etermed

Thanks for the quick reply, and fair points. Nothing in our setup is read-only on purpose. The trigger is transient: the host’s nightly backup briefly fills the disk, so opening or writing the .sqlite file fails with “database or disk is full (13)”. Your retry / delete-and-recreate logic would handle that fine, but it never gets a chance to run, because the error_log() call crashes first: esc_html() ends up in get_option() -> wp_cache_get() while $wp_object_cache is still null.

So the minimal fix I’d suggest is just one line in error_log(): use htmlspecialchars( $server_software, ENT_QUOTES, ‘UTF-8’ ) instead of esc_html(). That’s log output, not page output, so it shouldn’t conflict with plugin-check, and it lets your existing retry path work. Agreed on :memory: (we don’t use APCu, but I see the wp-cli problem) and on keeping wp_die(), so please treat those as our local workaround, not a request.

We’re also moving the backups off the production disk on our side. Thanks again!