Performance: ap_actor table grows unbounded, hourly refresh job doesn't scale #2763
Description
Summary
The ap_actor custom post type accumulates entries indefinitely from any ActivityPub interaction (comments, mentions, replies, etc.), while the hourly activitypub_update_remote_actors cron job attempts to refresh ALL cached actors regardless of whether they're followers or just one-time interactions.
On sites with significant ActivityPub activity, this causes severe performance degradation.
Example Scenario
A site with:
- 122 actual followers (stored in
ap_inbox) - 19,381 cached actor profiles (stored in
ap_actor)
The hourly job tries to refresh all 19,381 actors, causing:
- 180% CPU spike
- 502 timeouts
- php-fpm workers exhausted
Technical Analysis
Data Structure
ap_inbox- Stores follower relationships, has_activitypub_user_idmetaap_actor- Caches remote actor profiles, purely a cache with no follower relationship
These are separate. An ap_actor entry is created for ANY remote actor the site interacts with:
- Someone who comments once
- A user mentioned in a post
- A reply target
- Any ActivityPub interaction
The Hourly Job (update_remote_actors)
// class-scheduler.php:201-229
public static function update_remote_actors() {
$number = 5; // Only 5 per run (50 with DISABLE_WP_CRON)
$actors = Remote_Actors::get_outdated( $number );
foreach ( $actors as $actor ) {
$meta = get_remote_metadata_by_actor( $actor->guid, false ); // HTTP request
// ... updates actor
}
}// class-remote-actors.php:461-476 - "Outdated" definition
'date_query' => array(
'column' => 'post_modified_gmt',
'before' => gmdate( 'Y-m-d', time() - DAY_IN_SECONDS ), // > 24 hours old
),Problems:
- Queries ALL
ap_actorentries, not just followers - With 19k actors at 5/hour = 3,877 hours (161 days) to cycle through
- Each requires an HTTP request to the remote server
- No distinction between important actors (followers) and irrelevant ones (one-time commenters)
The Cleanup Job (cleanup_remote_actors)
Only removes actors that:
- Return a Tombstone (explicitly deleted on remote server)
- Fail 5+ consecutive fetch attempts
Does NOT clean up actors that are:
- Not followers
- Haven't interacted in months/years
- From defunct servers that still respond
Investigation Queries
-- Actors used by comments
SELECT COUNT(DISTINCT meta_value) FROM wp_commentmeta
WHERE meta_key = '_activitypub_remote_actor_id';
-- Result: 42
-- Actors that are followers
SELECT COUNT(DISTINCT p.ID) FROM wp_posts p
WHERE p.post_type = 'ap_actor'
AND p.guid IN (
SELECT DISTINCT pm.meta_value FROM wp_postmeta pm
INNER JOIN wp_posts inbox ON inbox.ID = pm.post_id
WHERE inbox.post_type = 'ap_inbox'
AND pm.meta_key = '_activitypub_activity_remote_actor'
);
-- Result: 23
-- Orphaned actors (not used by anything)
-- Result: 19,329 out of 19,381 totalSuggested Fixes
Option 1: Scope actor refresh to relevant actors only
Only refresh actors that are:
- Actual followers (have corresponding
ap_inboxentry) - Have interacted within a configurable timeframe
- Are being actively followed by the site
Option 2: Add proper cleanup for stale actors
Create a cleanup job that removes ap_actor entries for:
- Non-followers with no interaction in X days (configurable)
- Actors not referenced by any comments or interactions
Option 3: Add interaction tracking
Store _activitypub_last_interaction timestamp on actors to enable:
- Smart refresh prioritization (recent interactions first)
- Cleanup of truly stale actors
- Skip refreshing actors that haven't been relevant in months
Option 4: Reduce refresh frequency for non-followers
- Followers: refresh every 24 hours (current behavior)
- Non-followers: refresh only on interaction, or weekly/monthly
Workaround
Site owners can manually clean up orphaned actors:
-- Delete actors not used by comments or followers
DELETE wp_posts FROM wp_posts
LEFT JOIN wp_commentmeta cm ON wp_posts.ID = cm.meta_value
AND cm.meta_key = '_activitypub_remote_actor_id'
LEFT JOIN (
SELECT DISTINCT pm.meta_value as actor_guid
FROM wp_postmeta pm
INNER JOIN wp_posts inbox ON inbox.ID = pm.post_id
WHERE inbox.post_type = 'ap_inbox'
AND pm.meta_key = '_activitypub_activity_remote_actor'
) followers ON wp_posts.guid = followers.actor_guid
WHERE wp_posts.post_type = 'ap_actor'
AND cm.meta_value IS NULL
AND followers.actor_guid IS NULL;
-- Clean up orphaned postmeta
DELETE pm FROM wp_postmeta pm
LEFT JOIN wp_posts p ON pm.post_id = p.ID
WHERE p.ID IS NULL;Environment
- ActivityPub plugin 7.8.4
- PHP 8.3
- WordPress 6.x
- VPS with limited PHP workers (5-10)