Self-hosted project management for devs and their clients: tickets, time tracking, docs and meetings, with an AI assistant that cites its sources. AGPL-3.0.
Ruby agplai-assistantdockerdsgvogdprhotwireissue-trackerjira-alternativeknowledge-basellmmcp-servermeeting-notesopen-sourceproject-managementredmine-alternativeruby-on-railsself-hostedticketingtime-tracking
Sweep expired cache entries hourly and give the file store its own directory
Guido Boyke9232b59df21b
diff --git a/app/jobs/cache_cleanup_job.rb b/app/jobs/cache_cleanup_job.rbnew file mode 100644index 0000000..52583f9--- /dev/null+++ b/app/jobs/cache_cleanup_job.rb@@ -0,0 +1,23 @@+# frozen_string_literal: true++# Deletes expired cache entries, hourly (config/recurring.yml).+#+# Production caches in a file store (production.rb, tmp/cache/store), and a+# file store never deletes an expired entry on its own: it only skips it on+# read. Rack::Attack writes a counter per client and minute there, so without+# this sweep the directory grows by every counter ever written, and whatever a+# key named stays on the disk. Emptying tmp/ instead is no substitute: it also+# deletes the server's socket under tmp/sockets. Entries without an expiry+# stay, as in any store.+#+# Stores that expire entries themselves (Redis on staging, Memcache, Solid+# Cache) raise NotImplementedError from #cleanup; there is nothing to do.+class CacheCleanupJob < ApplicationJob+ queue_as :default++ def perform+ Rails.cache.cleanup+ rescue NotImplementedError+ nil+ end+enddiff --git a/config/environments/production.rb b/config/environments/production.rbindex 7e0de93..d804bd2 100644--- a/config/environments/production.rb+++ b/config/environments/production.rb@@ -119,8 +119,10 @@ Rails.application.configure doendend- # Use a different cache store in production.- # config.cache_store = :mem_cache_store+ # A file store, as Rails' default, but in a directory of its own: the default+ # is tmp/cache itself, which bootsnap and the asset caches share, and+ # CacheCleanupJob (FileStore#cleanup) reads every file below the store.+ config.cache_store = :file_store, Rails.root.join('tmp/cache/store')# Use a real queuing backend for Active Job. Solid Queue runs on the primary# database: the queue connection is pinned to :primary indiff --git a/config/recurring.yml b/config/recurring.ymlindex 5658ea7..d217af4 100644--- a/config/recurring.yml+++ b/config/recurring.yml@@ -15,6 +15,13 @@ production:clear_solid_queue_finished_jobs:command: "SolidQueue::Job.clear_finished_in_batches(sleep_between_batches: 0.3)"schedule: every hour at minute 12+ # The file store deletes no expired entry on its own (CacheCleanupJob says+ # why that matters). Hourly: a rate limit counter lives a minute, and nothing+ # should outlive its expiry by much more than an hour.+ cache_cleanup:+ class: CacheCleanupJob+ queue: default+ schedule: every hour at minute 25# GDPR: runs scheduled account deletions (scheduled_deletion_at exceeded)# automatically — previously NEVER ran automatically (only manually via rails runner).scheduled_account_deletions:@@ -191,6 +198,11 @@ staging:clear_solid_queue_finished_jobs:command: "SolidQueue::Job.clear_finished_in_batches(sleep_between_batches: 0.3)"schedule: every hour at minute 12+ # A no-op on Redis, here so the schedule is the same as on production.+ cache_cleanup:+ class: CacheCleanupJob+ queue: default+ schedule: every hour at minute 25scheduled_account_deletions:class: ScheduledDeletionJobqueue: defaultdiff --git a/spec/config/instance_settings_spec.rb b/spec/config/instance_settings_spec.rbindex 78901cb..8e82a59 100644--- a/spec/config/instance_settings_spec.rb+++ b/spec/config/instance_settings_spec.rb@@ -252,6 +252,18 @@ RSpec.describe 'config/environments/production.rb' doendend+ # The cache gets a directory of its own under tmp/cache. Rails' default is+ # tmp/cache itself, where bootsnap and the asset caches live too, and+ # CacheCleanupJob (FileStore#cleanup) reads every file below the store's+ # directory. In every environment of production.rb, the image included.+ describe 'the cache store' do+ it 'is a file store in tmp/cache/store, with or without LALABASE_HOST' do+ [evaluate, evaluate('LALABASE_HOST' => 'tickets.example.org')].each do |config|+ expect(config[:cache_store]).to eq([:file_store, Rails.root.join('tmp/cache/store')])+ end+ end+ end+# Lines a shared .env may carry from an old template. production.rb used to# ignore them, and it still must.context 'on our servers with ASSET_HOST, RAILS_LOG_TO_STDOUT and LALABASE_FORCE_SSL lines' dodiff --git a/spec/config/recurring_schedule_spec.rb b/spec/config/recurring_schedule_spec.rbindex d7492e0..a039afe 100644--- a/spec/config/recurring_schedule_spec.rb+++ b/spec/config/recurring_schedule_spec.rb@@ -29,6 +29,14 @@ RSpec.describe 'config/recurring.yml' doendend+ # Without a sweep the file store keeps every expired entry for good.+ it 'contains the hourly cache cleanup' do+ entry = schedule.fetch(env)['cache_cleanup']+ expect(entry).to be_present+ expect(entry['class']).to eq('CacheCleanupJob')+ expect(Fugit.parse(entry['schedule'])).to be_present+ end+it 'contains the documentation polling sync (Slice-2 auto-sync contract)' doentry = schedule.fetch(env)['documentation_polling_sync']expect(entry).to be_presentdiff --git a/spec/jobs/cache_cleanup_job_spec.rb b/spec/jobs/cache_cleanup_job_spec.rbnew file mode 100644index 0000000..1c2fd18--- /dev/null+++ b/spec/jobs/cache_cleanup_job_spec.rb@@ -0,0 +1,59 @@+# frozen_string_literal: true++require 'rails_helper'++# What: the hourly sweep of expired cache entries (config/recurring.yml).+# Type: Job spec against a real file store in a temporary directory: the+# point is which files stay on the disk, and a double has no disk.+# Why: Production caches in a file store, and a file store deletes nothing on+# its own. An expired entry (a rate limit counter, which until 2026-10-31+# named the raw access token in its file name) stays on the disk for good+# unless this job removes it.+RSpec.describe CacheCleanupJob do+ include ActiveSupport::Testing::TimeHelpers++ around do |example|+ Dir.mktmpdir do |dir|+ @dir = dir+ example.run+ end+ end++ let(:store) { ActiveSupport::Cache::FileStore.new(File.join(@dir, 'store')) }++ before { allow(Rails).to receive(:cache).and_return(store) }++ it 'deletes expired entries and keeps live ones' do+ store.write('expired', 1, expires_in: 1.minute)+ store.write('live', 2, expires_in: 1.hour)+ store.write('forever', 3)++ travel 2.minutes do+ described_class.perform_now++ expect(store.read('live')).to eq(2)+ expect(store.read('forever')).to eq(3)+ end+ expect(Dir.glob(File.join(@dir, 'store', '**', '*')).select { |f| File.file?(f) }.size).to eq(2)+ end++ # The production store has a directory of its own under tmp/cache+ # (production.rb). This pins what that is for: a file the store did not+ # write survives, so bootsnap and the asset caches next to it would as well.+ it 'leaves a file it did not write alone' do+ foreign = File.join(@dir, 'store', 'foreign.bin')+ FileUtils.mkdir_p(File.dirname(foreign))+ File.binwrite(foreign, "\x04\x08not an entry")++ described_class.perform_now++ expect(File.exist?(foreign)).to be(true)+ end++ # Redis and Memcache expire entries themselves and raise on cleanup.+ it 'does nothing on a store that cannot clean up' do+ allow(Rails).to receive(:cache).and_return(Class.new(ActiveSupport::Cache::Store).new)++ expect { described_class.perform_now }.not_to raise_error+ end+end