try { await app.refreshComboInNodes?.(); } catch (e) { /* non-fatal */ }
notify("success", "Done — new models are now available (reload the page if a dropdown still looks stale).", 8000);
} catch (e) {
notify("error", "Refresh failed: " + (e?.message || e), 10000);
} finally {
busy = false;
}
}
app.registerExtension({
name: "Tenaciousload.Refresh",
commands: [
{
id: "Tenaciousload.refresh",
label: "🔄 Refresh Models / LoRAs",
icon: "pi pi-refresh",
function: doRefresh,
},
],
// Adds the command under the top "Extensions" menu (and the command palette).
menuCommands: [
{
path: ["Extensions"],
commands: ["Tenaciousload.refresh"],
},
],
});