Bump release version to 0.9.35rc460

Dockerfile

@@ -10,10 +10,10 @@
 #       --build-context abx-plugins=../abx-plugins \
 #       -t archivebox/abx-dl:dev
 #   docker buildx build . -f Dockerfile \
-#       --build-arg ABX_DL_IMAGE=archivebox/abx-dl:1.12.274 \
+#       --build-arg ABX_DL_IMAGE=archivebox/abx-dl:1.12.281 \
 #       -t archivebox:multistage
 
-ARG ABX_DL_IMAGE=archivebox/abx-dl:1.12.274
+ARG ABX_DL_IMAGE=archivebox/abx-dl:1.12.281
 
 FROM archivebox/sonic:1.4.9 AS sonic
 FROM ${ABX_DL_IMAGE} AS archivebox-runtime-base
@@ -96,17 +96,15 @@ RUN --mount=type=cache,target=/var/cache/apt,sharing=locked,id=apt-$TARGETARCH$T
     --mount=type=bind,source=pyproject.toml,target=/app/pyproject.toml \
     <<'EOF'
 echo "[+] UV Installing ArchiveBox dependencies from pyproject.toml..."
-echo 'Binary::apt::APT::Keep-Downloaded-Packages "1";' > /etc/apt/apt.conf.d/99keep-cache
-echo 'APT::Install-Recommends "0";' > /etc/apt/apt.conf.d/99no-install-recommends
-echo 'APT::Install-Suggests "0";' > /etc/apt/apt.conf.d/99no-install-suggests
+export UV_NO_CACHE=true
 rm -f /etc/apt/apt.conf.d/docker-clean
 apt-get update -qq
-apt-get install -qq -y --no-install-recommends \
-    build-essential gcc libldap2-dev libsasl2-dev libssl-dev x11-utils
+apt-get install -qq -y --no-install-recommends x11-utils
 # Extend abx-dl's existing venv instead of clearing it. Clearing and later
 # copying a complete replacement over the base duplicates every venv byte in
 # the final overlay history even when most packages are unchanged.
-/usr/bin/uv pip install --no-cache setuptools pip wheel
+# Do not rewrite inherited native libraries merely to strip them again.
+/usr/bin/find /venv/lib/python3.*/site-packages -type f -name '*.so' -print > /tmp/archivebox-inherited-libraries
 
 mkdir -p /tmp/archivebox-uv-project
 /venv/bin/python - <<'PY'
@@ -155,25 +153,31 @@ PY
     --inexact \
     --no-install-project \
     --no-install-workspace \
+    --no-install-package abxbus \
+    --no-install-package abxpkg \
+    --no-install-package abx-plugins \
+    --no-install-package abx-dl \
     --no-sources
 builder_abxpkg_lib_dir=/tmp/archivebox-builder-abxpkg
 ABXPKG_NO_CACHE=True abxpkg env --install --binproviders=env,apt --lib="$builder_abxpkg_lib_dir" --overrides='{"apt":{"install_args":["binutils"]}}' strip >/dev/null
 /usr/bin/find /venv/lib/python3.*/site-packages -type f -name '*.so' -print0 > /tmp/archivebox-native-libraries
 while IFS= read -r -d '' native_library; do
+    if grep -Fxq "$native_library" /tmp/archivebox-inherited-libraries; then
+        continue
+    fi
     magic=''
     if IFS= read -r -N 4 magic < "$native_library" && [[ "$magic" == $'\x7fELF' ]]; then
         "$builder_abxpkg_lib_dir/env/bin/strip" --strip-unneeded "$native_library" || exit $?
     fi
 done < /tmp/archivebox-native-libraries
-rm -f /tmp/archivebox-native-libraries
+rm -f /tmp/archivebox-native-libraries /tmp/archivebox-inherited-libraries
 rm -f /venv/bin/uv /venv/bin/uvx
-abxpkg run --binproviders=env --lib="$builder_abxpkg_lib_dir" apt-get purge -y binutils build-essential gcc libldap2-dev libsasl2-dev libssl-dev
+abxpkg run --binproviders=env --lib="$builder_abxpkg_lib_dir" apt-get purge -y binutils
 abxpkg run --binproviders=env --lib="$builder_abxpkg_lib_dir" apt-get autoremove -y
 rm -rf "$builder_abxpkg_lib_dir"
 rm -rf /venv/lib/python3.*/site-packages/pip* \
     /venv/lib/python3.*/site-packages/wheel* \
     /venv/bin/pip /venv/bin/pip3 /venv/bin/pip3.* /venv/bin/wheel
-echo 'Binary::apt::APT::Keep-Downloaded-Packages "0";' > /etc/apt/apt.conf.d/99keep-cache
 rm -rf /var/lib/apt/lists/* /tmp/archivebox-uv-project
 EOF
 
@@ -202,7 +206,7 @@ RUN echo "[*] Installing ArchiveBox Python source code from $CODE_DIR..." \
     && test -f "$(/venv/bin/python -c 'import archivebox; print(archivebox.__cached__)')" \
     && /usr/bin/uv pip show archivebox | tee -a /VERSION.txt
 
-# The builder installs and purges compilers in one layer, so its final
+# The builder installs and purges stripping tools in one layer, so its final
 # filesystem is already runtime-clean. Preserve that ancestry: starting again
 # from archivebox-runtime-base and COPYing /venv would bake a second full venv
 # over the inherited one instead of recording only ArchiveBox's package delta.
@@ -225,7 +229,8 @@ COPY --chown=root:root --chmod=755 "etc/sonic.cfg" /etc/sonic.cfg
 # The builder invokes abxpkg as root for temporary ELF stripping, which can
 # rewrite derived state ownership. Restore UID 911 before resolving binaries
 # as the runtime user; the old copy-based final stage hid this dependency.
-RUN echo "[*] Setting up $ARCHIVEBOX_USER user uid=${DEFAULT_ARCHIVEBOX_UID}..." \
+RUN --mount=type=cache,target=/opt/archivebox/lib/cache,sharing=locked,mode=1777,id=archivebox-runtime-config-$TARGETARCH \
+    echo "[*] Setting up $ARCHIVEBOX_USER user uid=${DEFAULT_ARCHIVEBOX_UID}..." \
     && printf 'export PATH="/venv/bin:/opt/node/bin:$PATH"\n' > /etc/profile.d/archivebox-path.sh \
     && ln -sf /venv/bin/archivebox /usr/local/bin/archivebox \
     && ln -sf /venv/bin/daphne /usr/local/bin/daphne \
@@ -239,8 +244,8 @@ RUN echo "[*] Setting up $ARCHIVEBOX_USER user uid=${DEFAULT_ARCHIVEBOX_UID}..."
     && install -d -o "$DEFAULT_ARCHIVEBOX_UID" -g "$DEFAULT_ARCHIVEBOX_GID" "$DATA_DIR" "$TMP_DIR" "$CONFIG_DIR" "$ABXPKG_LIB_DIR" "$XDG_CACHE_HOME" "$PLAYWRIGHT_BROWSERS_PATH" \
     && install -d -o "$DEFAULT_ARCHIVEBOX_UID" -g "$DEFAULT_ARCHIVEBOX_GID" "/home/$ARCHIVEBOX_USER" \
     && chown "$DEFAULT_ARCHIVEBOX_UID:$DEFAULT_ARCHIVEBOX_GID" "$DATA_DIR" "$TMP_DIR" \
-    && chown -R "$DEFAULT_ARCHIVEBOX_UID:$DEFAULT_ARCHIVEBOX_GID" "$ABXPKG_LIB_DIR" \
-    && setpriv --reuid="$ARCHIVEBOX_USER" --regid="$ARCHIVEBOX_USER" --init-groups abxpkg load --binproviders=env sonic | tee -a /VERSION.txt \
+    && find "$ABXPKG_LIB_DIR" \( ! -user "$DEFAULT_ARCHIVEBOX_UID" -o ! -group "$DEFAULT_ARCHIVEBOX_GID" \) -exec chown -h "$DEFAULT_ARCHIVEBOX_UID:$DEFAULT_ARCHIVEBOX_GID" {} + \
+    && UV_NO_CACHE=true setpriv --reuid="$ARCHIVEBOX_USER" --regid="$ARCHIVEBOX_USER" --init-groups abxpkg load --binproviders=env sonic | tee -a /VERSION.txt \
     && openssl rand -hex 16 > /etc/machine-id \
     && echo -e "\nARCHIVEBOX_USER=$ARCHIVEBOX_USER ARCHIVEBOX_UID=$(id -u "$ARCHIVEBOX_USER") ARCHIVEBOX_GID=$(id -g "$ARCHIVEBOX_USER")" | tee -a /VERSION.txt \
     && echo -e "TMP_DIR=$TMP_DIR\nABXPKG_LIB_DIR=$ABXPKG_LIB_DIR\nPLAYWRIGHT_BROWSERS_PATH=$PLAYWRIGHT_BROWSERS_PATH\nMACHINE_ID=$(cat /etc/machine-id)\n" | tee -a /VERSION.txt

archivebox/api/v1_auth.py

@@ -4,7 +4,7 @@ from django.http import HttpRequest
 
 from ninja import Router, Schema
 
-from archivebox.api.auth import auth_using_token, auth_using_password, get_or_create_api_token
+from archivebox.api.auth import auth_using_token, auth_using_password, get_or_create_api_token, HeaderTokenAuth, BearerTokenAuth
 
 
 router = Router(tags=["Authentication"], auth=None)
@@ -63,3 +63,44 @@ def check_api_token(request: HttpRequest, token_data: TokenAuthSchema):
         return {"success": True, "user_id": str(user.pk)}
 
     return {"success": False, "user_id": None}
+
+
+@router.post(
+    "/browser_session",
+    auth=[HeaderTokenAuth(), BearerTokenAuth()],
+    summary="Exchange an administrator API key for a normal browser session",
+)
+def browser_session(request: HttpRequest):
+    from django.conf import settings
+    from django.contrib.auth import login
+    from django.http import JsonResponse
+    from django.utils import timezone
+    from ninja.errors import HttpError
+    from archivebox.core.routes_util import get_admin_base_url
+
+    user = request.auth
+    if not user.is_active or not user.is_superuser:
+        raise HttpError(403, "An active administrator is required")
+    login(request, user, backend="django.contrib.auth.backends.ModelBackend")
+    # A browser session must not outlive the key used to establish it.
+    token = request._api_token
+    lifetime = settings.SESSION_COOKIE_AGE
+    if token.expires:
+        lifetime = min(lifetime, max(0, int((token.expires - timezone.now()).total_seconds())))
+    request.session.set_expiry(lifetime)
+    request.session.save()
+    response = JsonResponse(
+        {
+            "admin_url": get_admin_base_url(request=request).rstrip("/") + "/admin/",
+            "cookie": {
+                "name": settings.SESSION_COOKIE_NAME,
+                "value": request.session.session_key,
+                "expires": request.session.get_expiry_date().timestamp(),
+                "secure": settings.SESSION_COOKIE_SECURE,
+            },
+        },
+    )
+    # The native client installs a host-only HttpOnly cookie on the admin origin;
+    # API and admin hosts can differ, so a Set-Cookie on the API host is insufficient.
+    response["Cache-Control"] = "no-store"
+    return response

archivebox/cli/archivebox_add.py

@@ -388,10 +388,6 @@ def main(**kwargs):
             raise click.BadParameter(str(err), param_hint="--snapshot-max-size") from err
         if kwargs.get("crawl_max_concurrent_snapshots") is not None and int(kwargs["crawl_max_concurrent_snapshots"]) < 1:
             raise click.BadParameter("crawl_max_concurrent_snapshots must be at least 1.", param_hint="--crawl-max-concurrent-snapshots")
-        if not kwargs.get("index_only"):
-            from archivebox.workers.supervisord_util import require_crawl_memory
-
-            require_crawl_memory()
 
         # Translate --only-new/--no-only-new into a crawl config override.
         # add() takes config overrides as a dict; no per-flag kwargs.

archivebox/cli/archivebox_server.py

@@ -363,12 +363,10 @@ def server(
     command = current_command(Process.TypeChoices.SERVER, data_dir=CONSTANTS.DATA_DIR, url=bind_url)
 
     def still_owns_runtime_stack() -> bool:
-        from django.db import connections
-
-        try:
-            return command_owns_runtime_stack(command, data_dir=CONSTANTS.DATA_DIR)
-        finally:
-            connections.close_all()
+        # This is polled at the log-tail cadence. Keep the autocommit
+        # connection open so SQLite can reuse its page cache between polls;
+        # reconnecting here repeatedly rereads the process table while idle.
+        return command_owns_runtime_stack(command, data_dir=CONSTANTS.DATA_DIR)
 
     shutdown_state = None
     try:
@@ -416,8 +414,13 @@ def server(
     except KeyboardInterrupt:
         pass
     finally:
-        if not shutdown_state or not shutdown_state.signal_name:
-            command.mark_exited()
+        from django.db import connections
+
+        try:
+            if not shutdown_state or not shutdown_state.signal_name:
+                command.mark_exited()
+        finally:
+            connections.close_all()
     print("\n[i][green][🟩] ArchiveBox server shut down gracefully.[/green][/i]")
 
 

archivebox/core/views.py

@@ -765,6 +765,11 @@ def _plugin_full_preview_response(
     response.headers["X-Content-Type-Options"] = "nosniff"
     response.headers["X-ArchiveBox-Security-Mode"] = request.archivebox_config.SERVER_SECURITY_MODE
     response.headers["Referrer-Policy"] = "no-referrer"
+    # Trusted viewers live on the snapshot origin, but the collection UI lives
+    # on web.* (and can be embedded by admin.*). Permit only those configured
+    # origins, not arbitrary sites or other snapshots. This does not grant the
+    # archived document access to the parent frame or its admin session cookies.
+    viewer_origins = " ".join(dict.fromkeys((build_web_url(request=request), build_admin_url(request=request))))
     response.headers["Content-Security-Policy"] = (
         "default-src 'self' data: blob:; "
         "script-src 'self' 'unsafe-inline' 'unsafe-eval' data: blob:; "
@@ -778,7 +783,7 @@ def _plugin_full_preview_response(
         "object-src 'none'; "
         "base-uri 'none'; "
         "form-action 'none'; "
-        "frame-ancestors 'self';"
+        f"frame-ancestors 'self' {viewer_origins};"
     )
     return response
 

archivebox/crawls/admin.py

@@ -1230,7 +1230,7 @@ class CrawlScheduleAdmin(BaseModelAdmin):
         (
             "Configuration",
             {
-                "fields": ("schedule", "template"),
+                "fields": ("is_enabled", "schedule", "template"),
                 "classes": ("card",),
             },
         ),
@@ -1276,7 +1276,21 @@ class CrawlScheduleAdmin(BaseModelAdmin):
 
     def change_view(self, request, object_id, form_url="", extra_context=None):
         self.request = request
-        return super().change_view(request, object_id, form_url, extra_context)
+        from django.db import connections, router
+
+        connection = connections[router.db_for_write(self.model)]
+        if request.method != "POST" or connection.vendor != "sqlite":
+            return super().change_view(request, object_id, form_url, extra_context)
+        # Reserve the writer before Django's atomic admin form reads. A
+        # deferred read transaction cannot safely upgrade while the runner
+        # writes, even when SQLite's existing busy timeout has time left.
+        connection.ensure_connection()
+        previous_mode = connection.transaction_mode
+        connection.transaction_mode = "IMMEDIATE"
+        try:
+            return super().change_view(request, object_id, form_url, extra_context)
+        finally:
+            connection.transaction_mode = previous_mode
 
     def add_view(self, request, form_url="", extra_context=None):
         return redirect("/add/#schedule")

archivebox/progressmonitor/collection.py

@@ -0,0 +1,54 @@
+"""Non-blocking, database-only collection totals for the progress monitor."""
+
+import logging
+from threading import Lock, Thread
+from time import time
+
+from django.core.cache import cache
+from django.db import close_old_connections, connections
+from django.db.models import Count, Sum
+
+
+_refresh_lock = Lock()
+
+
+def collection_summary(user):
+    user_id = None if user.is_superuser else user.pk
+    key = f"progress-collection:{'all' if user_id is None else user_id}"
+    summary = cache.get(key)
+    # Even an indexed SUM scans the index. Never make an HTTP request wait for
+    # that scan on a large collection. Serve stale totals while one worker refreshes;
+    # a single non-blocking lock prevents a queue of scans from concurrent clients.
+    if (summary is None or time() - summary["sampled_at"] >= 60) and _refresh_lock.acquire(blocking=False):
+        # Throttle failures too: a broken/locked database must not trigger work
+        # and log output on every sidebar poll. This cache is local to each worker.
+        if not cache.add(f"{key}:refresh", True, timeout=60):
+            _refresh_lock.release()
+            return summary
+
+        def refresh():
+            try:
+                close_old_connections()
+                from archivebox.core.models import Snapshot
+
+                snapshots = Snapshot.objects.all()
+                if user_id is not None:
+                    snapshots = snapshots.filter(crawl__created_by_id=user_id)
+                # COUNT(*) + SUM(output_size) can use Snapshot's covering size index.
+                # Snapshot.output_size is maintained when ArchiveResult sizes change.
+                totals = snapshots.aggregate(snapshots=Count("*"), bytes=Sum("output_size"))
+                totals["bytes"] = totals["bytes"] or 0
+                totals["sampled_at"] = time()
+                cache.set(key, totals, timeout=3600)
+            except Exception:
+                logging.getLogger(__name__).exception("Could not refresh collection totals")
+            finally:
+                connections.close_all()
+                _refresh_lock.release()
+
+        try:
+            Thread(target=refresh, name="collection-summary", daemon=True).start()
+        except RuntimeError:
+            _refresh_lock.release()
+            logging.getLogger(__name__).exception("Could not start collection totals refresh")
+    return summary

archivebox/progressmonitor/views.py

@@ -942,6 +942,10 @@ def live_progress_view(request):
             "queued_crawls_hidden": queued_crawls_hidden,
             "server_time": timezone.now().isoformat(),
         }
+        if is_admin and not snapshot_id_filter and not crawl_id_filter and request.GET.get("collection") == "1":
+            from archivebox.progressmonitor.collection import collection_summary
+
+            payload["collection"] = collection_summary(request.user)
         try:
             import ujson
 

archivebox/search/query.py

@@ -221,6 +221,7 @@ def iter_query_search_ids(
     search_mode: str | None = None,
     config: dict[str, Any] | None = None,
     max_results: int | None = None,
+    stop_event=None,
     **config_kwargs: Any,
 ):
     """Yield snapshot IDs from configured search backend modules."""
@@ -267,6 +268,8 @@ def iter_query_search_ids(
     seen: set[str] = set()
     try:
         for backend_name in backend_names:
+            if stop_event is not None and stop_event.is_set():
+                return
             backend = backends[backend_name]
             try:
                 if backend_name == "sonic":
@@ -292,14 +295,18 @@ def iter_query_search_ids(
                     env=search_backend_command_env(config=config),
                     cwd=CONSTANTS.DATA_DIR,
                     timeout=max(1, int(config.get("TIMEOUT", 60))) * 4,
+                    stop_event=stop_event,
                 )
-                for snapshot_id in ids:
-                    if snapshot_id in seen:
-                        continue
-                    seen.add(snapshot_id)
-                    yield snapshot_id
-                    if max_results and len(seen) >= max_results:
-                        return
+                try:
+                    for snapshot_id in ids:
+                        if snapshot_id in seen:
+                            continue
+                        seen.add(snapshot_id)
+                        yield snapshot_id
+                        if max_results and len(seen) >= max_results:
+                            return
+                finally:
+                    ids.close()
                 successful_backends += 1
             except Exception as err:
                 errors.append(err)

archivebox/search/views.py

@@ -6,7 +6,7 @@ import json
 import threading
 import time
 from copy import copy
-from queue import Full, Queue
+from queue import Empty, Full, Queue
 from urllib.parse import urlsplit
 from uuid import UUID
 
@@ -254,15 +254,18 @@ def iter_filtered_search_result_ids(iterator, queryset, *, flush_max_delay=0.05)
         yield from flush_batch()
 
 
-def iter_search_result_ids(query, base_queryset, *, search_mode, config):
+def iter_search_result_ids(query, base_queryset, *, search_mode, config, stop_event=None):
     """Yield filtered Snapshot IDs from the selected search provider."""
     search_mode_base = get_search_mode_base(search_mode, config=config)
     provider = (
         iter_meta_search_ids(query, base_queryset)
         if search_mode_base == "meta"
-        else iter_query_search_ids(query, search_mode=search_mode, config=config)
+        else iter_query_search_ids(query, search_mode=search_mode, config=config, stop_event=stop_event)
     )
-    yield from iter_filtered_search_result_ids(provider, base_queryset)
+    try:
+        yield from iter_filtered_search_result_ids(provider, base_queryset)
+    finally:
+        provider.close()
 
 
 def snapshot_search_stream_response(query, base_queryset, *, search_mode, config, cache_key, thread_name):
@@ -300,7 +303,7 @@ def snapshot_search_stream_response(query, base_queryset, *, search_mode, config
             iterator = None
             try:
                 close_old_connections()
-                iterator = iter_search_result_ids(query, base_queryset, search_mode=search_mode, config=config)
+                iterator = iter_search_result_ids(query, base_queryset, search_mode=search_mode, config=config, stop_event=stop_event)
                 for snapshot_id in iterator:
                     if stop_event.is_set():
                         break
@@ -326,10 +329,13 @@ def snapshot_search_stream_response(query, base_queryset, *, search_mode, config
                 emit(None)
 
         threading.Thread(target=run_search, name=thread_name, daemon=True).start()
-        yield f"0{stream_padding}\n"
         try:
+            yield f"0{stream_padding}\n"
             while True:
-                item = await asyncio.to_thread(queue.get)
+                try:
+                    item = await asyncio.to_thread(queue.get, True, 0.1)
+                except Empty:
+                    continue
                 if item is None:
                     break
                 if isinstance(item, BaseException):

archivebox/services/resource_admission.py

@@ -0,0 +1,91 @@
+"""Admission signals, not estimates of how much RAM an extractor will need."""
+
+import math
+import os
+import time
+from pathlib import Path
+
+
+RESOURCE_RECHECK_SECONDS = 2.0
+# Scheduling policy, not a minimum-RAM claim: stop increasing extraction
+# load when all non-idle work loses at least one second out of ten to memory
+# stalls. Require both recent and rolling pressure so transient reclaim does
+# not indefinitely starve otherwise viable work on small hosts.
+MEMORY_FULL_STALL_PERCENT = 10.0
+
+
+class ResourceAdmission:
+    def __init__(self):
+        self._memory_stalls: dict[str, tuple[int, float]] = {}
+        self._checked_at = float("-inf")
+        self._reason: str | None = None
+
+    @staticmethod
+    def cgroup_paths() -> list[Path]:
+        root = Path("/sys/fs/cgroup")
+        paths = [root]
+        try:
+            for line in Path("/proc/self/cgroup").read_text().splitlines():
+                if line.startswith("0::"):
+                    current = root / line[3:].lstrip("/")
+                    if current.is_dir() and current.is_relative_to(root):
+                        paths = [current, *[parent for parent in current.parents if parent.is_relative_to(root)]]
+                    break
+        except OSError:
+            pass
+        return paths
+
+    def snapshot_limit(self, configured: int) -> int:
+        cpus = len(os.sched_getaffinity(0)) if hasattr(os, "sched_getaffinity") else (os.cpu_count() or 1)
+        for path in self.cgroup_paths():
+            try:
+                quota, period = (path / "cpu.max").read_text().split()
+                if quota != "max":
+                    cpus = min(cpus, max(1, math.ceil(int(quota) / int(period))))
+            except (OSError, ValueError, ZeroDivisionError):
+                continue
+        # CPU quota limits parallel admission, not forward progress. A host
+        # with a fractional CPU must still be able to run one snapshot.
+        return max(1, min(configured, cpus))
+
+    def observe_memory_stalls(self, source: str, total: int, avg10: float, sampled_at: float) -> bool:
+        previous = self._memory_stalls.get(source)
+        self._memory_stalls[source] = (total, sampled_at)
+        if previous is None:
+            return avg10 >= MEMORY_FULL_STALL_PERCENT
+        previous_total, previous_time = previous
+        elapsed = sampled_at - previous_time
+        if elapsed <= 0 or total < previous_total:
+            return False
+        stalled_percent = (total - previous_total) / (elapsed * 1_000_000) * 100
+        return avg10 >= MEMORY_FULL_STALL_PERCENT and stalled_percent >= MEMORY_FULL_STALL_PERCENT
+
+    def memory_pressure(self) -> str | None:
+        now = time.monotonic()
+        if now - self._checked_at < RESOURCE_RECHECK_SECONDS:
+            return self._reason
+        self._checked_at = now
+        reasons = []
+        cgroups = self.cgroup_paths()
+        for path in cgroups:
+            try:
+                current = int((path / "memory.current").read_text())
+                for name in ("memory.high", "memory.max"):
+                    limit = (path / name).read_text().strip()
+                    if limit != "max" and current >= int(limit):
+                        reasons.append(f"{name} reached")
+            except (OSError, ValueError):
+                continue
+        for path in [Path("/proc/pressure/memory"), *[p / "memory.pressure" for p in cgroups]]:
+            try:
+                full = next(line for line in path.read_text().splitlines() if line.startswith("full "))
+                fields = dict(field.split("=", 1) for field in full.split()[1:])
+                if self.observe_memory_stalls(str(path), int(fields["total"]), float(fields["avg10"]), now):
+                    reasons.append("memory stalls")
+            except (OSError, ValueError, KeyError, StopIteration):
+                continue
+        self._reason = ", ".join(dict.fromkeys(reasons)) or None
+        return self._reason
+
+
+resource_admission = ResourceAdmission()

archivebox/services/runner.py

@@ -58,6 +58,8 @@ from abxbus import BaseEvent
 from abxbus.event_bus import EventBus, get_current_event, in_handler_context
 from abxbus.event_handler import EventHandlerAbortedError, EventHandlerCancelledError
 
+from archivebox.services.resource_admission import RESOURCE_RECHECK_SECONDS, resource_admission
+
 from archivebox.config.common import (
     ArchiveBoxBaseConfig,
     normalize_runtime_config,
@@ -221,6 +223,8 @@ class CrawlRunner:
         self.snapshot_tasks: dict[str, asyncio.Task[None]] = {}
         self.snapshot_semaphore = asyncio.Semaphore(1)
         self.max_concurrent_snapshots = 1
+        self._resource_deferred = False
+        self._initial_snapshot_admitted = False
         self.persona = None
         self.base_config: ArchiveBoxBaseConfig | dict[str, Any] = {}
         self.derived_config: dict[str, Any] = {}
@@ -294,6 +298,7 @@ class CrawlRunner:
     async def run(self) -> None:
         root_snapshot_id: str | None = None
         bus_destroyed = False
+        run_state_loaded = False
         try:
             first_signal_message = (
                 "\n[🛑] Got {signal_name}, aborting the active hook...\n"
@@ -313,8 +318,15 @@ class CrawlRunner:
                 on_signal=self._request_abort_from_signal,
                 raise_on_first_signal=False,
             ):
+                if await sync_to_async(defer_crawl_for_resources, thread_sensitive=True)(self.crawl):
+                    return
+                # Startup and its first snapshot are one admission. Browser
+                # startup can itself cause reclaim; checking again immediately
+                # would tear it down and repeat startup without doing any work.
+                self._initial_snapshot_admitted = True
+                run_state_loaded = True
                 snapshot_ids = await sync_to_async(self.load_run_state, thread_sensitive=True)()
-                max_concurrent_snapshots = max(1, int(self.base_config.get("CRAWL_MAX_CONCURRENT_SNAPSHOTS", 1)))
+                max_concurrent_snapshots = resource_admission.snapshot_limit(int(self.base_config.get("CRAWL_MAX_CONCURRENT_SNAPSHOTS", 1)))
                 self.max_concurrent_snapshots = max_concurrent_snapshots
                 self.snapshot_semaphore = asyncio.Semaphore(max_concurrent_snapshots)
                 live_ui = self._create_live_ui()
@@ -344,10 +356,19 @@ class CrawlRunner:
                 except Exception:
                     pass
                 self._live_stream = None
-            await sync_to_async(project_abxpkg_derived_cache_to_db, thread_sensitive=True)(self.base_config.get("ABXPKG_LIB_DIR"))
-            await sync_to_async(self.finalize_run_state, thread_sensitive=True)()
+            if run_state_loaded:
+                await sync_to_async(project_abxpkg_derived_cache_to_db, thread_sensitive=True)(self.base_config.get("ABXPKG_LIB_DIR"))
+                await sync_to_async(self.finalize_run_state, thread_sensitive=True)()
 
     async def enqueue_snapshot(self, snapshot_id: str, crawl_start_event: CrawlStartEvent | None = None) -> None:
+        if self._resource_deferred:
+            return
+        if not self._initial_snapshot_admitted and await sync_to_async(defer_crawl_for_resources, thread_sensitive=True)(
+            self.crawl,
+            excluded_snapshot_ids=list(self.snapshot_tasks),
+        ):
+            self._resource_deferred = True
+            return
         if await self.crawl_is_cancelled():
             return
         if await self.crawl_is_paused() and not self.allow_maintenance_on_inactive_crawl:
@@ -362,6 +383,7 @@ class CrawlRunner:
             return
         else:
             task = asyncio.create_task(self.run_snapshot(snapshot_id), context=_runner_task_context())
+        self._initial_snapshot_admitted = False
         self.snapshot_tasks[snapshot_id] = task
 
     async def stop_snapshot_tasks(self) -> None:
@@ -488,6 +510,8 @@ class CrawlRunner:
         from archivebox.core.models import Snapshot
         from archivebox.config.common import get_config
 
+        if self._resource_deferred:
+            return
         if not isinstance(get_current_event(), CrawlStartEvent):
             return
         if await self.crawl_is_cancelled():
@@ -497,7 +521,7 @@ class CrawlRunner:
 
         await sync_to_async(self.crawl.refresh_from_db, thread_sensitive=True)()
         config = await sync_to_async(lambda: get_config(crawl=self.crawl), thread_sensitive=True)()
-        self.max_concurrent_snapshots = max(1, int(config["CRAWL_MAX_CONCURRENT_SNAPSHOTS"]))
+        self.max_concurrent_snapshots = resource_admission.snapshot_limit(int(config["CRAWL_MAX_CONCURRENT_SNAPSHOTS"]))
 
         active_snapshot_ids = [snapshot_id for snapshot_id, task in self.snapshot_tasks.items() if not task.done()]
         available_slots = max(0, self.max_concurrent_snapshots - len(active_snapshot_ids))
@@ -1266,6 +1290,29 @@ def run_snapshot_maintenance(snapshot_id: str, *, output_dir: Path | None = None
     return True
 
 
+def defer_crawl_for_resources(crawl, *, excluded_snapshot_ids=()) -> bool:
+    """Leave extraction durable; never delay cancellation or final cleanup."""
+    from archivebox.core.models import Snapshot
+
+    if crawl.status not in (crawl.StatusChoices.QUEUED, crawl.StatusChoices.STARTED):
+        return False
+    reason = resource_admission.memory_pressure()
+    if reason is None:
+        return False
+    pending = crawl.snapshot_set.filter(status__in=Snapshot.RUNNABLE_STATES).exclude(id__in=excluded_snapshot_ids)
+    if crawl.snapshot_set.exists() and not pending.exists():
+        return False
+    now = timezone.now()
+    retry_at = now + timedelta(seconds=RESOURCE_RECHECK_SECONDS)
+    # Conditional short updates cannot resurrect a concurrent pause/cancel.
+    type(crawl).objects.filter(id=crawl.id, status__in=[crawl.StatusChoices.QUEUED, crawl.StatusChoices.STARTED]).update(
+        retry_at=retry_at,
+        modified_at=now,
+    )
+    pending.filter(retry_at__lte=now).update(retry_at=retry_at, modified_at=now)
+    return True
+
+
 def run_due_crawl(crawl, *, lock_seconds: int, interactive_interrupts: bool = False) -> bool:
     with crawl_lifecycle_lock(str(crawl.id)):
         return _run_due_crawl_locked(

archivebox/templates/core/snapshot.html

@@ -1890,14 +1890,13 @@
 
             function attachPreviewFrameHandlers(frame) {
                 resetPreviewFrameSize(frame)
-                if (frame.src.endsWith('.pdf')) {
+                if (new URL(frame.src || 'about:blank', window.location.href).pathname.endsWith('.pdf')) {
                     frame.removeAttribute('sandbox')
                     frame.src = frame.src + '#toolbar=0'
                 }
                 frame.onload = function() {
-                    if (this.src.includes('.pdf')) {
+                    if (new URL(this.src || 'about:blank', window.location.href).pathname.endsWith('.pdf')) {
                         this.removeAttribute('sandbox')
-                        this.src = this.src.split('?autoplay=')[0] + '#toolbar=0'
                     }
                     tryCenterImageFrame(this)
                     try {
@@ -2083,7 +2082,10 @@
                     nextSrcAbs = new URL(nextSrc, window.location.href).href
                 } catch (err) {}
                 const iframe_elem = ensureMainFrame(currentSrc !== nextSrcAbs)
-                if (target.endsWith('.pdf')) {
+                // Preview wrappers retain the .pdf path but add ?preview=1.
+                // The native PDF renderer must be allowed before navigation,
+                // not after a failed sandboxed load has already completed.
+                if (new URL(target, window.location.href).pathname.endsWith('.pdf')) {
                     iframe_elem.removeAttribute('sandbox')
                 } else {
                     iframe_elem.sandbox = "allow-same-origin allow-top-navigation-by-user-activation allow-scripts allow-forms"

archivebox/templates/core/system_warnings_banner.html

@@ -34,7 +34,7 @@
     {% elif mode == "high_load" %}
       15-min loadavg <code style="background:#000;color:#fff;padding:1px 6px;border-radius:3px;font:700 11px/1.3 ui-monospace,Menlo,Consolas,monospace;text-transform:none;">{{ load_15 }}</code>
       exceeds <code style="background:#000;color:#fff;padding:1px 6px;border-radius:3px;font:700 11px/1.3 ui-monospace,Menlo,Consolas,monospace;text-transform:none;">{{ load_threshold }}</code>
-      ({{ cpu_count }} cores × 3) — the host is saturated; crawls are queuing faster than the runner can finish them.
+      ({{ cpu_count }} cores × 3) — recent CPU or I/O contention may slow crawls. This average can stay elevated after load subsides; it does not measure queue growth.
     {% elif mode == "unsafe" %}
       ArchiveBox single-domain mode — archived pages share an origin with this site
       <a href="https://github.com/ArchiveBox/ArchiveBox/wiki/Security-Modes" target="_blank" rel="noopener noreferrer" style="color:#fff;text-decoration:underline;margin-left:6px;">why?</a>

archivebox/tests/test_api_v1_auth_browser_session.py

@@ -0,0 +1,56 @@
+from datetime import timedelta
+from urllib.parse import urlsplit
+
+import pytest
+from django.contrib.sessions.models import Session
+from django.utils import timezone
+
+from archivebox.api.models import APIToken
+from archivebox.tests.conftest import API_TEST_HOST, api_client_request
+
+pytestmark = pytest.mark.django_db(transaction=True)
+
+
+@pytest.mark.parametrize("header", ["HTTP_AUTHORIZATION", "HTTP_X_ARCHIVEBOX_API_KEY"])
+def test_browser_session_authenticates_admin_and_expires_with_key(client, api_admin_user, header):
+    token = APIToken.objects.create(created_by=api_admin_user, expires=timezone.now() + timedelta(minutes=5))
+    response = api_client_request(
+        client,
+        "post",
+        "/api/v1/auth/browser_session",
+        payload={},
+        headers={
+            "HTTP_HOST": API_TEST_HOST,
+            header: f"Bearer {token.token}" if header == "HTTP_AUTHORIZATION" else token.token,
+        },
+    )
+    assert response.status_code == 200, response.content
+    assert response["Cache-Control"] == "no-store"
+    data = response.json()
+    session = Session.objects.get(session_key=data["cookie"]["value"])
+    assert session.get_decoded()["_auth_user_id"] == str(api_admin_user.pk)
+    assert timezone.now() < session.expire_date <= token.expires
+    client.cookies.clear()
+    client.cookies[data["cookie"]["name"]] = data["cookie"]["value"]
+    response = client.get("/admin/", HTTP_HOST=urlsplit(data["admin_url"]).netloc)
+    assert response.status_code == 200, response.content
+    assert response.wsgi_request.user.pk == api_admin_user.pk
+
+
+@pytest.mark.parametrize("kind", ["missing", "invalid", "expired", "inactive", "nonadmin", "query"])
+def test_browser_session_rejects_invalid_credentials(client, api_admin_user, kind):
+    token = APIToken.objects.create(created_by=api_admin_user, expires=timezone.now() + timedelta(minutes=5))
+    if kind == "expired":
+        token.expires = timezone.now() - timedelta(seconds=1)
+        token.save()
+    if kind in ("inactive", "nonadmin"):
+        setattr(api_admin_user, "is_active" if kind == "inactive" else "is_superuser", False)
+        api_admin_user.save()
+    headers = {"HTTP_HOST": API_TEST_HOST}
+    if kind not in ("missing", "query"):
+        headers["HTTP_AUTHORIZATION"] = "Bearer " + ("invalid" if kind == "invalid" else token.token)
+    path = "/api/v1/auth/browser_session" + (f"?api_key={token.token}" if kind == "query" else "")
+    before = Session.objects.count()
+    response = api_client_request(client, "post", path, payload={}, headers=headers)
+    assert response.status_code in (401, 403), response.content
+    assert Session.objects.count() == before

archivebox/tests/test_cli_add.py

@@ -230,27 +230,23 @@ def test_add_single_url_records_url_in_crawl(initialized_archive):
     assert snapshots == []
 
 
-def test_add_memory_preflight_fails_before_creating_crawl(initialized_archive, capsys):
-    from archivebox.workers.supervisord_util import MIN_CRAWL_AVAILABLE_MEMORY_BYTES, require_crawl_memory
-
-    assert MIN_CRAWL_AVAILABLE_MEMORY_BYTES == 512 * 1024 * 1024
-    require_crawl_memory(MIN_CRAWL_AVAILABLE_MEMORY_BYTES)
-
+def test_add_bg_persists_queue_without_starting_extraction_workers(initialized_archive):
+    result = run_archivebox_cmd(
+        ["add", "--bg", "--plugins=wget", "--crawl-max-concurrent-snapshots=4", "https://example.com", "https://example.org"],
+        cwd=initialized_archive,
+        env=cli_env(),
+    )
+    assert result.returncode == 0, result.stderr
     with use_archivebox_db(initialized_archive):
-        before_crawls = Crawl.objects.count()
-        before_processes = Process.objects.count()
-
-        with pytest.raises(SystemExit) as err:
-            require_crawl_memory(MIN_CRAWL_AVAILABLE_MEMORY_BYTES - 1)
-
-        assert Crawl.objects.count() == before_crawls
-        assert Process.objects.count() == before_processes
-
-    assert err.value.code == 1
-    output = capsys.readouterr().err
-    assert "Not enough available memory to archive a crawl" in output
-    assert "No crawl, runner, or Sonic workers were started" in output
-    assert "configure swap" in output
+        crawl = Crawl.objects.get()
+        assert crawl.status == Crawl.StatusChoices.QUEUED
+        assert crawl.retry_at is not None
+        assert crawl.get_urls_list() == ["https://example.com", "https://example.org"]
+        assert crawl.config["PLUGINS"] == "wget"
+        assert crawl.config["CRAWL_MAX_CONCURRENT_SNAPSHOTS"] == 4
+        assert not Snapshot.objects.exists()
+        assert not ArchiveResult.objects.exists()
+        assert not Process.objects.filter(process_type__in=[Process.TypeChoices.ORCHESTRATOR, Process.TypeChoices.HOOK]).exists()
 
 
 def test_initial_crawl_creation_does_not_warn_as_an_outside_runner_update(initialized_archive, caplog):

archivebox/tests/test_resource_admission.py

@@ -0,0 +1,30 @@
+import os
+
+from archivebox.services.resource_admission import ResourceAdmission
+
+
+def test_cpu_admission_respects_process_affinity():
+    admission = ResourceAdmission()
+    available = len(os.sched_getaffinity(0)) if hasattr(os, "sched_getaffinity") else os.cpu_count()
+    assert 1 <= admission.snapshot_limit(100000) <= available
+    assert admission.snapshot_limit(1) == 1
+
+
+def test_memory_stall_counter_recovers_without_resetting_monitor():
+    # Real kernel PSI units: cumulative microseconds and ten-second average.
+    # Decision samples exercise the policy, not simulated hooks/processes.
+    admission = ResourceAdmission()
+    assert admission.observe_memory_stalls("host", 100, 0.0, 0.0) is False
+    assert admission.observe_memory_stalls("host", 400100, 20.0, 2.0) is True
+    assert admission.observe_memory_stalls("host", 400100, 20.0, 4.0) is False
+    assert admission.observe_memory_stalls("new-cgroup", 900, 20.0, 0.0) is True
+    assert admission.observe_memory_stalls("new-cgroup", 900, 20.0, 2.0) is False
+
+
+def test_transient_and_low_memory_stalls_do_not_starve_admission():
+    admission = ResourceAdmission()
+    for sample in range(20):
+        assert admission.observe_memory_stalls("host", sample * 2600, 0.13, sample * 2.0) is False
+    # One large interval alone is not sustained pressure; the rolling average
+    # must agree before deferring work.
+    assert admission.observe_memory_stalls("host", 1_000_000, 0.13, 40.0) is False

archivebox/tests/test_search.py

@@ -1289,3 +1289,75 @@ class TestSearchBackendsE2E:
             fixture_server.shutdown()
             fixture_server.server_close()
             fixture_thread.join(timeout=5)
+
+
+@pytest.mark.timeout(90)
+def test_live_public_search_disconnect_reaps_backend(initialized_archive):
+    """A real HTTP disconnect cancels a silent ripgrep process and its wrapper."""
+    import signal
+    from uuid import uuid4
+
+    import psutil
+
+    from archivebox.core.models import Snapshot
+    from archivebox.tests.test_orm_helpers import use_archivebox_db
+
+    rg = shutil.which("rg")
+    assert rg
+    query = "disconnect" + uuid4().hex
+    port = get_free_port()
+    env = cli_env(
+        live=True,
+        PUBLIC_INDEX="True",
+        PERMISSIONS="public",
+        SEARCH_BACKEND_ENGINE="ripgrep",
+        SEARCH_BACKEND_RIPGREP_ENABLED="True",
+        RIPGREP_BINARY=rg,
+        RIPGREP_ARGS_EXTRA='["--no-mmap", "--text", "--sort", "path"]',
+        RIPGREP_TIMEOUT="30",
+        ALLOWED_HOSTS="*",
+    )
+    created = run_archivebox_cmd(
+        ["snapshot", "create", "--status=sealed", "https://example.com/disconnect-test"],
+        cwd=initialized_archive,
+        env=env,
+    )
+    assert created.returncode == 0, created.stderr
+    with use_archivebox_db(initialized_archive):
+        snapshot = Snapshot.objects.get(url="https://example.com/disconnect-test")
+        directory = snapshot.output_dir / "dom"
+        directory.mkdir(parents=True, exist_ok=True)
+        (directory / "a.txt").write_text(query + "\n")
+        with (directory / "z.txt").open("wb") as large:
+            large.truncate(8 * 1024**3)
+            for offset in range(0, 8 * 1024**3, 1024**2):
+                large.seek(offset)
+                large.write(b"\n")
+    server = start_archivebox_server(initialized_archive, port=port, env=env, daemonize=False, log_name="disconnect-server.log")
+    children = []
+    try:
+        with requests.get(
+            f"http://127.0.0.1:{port}/public/search-stream/",
+            params={"q": query, "search_mode": "deep:ripgrep"},
+            headers={"Host": f"web.archivebox.localhost:{port}"},
+            stream=True,
+            timeout=(5, 10),
+        ) as response:
+            assert response.status_code == 200, response.text
+            lines = response.iter_lines()
+            assert int(next(lines).strip()) == 0
+            assert int(next(lines).strip()) == 1
+            children = [child for child in psutil.Process(server.pid).children(recursive=True) if query in " ".join(child.cmdline())]
+            assert len(children) == 2
+            for child in children:
+                if Path(child.exe()).name == Path(rg).name:
+                    child.send_signal(signal.SIGSTOP)
+        _, alive = psutil.wait_procs(children, timeout=5)
+        assert not alive, [(child.pid, child.status()) for child in alive]
+    finally:
+        for child in children:
+            try:
+                child.kill()
+            except psutil.NoSuchProcess:
+                pass
+        stop_archivebox_process(server)

archivebox/tests/test_ui_admin_crawl.py

@@ -14,6 +14,7 @@ pytestmark = pytest.mark.django_db
 
 
 class TestCrawlScheduleAdmin:
+    @pytest.mark.django_db(transaction=True)
     def test_crawlschedule_change_view_renders_and_saves(self, client, admin_user, crawl):
         from archivebox.crawls.models import CrawlSchedule
 
@@ -35,24 +36,33 @@ class TestCrawlScheduleAdmin:
         assert b"No Crawls yet..." not in get_response.content
         assert b"No Snapshots yet..." not in get_response.content
 
-        post_response = client.post(
-            change_url,
-            {
-                "label": "Morning crawl",
-                "notes": "updated",
-                "schedule": "0 8 * * *",
-                "template": str(crawl.pk),
-                "created_by": str(admin_user.pk),
-                "_save": "Save",
-            },
-            HTTP_HOST=ADMIN_TEST_HOST,
-        )
+        from django.db import connection
+        from django.test.utils import CaptureQueriesContext
+
+        previous_mode = getattr(connection, "transaction_mode", None)
+        with CaptureQueriesContext(connection) as queries:
+            post_response = client.post(
+                change_url,
+                {
+                    "label": "Morning crawl",
+                    "notes": "updated",
+                    "schedule": "0 8 * * *",
+                    "template": str(crawl.pk),
+                    "created_by": str(admin_user.pk),
+                    "_save": "Save",
+                },
+                HTTP_HOST=ADMIN_TEST_HOST,
+            )
+        if connection.vendor == "sqlite":
+            assert any(query["sql"] == "BEGIN IMMEDIATE" for query in queries)
+            assert connection.transaction_mode == previous_mode
 
         assert post_response.status_code == 302
         schedule.refresh_from_db()
         assert schedule.label == "Morning crawl"
         assert schedule.notes == "updated"
         assert schedule.schedule == "0 8 * * *"
+        assert schedule.is_enabled is False
         assert schedule.template_id == crawl.pk
         assert schedule.created_by_id == admin_user.pk
 

archivebox/tests/test_ui_live_progress.py

@@ -104,6 +104,74 @@ def real_crawl_setup_process(snapshot, hermetic_lib_dir):
 
 
 class TestLiveProgressView:
+    def test_collection_summary_reports_stored_sizes(self, client, admin_user, snapshot):
+        import time
+        from django.core.cache import cache
+        from django.db.models import Count, Sum
+        from archivebox.core.models import Snapshot
+
+        Snapshot.objects.filter(pk=snapshot.pk).update(output_size=12345)
+        expected = Snapshot.objects.aggregate(snapshots=Count("*"), bytes=Sum("output_size"))
+        cache.delete_many(["progress-collection:all", "progress-collection:all:refresh"])
+        client.force_login(admin_user)
+        deadline = time.monotonic() + 5
+        while True:
+            response = client.get(reverse("live_progress"), {"collection": "1"}, HTTP_HOST=ADMIN_TEST_HOST)
+            assert response.status_code == 200
+            summary = response.json().get("collection")
+            if summary is not None or time.monotonic() >= deadline:
+                break
+            time.sleep(0.05)
+        assert summary is not None
+        assert summary["snapshots"] == expected["snapshots"]
+        assert summary["bytes"] == expected["bytes"]
+        from django.db import connection
+        from django.test.utils import CaptureQueriesContext
+
+        if connection.vendor == "sqlite":
+            with connection.cursor() as cursor:
+                cursor.execute("EXPLAIN QUERY PLAN SELECT COUNT(*), SUM(output_size) FROM core_snapshot")
+                plan = " ".join(str(row) for row in cursor.fetchall())
+            assert "COVERING INDEX" in plan, plan
+
+        timings = []
+        with CaptureQueriesContext(connection) as queries:
+            for _ in range(20):
+                start = time.perf_counter()
+                response = client.get(reverse("live_progress"), {"collection": "1"}, HTTP_HOST=ADMIN_TEST_HOST)
+                timings.append((time.perf_counter() - start) * 1000)
+                assert response.status_code == 200
+                assert response.json()["collection"] == summary
+        assert not any('AS "bytes"' in query["sql"] for query in queries)
+        print(f"cached progress response: median={sorted(timings)[10]:.1f}ms max={max(timings):.1f}ms")
+
+    def test_collection_summary_is_scoped_to_staff_owner(self, client, admin_user, snapshot):
+        import time
+        from django.core.cache import cache
+        from archivebox.core.models import Snapshot
+
+        admin_user.is_superuser = False
+        admin_user.save(update_fields=["is_superuser"])
+        from django.contrib.auth import get_user_model
+
+        # Give the snapshot another owner; staff must not learn its size/count.
+        snapshot.crawl.created_by = get_user_model().objects.create_user(username="other-summary-owner")
+        snapshot.crawl.save(update_fields=["created_by"])
+        Snapshot.objects.filter(pk=snapshot.pk).update(output_size=12345)
+        cache.delete_many([f"progress-collection:{admin_user.pk}", f"progress-collection:{admin_user.pk}:refresh"])
+        client.force_login(admin_user)
+        deadline = time.monotonic() + 5
+        while True:
+            response = client.get(reverse("live_progress"), {"collection": "1"}, HTTP_HOST=ADMIN_TEST_HOST)
+            assert response.status_code == 200
+            summary = response.json().get("collection")
+            if summary is not None or time.monotonic() >= deadline:
+                break
+            time.sleep(0.05)
+        assert summary is not None
+        assert summary["snapshots"] == 0
+        assert summary["bytes"] == 0
+
     def test_live_progress_rejects_unauthenticated_unscoped_request(self, client):
         response = client.get(reverse("live_progress"), HTTP_HOST=ADMIN_TEST_HOST)
 

archivebox/tests/test_urls.py

@@ -1431,7 +1431,7 @@ class TestUrlRouting:
             [
                 "add",
                 "--depth=0",
-                "--plugins=consolelog,screenshot,chrome_mhtml",
+                "--plugins=consolelog,screenshot,pdf,chrome_mhtml",
                 console_source_url,
             ],
             cwd=self.data_dir,
@@ -1475,6 +1475,14 @@ class TestUrlRouting:
             resp = client.get(f"/{screenshot_rel}?preview=1", HTTP_HOST=snapshot_host)
             assert resp.status_code == 200
             assert resp["Content-Type"].startswith("text/html")
+            assert f"frame-ancestors 'self' http://{web_host}" in resp["Content-Security-Policy"]
+
+            resp = client.get("/pdf/output.pdf?preview=1", HTTP_HOST=snapshot_host)
+            assert resp.status_code == 200
+            pdf_preview = response_body(resp).decode("utf-8")
+            assert '<iframe ' in pdf_preview
+            assert '<embed ' not in pdf_preview
+            assert "object-src 'none'" in resp["Content-Security-Policy"]
 
             root_screenshot = screenshot_file.read_bytes()
             import shutil
@@ -1507,6 +1515,58 @@ class TestUrlRouting:
             """.replace("__CONSOLE_MARKER__", console_marker),
         )
 
+        # Exercise the same real captured files with each request-time posture,
+        # including both auto branches and an HTTPS canonical reverse-proxy URL.
+        for mode, base in (
+            ("auto", "http://archivebox.localhost:8000"),
+            ("auto", "http://archivebox.example:8000"),
+            ("safe-subdomains-fullreplay", "https://archivebox.example"),
+            ("safe-onedomain-nojsreplay", "http://archivebox.example:8000"),
+            ("unsafe-onedomain-noadmin", "http://archivebox.example:8000"),
+            ("danger-onedomain-fullreplay", "http://archivebox.example:8000"),
+        ):
+            self._run(
+                """
+                from urllib.parse import urlsplit
+                from archivebox.config.common import get_request_config
+                snapshot = get_snapshot()
+                client = Client()
+                for path in ('screenshot/screenshot.png', 'pdf/output.pdf', 'chrome_mhtml/snapshot.mhtml'):
+                    target = urlsplit(build_snapshot_url(str(snapshot.id), path))
+                    response = client.get(target.path + '?preview=1', HTTP_HOST=target.netloc)
+                    assert response.status_code == 200, (target, response.status_code)
+                    csp = response['Content-Security-Policy']
+                    config = get_request_config(response.wsgi_request)
+                    origins = tuple(dict.fromkeys((get_web_base_url(config=config), get_admin_base_url(config=config))))
+                    ancestors = next(d.strip() for d in csp.split(';') if d.strip().startswith('frame-ancestors'))
+                    assert ancestors == "frame-ancestors 'self' " + ' '.join(origins), ancestors
+                    assert '*' not in ancestors  # no unrelated snapshots/sites
+                    assert "object-src 'none'" in csp
+                    assert "form-action 'none'" in csp
+                    assert "base-uri 'none'" in csp
+                    assert "script-src 'self'" in csp  # trusted viewer can run
+                    if path.endswith('.mhtml'):
+                        assert 'sandbox="allow-popups allow-popups-to-escape-sandbox allow-same-origin"' in response_body(response).decode()
+
+                dangerous, _, _ = write_replay_fixtures(snapshot)
+                target = urlsplit(build_snapshot_url(str(snapshot.id), dangerous))
+                response = client.get(target.path, HTTP_HOST=target.netloc)
+                assert response.status_code == 200
+                config = get_request_config(response.wsgi_request)
+                csp = response.headers.get('Content-Security-Policy', '')
+                assert ("script-src 'none'" in csp) == config.SHOULD_NEUTER_RISKY_REPLAY
+                assert ('sandbox;' in csp) == config.SHOULD_NEUTER_RISKY_REPLAY
+                admin = client.get('/admin/login/', HTTP_HOST=get_admin_host())
+                assert admin.status_code == (200 if config.CONTROL_PLANE_ENABLED else 403)
+                if not config.CONTROL_PLANE_ENABLED:
+                    assert client.get('/api/v1/docs', HTTP_HOST=get_api_host()).status_code == 403
+                    assert client.post('/public/', HTTP_HOST=get_web_host()).status_code == 403
+                print('OK')
+                """,
+                mode=mode,
+                env_overrides={"BASE_URL": base},
+            )
+
     def test_api_available_on_admin_and_api_hosts(self) -> None:
         self._run(
             """

archivebox/workers/supervisord_util.py

@@ -47,7 +47,6 @@ _SUPERVISORD_ERRORS = (XmlRpcError, OSError, RuntimeError, TimeoutError)
 _PROCESS_STATE_ERRORS = (DatabaseError, OSError, RuntimeError, ValueError, psutil.Error)
 MIN_SERVER_WORKER_AVAILABLE_MEMORY_BYTES = 256 * 1024 * 1024
 MIN_DEPENDENCY_INSTALL_AVAILABLE_MEMORY_BYTES = MIN_SERVER_WORKER_AVAILABLE_MEMORY_BYTES
-MIN_CRAWL_AVAILABLE_MEMORY_BYTES = 512 * 1024 * 1024
 
 
 def _shell_join(args: list[str]) -> str:
@@ -118,16 +117,6 @@ def require_dependency_install_memory(available_bytes: int | None = None) -> Non
     )
 
 
-def require_crawl_memory(available_bytes: int | None = None) -> None:
-    available_bytes = effective_available_memory_bytes() if available_bytes is None else available_bytes
-    _require_available_memory(
-        "archive a crawl",
-        "    No crawl, runner, or Sonic workers were started.",
-        available_bytes,
-        MIN_CRAWL_AVAILABLE_MEMORY_BYTES,
-    )
-
-
 def archivebox_cmd(*args: str) -> list[str]:
     return [str(resolve_env_binary("archivebox")), *args]
 
@@ -1280,11 +1269,18 @@ def tail_multiple_worker_logs(log_files: list[str], follow=True, proc=None, keep
 
     print()
 
+    # Log display needs a short refresh interval, but each ownership check
+    # queries the database and validates OS processes. Check leadership once
+    # per second even when workers continuously produce log output.
+    next_ownership_check = 0.0
     try:
         while follow:
-            if keep_running is not None and not keep_running():
-                print("\n[newer ArchiveBox process is now running the orchestrator and server]")
-                return "transferred"
+            now = time.monotonic()
+            if keep_running is not None and now >= next_ownership_check:
+                if not keep_running():
+                    print("\n[newer ArchiveBox process is now running the orchestrator and server]")
+                    return "transferred"
+                next_ownership_check = time.monotonic() + 1.0
 
             # Check if the monitored process has exited
             if proc is not None and proc.poll() is not None:

bin/collect_ui_screenshots.sh

@@ -193,6 +193,22 @@ echo "[*] Creating a temporary screenshot admin"
 )
 CREATED_TEMP_USER=1
 
+# Configure the optional UI before the server snapshots its startup config.
+OPENCODE_PORT="$(uv run --no-cache --project "$REPO_DIR" python - <<'PYPORT'
+import socket
+
+with socket.socket() as sock:
+    sock.bind(("127.0.0.1", 0))
+    print(sock.getsockname()[1])
+PYPORT
+)"
+(
+    cd "$DATA_DIR"
+    uv run --no-cache --project "$REPO_DIR" archivebox config --set \
+        OPENCODE_ENABLED=True "OPENCODE_PORT=$OPENCODE_PORT" TLSNOTARY_ENABLED=True
+    uv run --no-cache --project "$REPO_DIR" archivebox install opencode --binproviders=env,pnpm
+)
+
 echo "[*] Starting ArchiveBox on port $PORT"
 (
     cd "$DATA_DIR"
@@ -248,6 +264,10 @@ while [[ "$capture_index" -lt "${#VIEWS[@]}" ]]; do
     capture_index=$((capture_index + 1))
     slug="$(printf '%s' "$name" | tr '[:upper:]' '[:lower:]' | tr -cs '[:alnum:]' '-' | sed 's/^-//; s/-$//')"
     expected_plugin=""
+    expected_frame_text=""
+    if [[ "$capture_mode" == wait-frame-text:* ]]; then
+        expected_frame_text="${capture_mode#wait-frame-text:}"
+    fi
     if [[ "$name" == "Snapshot View ("*")" && "$capture_mode" != "live-progress" && "$capture_mode" != "snapshot-collapsed" ]]; then
         expected_plugin="${name#Snapshot View (}"
         expected_plugin="${expected_plugin%)}"
@@ -407,6 +427,7 @@ PY
                     SCREENSHOT_VARIANTS_JSON="$output_variants" \
                     SCREENSHOT_COLLAPSE_FILTERS=1 \
                     SCREENSHOT_EXPECT_PLUGIN="$expected_plugin" \
+                    SCREENSHOT_EXPECT_FRAME_TEXT="$expected_frame_text" \
                     node "$REPO_DIR/bin/take_screenshot.js" "$url" "$screenshot_path" >"$capture_dir/report.json"
                 view_timing_report="$capture_dir/report.json"
                 timing_report_path="$view_timing_report"
@@ -494,21 +515,6 @@ PY
             CREATE_WEBHOOK="$CREATE_WEBHOOK" \
             node "$REPO_DIR/bin/setup_ui_screenshot_data.js" "$ADMIN_BASE_URL" "$USERNAME"
 
-        OPENCODE_PORT="$(uv run --no-cache --project "$REPO_DIR" python - <<'PY'
-import socket
-
-with socket.socket() as sock:
-    sock.bind(("127.0.0.1", 0))
-    print(sock.getsockname()[1])
-PY
-)"
-        (
-            cd "$DATA_DIR"
-            OPENCODE_PORT="$OPENCODE_PORT" uv run --no-cache --project "$REPO_DIR" archivebox manage shell --no-imports -c \
-                'import os; from archivebox.machine.models import Machine; Machine.from_json({"config": {"OPENCODE_ENABLED": True, "OPENCODE_PORT": int(os.environ["OPENCODE_PORT"])}})'
-            uv run --no-cache --project "$REPO_DIR" archivebox install opencode --binproviders=env,pnpm
-        )
-
         SWEETING_CAPTURE_STARTED_AT="$( (
             cd "$DATA_DIR"
             uv run --no-cache --project "$REPO_DIR" archivebox manage shell --no-imports -c \
@@ -561,7 +567,7 @@ PY
         VIEWS+=(
             "Add URLs|$ADMIN_BASE_URL/add/|/add/|archivebox/core/views.py"
             "Admin dashboard|$ADMIN_BASE_URL/admin/|/admin/|archivebox/core/admin_site.py"
-            "AI agent|$ADMIN_BASE_URL/admin/agent/|/admin/agent/|abx_plugins/plugins/opencode/views.py|wait-text:ArchiveBox AI Agent"
+            "AI agent|$ADMIN_BASE_URL/admin/agent/|/admin/agent/|archivebox/opencode/views.py|wait-text:ArchiveBox AI Agent"
             "Snapshots table|$ADMIN_BASE_URL/admin/core/snapshot/|/admin/core/snapshot/|archivebox/core/admin_snapshots.py"
             "Snapshots grid|$ADMIN_BASE_URL/admin/core/snapshot/grid/|/admin/core/snapshot/grid/|archivebox/templates/admin/snapshots_grid.html"
             "Snapshot admin detail|$ADMIN_BASE_URL/admin/core/snapshot/$SNAPSHOT_ID/change/|/admin/core/snapshot/$SNAPSHOT_ID/change/|archivebox/core/admin_snapshots.py"
@@ -625,11 +631,17 @@ PY
             SCREENSHOT_HEIGHT=1000 \
             node "$REPO_DIR/bin/take_screenshot.js" "$LIVE_SNAPSHOT_VIEW_URL" "$CAPTURE_ROOT/snapshot-output-discovery.png" >"$SNAPSHOT_DISCOVERY_REPORT"
         SNAPSHOT_OUTPUT_PLUGINS="$(UI_SCREENSHOT_DISCOVERY_REPORT="$SNAPSHOT_DISCOVERY_REPORT" uv run --no-cache --project "$REPO_DIR" python -c \
-            'import json, os; from urllib.parse import urlsplit; report=json.load(open(os.environ["UI_SCREENSHOT_DISCOVERY_REPORT"])); print("\n".join("{}\t{}".format(output["plugin"], "wait-replay:Nick Sweeting" if urlsplit(output["previewUrl"]).path.endswith(".wacz") else "") for output in report["checks"]["snapshotOutputs"]))')"
+            'import json, os; from urllib.parse import urlsplit; report=json.load(open(os.environ["UI_SCREENSHOT_DISCOVERY_REPORT"])); print("\n".join("{}\t{}".format(output["plugin"], "wait-frame-text:Verified · signature and archived response match" if output["plugin"] == "tlsnotary" else "wait-replay:Nick Sweeting" if urlsplit(output["previewUrl"]).path.endswith(".wacz") else "") for output in report["checks"]["snapshotOutputs"]))')"
         if [[ -z "$SNAPSHOT_OUTPUT_PLUGINS" ]]; then
             echo "[!] The Sweeting.me snapshot detail page exposed no selectable outputs" >&2
             exit 1
         fi
+        if ! printf '%s\n' "$SNAPSHOT_OUTPUT_PLUGINS" | cut -f1 | grep -qx tlsnotary; then
+            echo "[!] Live capture did not expose the required TLSNotary verifier output" >&2
+            tail -200 "$CAPTURE_ROOT/sweeting-live-capture.log" >&2
+            cat "$SNAPSHOT_DISCOVERY_REPORT" >&2
+            exit 1
+        fi
         while IFS=$'\t' read -r plugin_name output_capture_mode; do
             [[ -z "$plugin_name" ]] && continue
             VIEWS+=("Snapshot View ($plugin_name)|$LIVE_SNAPSHOT_VIEW_URL#$plugin_name|/|archivebox/templates/core/snapshot.html|$output_capture_mode")

bin/take_screenshot.js

@@ -32,6 +32,7 @@ Environment:
   SCREENSHOT_SNAPSHOT_VIEW   Set to list or grid before loading the page
   SCREENSHOT_SNAPSHOT_HEADER Set to expanded or collapsed before loading a snapshot detail page
   SCREENSHOT_EXPECT_PLUGIN   Require this snapshot output plugin to be selected
+  SCREENSHOT_EXPECT_FRAME_TEXT  Require this text in the selected output frame
   SCREENSHOT_EXPECT_LIVE_PROGRESS  Require real progress bars and a loaded screencast frame
   SCREENSHOT_COLLAPSE_FILTERS Set to 1 to keep admin filters out of screenshots
   SCREENSHOT_RESET_FILTERS   Set to 1 to clear the admin filter collapsed preference
@@ -231,6 +232,12 @@ async function main() {
           }
         });
       }, { timeout: 45000, polling: 250 }).catch(() => {});
+      if (process.env.SCREENSHOT_EXPECT_FRAME_TEXT) {
+        const frameElement = await page.$('#main-frame');
+        const frame = await frameElement.contentFrame();
+        await frame.waitForFunction((text) => document.body.innerText.includes(text),
+          { timeout: 45000 }, process.env.SCREENSHOT_EXPECT_FRAME_TEXT);
+      }
     }
 
     if (process.env.SCREENSHOT_EXPECT_LIVE_PROGRESS === '1') {

docs/Configuration.md

@@ -411,6 +411,8 @@ ArchiveBox splits its surfaces across three logical hosts: `admin.*` (Django adm
 
 SingleFile output is served as ordinary HTML in every mode; it remains usable in no-JS modes because SingleFile removes the page's scripts during capture. ArchiveWeb.page/ReplayWeb.page and MHTML use their existing trusted preview templates. ArchiveBox always attempts to load these viewers, including when the incoming request is plain HTTP, because HTTPS may be terminated by an upstream proxy. Browser service-worker rules still require ReplayWeb.page to be reached through HTTPS or localhost for replay to initialize.
 
+Trusted preview wrappers allow embedding by their own origin and the configured web/admin origins, so the collection UI can display snapshot-host previews without opening them to arbitrary embedding sites. PDF previews use a document frame rather than an object/embed, retaining `object-src 'none'` on the wrapper. Allowing a trusted viewer's scripts does not exempt the archived documents inside it from their replay policy; the MHTML viewer keeps its archived child document sandboxed without scripts.
+
 > [!WARNING]
 > Switching to any mode whose name starts with `unsafe-` or `danger-` is logged at startup and surfaces a banner in the UI. **Don't use these modes on a public hostname** — archived JavaScript will run on the same origin as your admin session.
 

etc/package.json

@@ -1,6 +1,6 @@
 {
   "name": "archivebox",
-  "version": "0.9.35rc449",
+  "version": "0.9.35rc460",
   "repository": "github:ArchiveBox/ArchiveBox",
   "license": "MIT",
   "dependencies": {

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "archivebox"
-version = "0.9.35rc449"
+version = "0.9.35rc460"
 requires-python = ">=3.13"
 description = "Self-hosted internet archiving solution."
 authors = [{name = "Nick Sweeting", email = "pyproject.toml@archivebox.io"}]
@@ -79,8 +79,8 @@ dependencies = [
     ### Binary/Package Management
     "abxbus==2.5.56",                 # direct imports only; version constrained by abx-dl -> abx-plugins
     "abxpkg==1.12.117",                # direct imports only; version constrained by abx-dl -> abx-plugins
-    "abx-plugins[opencode]==1.12.236", # optional AI runtime clients are owned by the plugin
-    "abx-dl==1.12.274",       # shared ArchiveBox downloader package
+    "abx-plugins[opencode]==1.12.241", # optional AI runtime clients are owned by the plugin
+    "abx-dl==1.12.281",       # shared ArchiveBox downloader package
     ### UUID7 backport for Python <3.14
     "uuid7>=0.1.0; python_version < '3.14'",  # provides the uuid_extensions module on Python 3.13
     "browser-cookie3>=0.20.1",
@@ -342,7 +342,7 @@ Donate = "https://github.com/ArchiveBox/ArchiveBox/wiki/Donations"
 
 
 [tool.bumpver]
-current_version = "v0.9.35rc449"
+current_version = "v0.9.35rc460"
 version_pattern = "vMAJOR.MINOR.PATCH[PYTAGNUM]"
 commit_message = "bump version {old_version} -> {new_version}"
 tag_message = "{new_version}"

uv.lock

@@ -13,18 +13,18 @@ supported-markers = [
 ]
 
 [options]
-exclude-newer = "2026-09-11T22:03:43.68620639Z"
+exclude-newer = "2026-09-12T23:27:00.678786196Z"
 exclude-newer-span = "P5D"
 
 [options.exclude-newer-package]
-abxbus = { timestamp = "2026-09-16T22:03:42.686219334Z", span = "PT1S" }
+abxbus = { timestamp = "2026-09-17T23:26:59.678836159Z", span = "PT1S" }
 abx-plugins = "2100-01-01T00:00:00Z"
 abx-dl = "2100-01-01T00:00:00Z"
 abxpkg = "2100-01-01T00:00:00Z"
 
 [[package]]
 name = "abx-dl"
-version = "1.12.274"
+version = "1.12.281"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "abx-plugins", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
@@ -38,14 +38,14 @@ dependencies = [
     { name = "rich", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "rich-click", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/c6/69/223fb0cfd106eda015a1a73daa58fc9aa87bb7b974094c5eadc65c15b693/abx_dl-1.12.274.tar.gz", hash = "sha256:58b7f85aec60db8233d4ea33a7683edf12be63fe32b8eccbd45d57edcab1a35b", size = 87257 }
+sdist = { url = "https://files.pythonhosted.org/packages/38/9d/8e20554592e09e4db3d76a0d85cd03971a4c9cb44210610c35bd8a8b4515/abx_dl-1.12.281.tar.gz", hash = "sha256:37bcf7b38d85776d9e727e4e65e04d72ce355cf15efebc6e933a9c9df51f18d6", size = 88246 }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/ab/36/b12628d4ce600cbcd1c3ba87c6f76721a1c21e074e1f30782baf2eb2eb06/abx_dl-1.12.274-py3-none-any.whl", hash = "sha256:5481d25ab82da2449a3a0a47079fd37a8768f1175ed2068bef45392beb4ee684", size = 92246 },
+    { url = "https://files.pythonhosted.org/packages/8f/e7/881a49809fd6a4f6bdcbfa09b47bdf3feb61e7e52faeb2a51e17d5b41840/abx_dl-1.12.281-py3-none-any.whl", hash = "sha256:4428ae92627819a9fa3d63bd601c460008aa2148aca92cf614e077e234e1d2db", size = 93261 },
 ]
 
 [[package]]
 name = "abx-plugins"
-version = "1.12.236"
+version = "1.12.241"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "abxbus", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
@@ -54,9 +54,9 @@ dependencies = [
     { name = "rich-click", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
     { name = "uv", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/11/90/617fe56f24af0ed1a6bcc255233703c9378f49d381d2192eb76a0a1c9ad7/abx_plugins-1.12.236.tar.gz", hash = "sha256:1c0a23a4878787a64109c69160c71df74885c90e5fca9def7c13b8ea731b1add", size = 269546, upload-time = "2026-09-16T21:51:20.39Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/00/aa/b5a07dd20c60635b57c1fd11c325b47d38a0d2c2a1e203fadbf64933148a/abx_plugins-1.12.241.tar.gz", hash = "sha256:001ca84516b0e51169efb1cf87b73b8504bea4fad61c8bf40b7ee53a6cee0b9b", size = 293160, upload-time = "2026-09-17T23:05:59.222Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/6c/ed/ac9682624eea2604cb3d35ac6f952c92c95e52c064401044d43c3df5864a/abx_plugins-1.12.236-py3-none-any.whl", hash = "sha256:0bf8738c64d27dd88d2ac4e91c4c85504352cc1a47bdc9aa863a12a9b7f8a5da", size = 424591, upload-time = "2026-09-16T21:51:21.816Z" },
+    { url = "https://files.pythonhosted.org/packages/fc/b9/94c0317de5ab364a91c1e2cdd16951022d37da3f2a2c6020b8f79a35f82f/abx_plugins-1.12.241-py3-none-any.whl", hash = "sha256:27170b2a155d7dcda43e80bc5e5209583a045b3b9896ca57c5b8085f24d6e9e1", size = 457274, upload-time = "2026-09-17T23:05:57.452Z" },
 ]
 
 [package.optional-dependencies]
@@ -127,7 +127,7 @@ wheels = [
 
 [[package]]
 name = "archivebox"
-version = "0.9.35rc449"
+version = "0.9.35rc460"
 source = { editable = "." }
 dependencies = [
     { name = "abx-dl", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
@@ -226,8 +226,8 @@ dev = [
 
 [package.metadata]
 requires-dist = [
-    { name = "abx-dl", specifier = "==1.12.274" },
-    { name = "abx-plugins", extras = ["opencode"], specifier = "==1.12.236" },
+    { name = "abx-dl", specifier = "==1.12.281" },
+    { name = "abx-plugins", extras = ["opencode"], specifier = "==1.12.241" },
     { name = "abxbus", specifier = "==2.5.56" },
     { name = "abxpkg", specifier = "==1.12.117" },
     { name = "archivebox", extras = ["sonic", "ldap", "debug"], marker = "extra == 'all'" },