Compare commits

...
Author SHA1 Message Date
ksmithandClaude Sonnet 5 f2ee793cf8 upgrade.sh: auto-bump EVECTION_PWA_CACHE_VERSION on every deploy
The release-notes reminder to manually bump this and restart gunicorn
was easy to miss or forget, leaving installed PWAs on a stale cached
shell/pages after an upgrade that changed static assets. upgrade.sh
now sets it to the deployed tag itself before restarting services, so
there's no manual step and no cache-invalidation window to miss.
install.sh does the same on first install for consistency.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-10-31 12:08:05 -06:00
ksmithandClaude Sonnet 5 b7d9ee672c Add an atomic per-feed claim to poll_due_feeds (T1)
poll_due_feeds scanned for due feeds and dispatched fetch_feed without
marking them claimed, so overlapping beat ticks or a slow-running
fetch could dispatch the same feed twice, racing on ETag/status/error-
count writes. Feed.claimed_at is now set via an UPDATE ... WHERE
claimed_at IS NULL OR claimed_at < stale-cutoff, so only one caller's
UPDATE can ever match a given row; fetch_feed clears it when done.
A claim older than CLAIM_STALE_MINUTES is treated as abandoned so a
worker dying mid-fetch doesn't wedge the feed forever.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-10-31 12:03:18 -06:00
ksmithandClaude Sonnet 5 d7b433e580 Purge the service worker's authenticated-page cache on logout (S3)
PAGES_CACHE held every cached navigation response (including
authenticated article pages) with no logout-time purge, so a later
offline navigation on a shared or account-switched device could still
serve another user's cached pages. Logout is a plain POST, so the
fetch handler now special-cases that request: it deletes PAGES_CACHE
and lets the real logout request through, with no race against the
navigation that follows since it's synchronous with the same request.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-10-31 12:03:10 -06:00
ksmithandClaude Sonnet 5 b52ba2be8a Make first-run setup race-safe with an atomic SetupLock claim (S4)
The setup view's User.objects.exists() check and the admin account's
save() weren't in the same transaction, so two concurrent requests
during first deployment could both pass the check and both create an
admin account. SetupLock is a singleton row (pk=1); claiming it and
creating the admin now happen in one atomic block, so only the first
of any racing requests can succeed — the rest hit IntegrityError and
are sent to the login page instead.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-10-31 12:03:05 -06:00
ksmithandClaude Sonnet 5 56eb2bc498 Fail closed in production if SECRET_KEY is missing or the placeholder default (S2)
base.py's env.str() fallback is a publicly known string in this repo's
history. Running prod settings with it (an unset or forgotten .env
value) would sign sessions/CSRF tokens with a key anyone can read.
install.sh already generates a real random key for fresh installs, so
this only blocks the broken case.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-10-31 12:02:58 -06:00
14 changed files with 235 additions and 21 deletions

No files matched your search

@@ -0,0 +1,22 @@
# Generated by Django 5.2.18 on 2026-10-31 18:00
from django.db import migrations, models
class Migration(migrations.Migration):
dependencies = [
("accounts", "0001_initial"),
]
operations = [
migrations.CreateModel(
name="SetupLock",
fields=[
(
"id",
models.PositiveSmallIntegerField(default=1, primary_key=True, serialize=False),
),
],
),
]
+12
View File
@@ -5,3 +5,15 @@ from django.db import models
class User(AbstractUser): class User(AbstractUser):
timezone = models.CharField(max_length=64, default="UTC") timezone = models.CharField(max_length=64, default="UTC")
created_at = models.DateTimeField(auto_now_add=True) created_at = models.DateTimeField(auto_now_add=True)
class SetupLock(models.Model):
"""Singleton row that gates first-run setup against a race between
concurrent requests. The setup view always tries to INSERT the row
with pk=1 in the same transaction as creating the first admin user:
the database's primary-key uniqueness makes exactly one of those
INSERTs win even under concurrent requests, so only one admin account
can ever be created this way, instead of the User.objects.exists()
check alone (which has a window between the check and the save)."""
id = models.PositiveSmallIntegerField(primary_key=True, default=1)
+23 -1
View File
@@ -1,7 +1,7 @@
import pytest import pytest
from django.urls import reverse from django.urls import reverse
from .models import User from .models import SetupLock, User
# SetupWizardMiddleware's cache is reset around every test by the autouse # SetupWizardMiddleware's cache is reset around every test by the autouse
# fixture in the top-level conftest.py, not just here. # fixture in the top-level conftest.py, not just here.
@@ -43,6 +43,28 @@ def test_setup_creates_admin_and_logs_in(client):
assert stream_response.status_code == 200 assert stream_response.status_code == 200
@pytest.mark.django_db
def test_setup_refuses_a_second_concurrent_submission(client):
# Simulates two requests racing the setup form: the first claims the
# SetupLock row directly (standing in for a request that's already
# past the User.objects.exists() check), so a second submission's
# INSERT must lose instead of also creating an admin account.
SetupLock.objects.create()
response = client.post(
reverse("accounts:setup"),
{
"username": "raced",
"email": "raced@example.com",
"password": "correct-horse-battery",
"password_confirm": "correct-horse-battery",
},
)
assert response.status_code == 302
assert response.url == reverse("accounts:login")
assert not User.objects.filter(username="raced").exists()
@pytest.mark.django_db @pytest.mark.django_db
def test_setup_is_unreachable_once_a_user_exists(client): def test_setup_is_unreachable_once_a_user_exists(client):
User.objects.create_user("existing", password="irrelevant") User.objects.create_user("existing", password="irrelevant")
+17 -6
View File
@@ -1,11 +1,12 @@
from django.contrib import messages from django.contrib import messages
from django.contrib.auth import login from django.contrib.auth import login
from django.contrib.auth.decorators import user_passes_test from django.contrib.auth.decorators import user_passes_test
from django.db import IntegrityError, transaction
from django.shortcuts import redirect, render from django.shortcuts import redirect, render
from .forms import CreateUserForm, SetupForm from .forms import CreateUserForm, SetupForm
from .middleware import SetupWizardMiddleware from .middleware import SetupWizardMiddleware
from .models import User from .models import SetupLock, User
is_admin = user_passes_test(lambda u: u.is_authenticated and u.is_staff) is_admin = user_passes_test(lambda u: u.is_authenticated and u.is_staff)
@@ -16,11 +17,21 @@ def setup(request):
form = SetupForm(request.POST or None) form = SetupForm(request.POST or None)
if request.method == "POST" and form.is_valid(): if request.method == "POST" and form.is_valid():
user = form.save(commit=False) try:
user.set_password(form.cleaned_data["password"]) with transaction.atomic():
user.is_staff = True # Claims the one-time setup slot: the first of any
user.is_superuser = True # concurrent requests to reach this INSERT wins it, and
user.save() # every other one raises IntegrityError instead of also
# creating an admin account. See SetupLock's docstring.
SetupLock.objects.create()
user = form.save(commit=False)
user.set_password(form.cleaned_data["password"])
user.is_staff = True
user.is_superuser = True
user.save()
except IntegrityError:
messages.info(request, "Setup was already completed. Log in with that account.")
return redirect("accounts:login")
SetupWizardMiddleware._a_user_exists = True SetupWizardMiddleware._a_user_exists = True
login(request, user) login(request, user)
messages.success(request, "Welcome to Evection! Your admin account is ready.") messages.success(request, "Welcome to Evection! Your admin account is ready.")
@@ -0,0 +1,18 @@
# Generated by Django 5.2.18 on 2026-10-31 18:00
from django.db import migrations, models
class Migration(migrations.Migration):
dependencies = [
("feeds", "0002_remove_folder"),
]
operations = [
migrations.AddField(
model_name="feed",
name="claimed_at",
field=models.DateTimeField(blank=True, null=True),
),
]
+9
View File
@@ -40,6 +40,15 @@ class Feed(models.Model):
fetch_error_count = models.PositiveIntegerField(default=0) fetch_error_count = models.PositiveIntegerField(default=0)
is_active = models.BooleanField(default=True) is_active = models.BooleanField(default=True)
# Set by poll_due_feeds' atomic claim (an UPDATE ... WHERE clause, not
# select_for_update — Celery Beat and fetch_feed run in separate
# transactions) and cleared once fetch_feed finishes, so overlapping
# beat ticks or a fetch that runs long can't both dispatch the same
# feed. A claim older than CLAIM_STALE_MINUTES (tasks.py) is treated as
# abandoned and can be reclaimed, so a worker dying mid-fetch doesn't
# wedge the feed forever.
claimed_at = models.DateTimeField(null=True, blank=True)
created_at = models.DateTimeField(auto_now_add=True) created_at = models.DateTimeField(auto_now_add=True)
def __str__(self): def __str__(self):
+29 -4
View File
@@ -4,6 +4,7 @@ from io import BytesIO
from celery import shared_task from celery import shared_task
from django.contrib.auth import get_user_model from django.contrib.auth import get_user_model
from django.db.models import Q
from django.utils import timezone from django.utils import timezone
from apps.articles.tasks import apply_filter_rules_for_feed from apps.articles.tasks import apply_filter_rules_for_feed
@@ -14,13 +15,23 @@ from .services.fetcher import effective_interval_minutes, fetch_and_ingest
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)
# How long a claimed-but-unfinished fetch blocks the same feed from being
# claimed again — must comfortably exceed FETCH_TIMEOUT_SECONDS plus parse/
# ingest time, so a slow-but-healthy fetch is never raced by the next tick,
# while a worker that died mid-fetch doesn't wedge the feed forever.
CLAIM_STALE_MINUTES = 10
@shared_task @shared_task
def poll_due_feeds(): def poll_due_feeds():
"""Find every active Feed whose polling interval has elapsed and """Find every active Feed whose polling interval has elapsed, atomically
dispatch a fetch_feed task for each — fan-out happens here, not via one claim it (so overlapping beat ticks or a slow-running previous fetch
Celery Beat entry per feed. can't dispatch the same feed twice), and dispatch a fetch_feed task for
each claimed feed — fan-out happens here, not via one Celery Beat entry
per feed.
""" """
now = timezone.now()
claim_stale_before = now - timedelta(minutes=CLAIM_STALE_MINUTES)
due_feed_ids = [ due_feed_ids = [
feed.id feed.id
for feed in Feed.objects.filter(is_active=True).only( for feed in Feed.objects.filter(is_active=True).only(
@@ -28,9 +39,22 @@ def poll_due_feeds():
) )
if _is_due(feed) if _is_due(feed)
] ]
claimed_ids = []
for feed_id in due_feed_ids: for feed_id in due_feed_ids:
# The UPDATE's WHERE clause re-checks staleness at the database
# level: it only claims rows nobody else has already claimed more
# recently than CLAIM_STALE_MINUTES ago, so two workers racing this
# same loop can't both get rowcount=1 for the same feed.
claimed = (
Feed.objects.filter(id=feed_id, is_active=True)
.filter(Q(claimed_at__isnull=True) | Q(claimed_at__lt=claim_stale_before))
.update(claimed_at=now)
)
if claimed:
claimed_ids.append(feed_id)
for feed_id in claimed_ids:
fetch_feed.delay(feed_id) fetch_feed.delay(feed_id)
return len(due_feed_ids) return len(claimed_ids)
def _is_due(feed: Feed) -> bool: def _is_due(feed: Feed) -> bool:
@@ -48,6 +72,7 @@ def fetch_feed(feed_id: int):
return None return None
result = fetch_and_ingest(feed) result = fetch_and_ingest(feed)
feed.claimed_at = None
feed.save() feed.save()
changed_article_ids = result.new_article_ids + result.updated_article_ids changed_article_ids = result.new_article_ids + result.updated_article_ids
+52
View File
@@ -40,6 +40,58 @@ def test_poll_due_feeds_dispatches_only_due_feeds():
assert inactive_but_overdue.id not in dispatched_ids assert inactive_but_overdue.id not in dispatched_ids
@pytest.mark.django_db
def test_poll_due_feeds_does_not_reclaim_a_feed_it_just_claimed():
# Models a second beat tick (or an overlapping worker) running
# poll_due_feeds before the first claimed fetch_feed has finished —
# the feed must only be dispatched once.
overdue = Feed.objects.create(
feed_url="https://c.example.com/feed.xml",
last_fetched_at=timezone.now() - timedelta(hours=2),
fetch_interval_minutes=60,
)
with patch("apps.feeds.tasks.fetch_feed.delay") as mock_delay:
first_count = poll_due_feeds()
second_count = poll_due_feeds()
assert first_count == 1
assert second_count == 0
assert mock_delay.call_count == 1
overdue.refresh_from_db()
assert overdue.claimed_at is not None
@pytest.mark.django_db
def test_poll_due_feeds_reclaims_a_stale_claim():
from apps.feeds.tasks import CLAIM_STALE_MINUTES
overdue = Feed.objects.create(
feed_url="https://c.example.com/feed.xml",
last_fetched_at=timezone.now() - timedelta(hours=2),
fetch_interval_minutes=60,
claimed_at=timezone.now() - timedelta(minutes=CLAIM_STALE_MINUTES + 1),
)
with patch("apps.feeds.tasks.fetch_feed.delay") as mock_delay:
dispatched_count = poll_due_feeds()
assert dispatched_count == 1
assert mock_delay.call_args.args[0] == overdue.id
@pytest.mark.django_db
@responses.activate
def test_fetch_feed_task_clears_its_claim():
feed = Feed.objects.create(feed_url="https://example.com/feed.xml", claimed_at=timezone.now())
responses.add(responses.GET, feed.feed_url, body=GOOD_FEED_XML, status=200)
fetch_feed(feed.id)
feed.refresh_from_db()
assert feed.claimed_at is None
GOOD_FEED_XML = """<?xml version="1.0"?> GOOD_FEED_XML = """<?xml version="1.0"?>
<rss version="2.0"><channel> <rss version="2.0"><channel>
<title>Example Blog</title> <title>Example Blog</title>
+5
View File
@@ -32,6 +32,11 @@ def test_service_worker_is_served_at_root_with_correct_type_and_scope(client):
assert 'self.addEventListener("fetch"' in body assert 'self.addEventListener("fetch"' in body
# Mutations must stay network-only, never intercepted. # Mutations must stay network-only, never intercepted.
assert 'request.method !== "GET"' in body assert 'request.method !== "GET"' in body
# Logging out must purge PAGES_CACHE so a later offline navigation on
# a shared/account-switched device can't serve another user's
# previously cached authenticated pages.
assert "caches.delete(PAGES_CACHE)" in body
assert reverse("accounts:logout") in body
@pytest.mark.django_db @pytest.mark.django_db
+15 -1
View File
@@ -1,8 +1,22 @@
from django.core.exceptions import ImproperlyConfigured
from .base import * # noqa: F403 from .base import * # noqa: F403
from .base import env from .base import SECRET_KEY, env
DEBUG = False DEBUG = False
# base.py falls back to a publicly known placeholder if SECRET_KEY isn't
# set — fine for a throwaway dev DB, but production must never sign
# sessions/CSRF tokens with a key anyone can read in this repo's history.
# Fail startup rather than quietly running with a known key.
if not SECRET_KEY or SECRET_KEY == "django-insecure-change-me-in-production":
raise ImproperlyConfigured(
"SECRET_KEY is missing or still the placeholder default. Set a unique, "
"random SECRET_KEY in .env before running with prod settings — generate "
'one with: python -c "from django.core.management.utils import '
'get_random_secret_key; print(get_random_secret_key())"'
)
# Health checks (scripts/upgrade.sh, an external uptime monitor, ...) hit # Health checks (scripts/upgrade.sh, an external uptime monitor, ...) hit
# gunicorn directly over loopback, bypassing the reverse proxy entirely — # gunicorn directly over loopback, bypassing the reverse proxy entirely —
# that request's Host header is "127.0.0.1:<port>", not the public domain # that request's Host header is "127.0.0.1:<port>", not the public domain
+5 -4
View File
@@ -295,6 +295,7 @@ git fetch --tags --force
LATEST_TAG="$(git tag --sort=-creatordate | head -n1)" LATEST_TAG="$(git tag --sort=-creatordate | head -n1)"
git checkout --detach "$LATEST_TAG" git checkout --detach "$LATEST_TAG"
chown -R evection:evection /opt/evection/app chown -R evection:evection /opt/evection/app
sed -i "s/^EVECTION_PWA_CACHE_VERSION=.*/EVECTION_PWA_CACHE_VERSION=$LATEST_TAG/" .env
sudo -u evection -H /opt/evection/venv/bin/pip install . sudo -u evection -H /opt/evection/venv/bin/pip install .
sudo -u evection -H env DJANGO_SETTINGS_MODULE=config.settings.prod \ sudo -u evection -H env DJANGO_SETTINGS_MODULE=config.settings.prod \
/opt/evection/venv/bin/python manage.py migrate /opt/evection/venv/bin/python manage.py migrate
@@ -311,7 +312,7 @@ git tag v$(date +%Y.%m.%d)
git push --tags git push --tags
``` ```
Bump `EVECTION_PWA_CACHE_VERSION` in `.env` whenever a release changes cached `upgrade.sh` sets `EVECTION_PWA_CACHE_VERSION` in `.env` to the deployed tag on every
static assets (CSS/JS/icons) — it's baked into the service worker's cache name, so run — it's baked into the service worker's cache name, so this forces every installed
changing it forces every installed PWA to drop its old cache instead of serving PWA to drop its old cache instead of serving stale assets after the upgrade. You don't
stale assets after the upgrade. need to bump it by hand.
+1 -1
View File
@@ -204,7 +204,7 @@ REDIS_URL=${EVECTION_REDIS_URL:-redis://localhost:6379/0}
EVECTION_GUNICORN_BIND=$GUNICORN_BIND EVECTION_GUNICORN_BIND=$GUNICORN_BIND
EVECTION_SIGNUP_ENABLED=False EVECTION_SIGNUP_ENABLED=False
EVECTION_MIN_FETCH_INTERVAL_MINUTES=15 EVECTION_MIN_FETCH_INTERVAL_MINUTES=15
EVECTION_PWA_CACHE_VERSION=v1 EVECTION_PWA_CACHE_VERSION=${LATEST_TAG:-v1}
EOF EOF
chown "$EVECTION_USER:$EVECTION_USER" "$ENV_FILE" chown "$EVECTION_USER:$EVECTION_USER" "$ENV_FILE"
chmod 600 "$ENV_FILE" chmod 600 "$ENV_FILE"
+12 -3
View File
@@ -72,6 +72,18 @@ echo "Deploying $LATEST_TAG (currently on: $CURRENT_REF)"
git checkout --quiet --detach "$LATEST_TAG" git checkout --quiet --detach "$LATEST_TAG"
chown -R "$EVECTION_USER:$EVECTION_USER" "$APP_DIR" chown -R "$EVECTION_USER:$EVECTION_USER" "$APP_DIR"
# Bump the PWA's cache-busting version to this release's tag on every
# deploy, not just ones that happened to touch static assets — a missed
# manual bump otherwise leaves installed PWAs serving a stale cached
# shell/pages indefinitely. Cheap even when nothing cacheable changed:
# it just makes the service worker open a fresh cache name next time it
# activates.
if grep -q '^EVECTION_PWA_CACHE_VERSION=' "$APP_DIR/.env"; then
sed -i "s/^EVECTION_PWA_CACHE_VERSION=.*/EVECTION_PWA_CACHE_VERSION=$LATEST_TAG/" "$APP_DIR/.env"
else
echo "EVECTION_PWA_CACHE_VERSION=$LATEST_TAG" >> "$APP_DIR/.env"
fi
echo "==> Installing updated dependencies" echo "==> Installing updated dependencies"
sudo -u "$EVECTION_USER" -H "$VENV_DIR/bin/pip" install --upgrade "$APP_DIR" sudo -u "$EVECTION_USER" -H "$VENV_DIR/bin/pip" install --upgrade "$APP_DIR"
@@ -131,7 +143,4 @@ systemctl status evection-gunicorn --no-pager -l | head -10
echo echo
echo "==> Deployed $LATEST_TAG." echo "==> Deployed $LATEST_TAG."
echo " If this release changed cached static assets (CSS/JS/icons), bump"
echo " EVECTION_PWA_CACHE_VERSION in $APP_DIR/.env and restart evection-gunicorn"
echo " again so installed PWAs drop their old cache instead of serving stale assets."
echo " journalctl -u evection-gunicorn -f to watch logs." echo " journalctl -u evection-gunicorn -f to watch logs."
+15 -1
View File
@@ -66,8 +66,23 @@ async function cacheFirstForAsset(request) {
return response; return response;
} }
const LOGOUT_PATH = "{% url 'accounts:logout' %}";
self.addEventListener("fetch", (event) => { self.addEventListener("fetch", (event) => {
const { request } = event; const { request } = event;
const url = new URL(request.url);
// Logging out ends this device's session, so any authenticated HTML
// sitting in PAGES_CACHE must not outlive it — otherwise a later
// offline navigation (shared device, different account logging in
// next) could still serve a previous user's cached article pages.
// Purging it synchronously with the logout request itself, rather
// than via a postMessage from the page, means there's no race with
// the navigation that follows.
if (request.method === "POST" && url.pathname === LOGOUT_PATH) {
event.respondWith(caches.delete(PAGES_CACHE).then(() => fetch(request)));
return;
}
// Mutations (mark-read, star, OPML upload, forms...) are always // Mutations (mark-read, star, OPML upload, forms...) are always
// network-only — never intercept anything but GET. // network-only — never intercept anything but GET.
@@ -78,7 +93,6 @@ self.addEventListener("fetch", (event) => {
return; return;
} }
const url = new URL(request.url);
if (url.pathname.startsWith("/static/")) { if (url.pathname.startsWith("/static/")) {
event.respondWith(cacheFirstForAsset(request)); event.respondWith(cacheFirstForAsset(request));
} }