During an incident

Two runtime settings change every rule at once, without a deploy, and end by themselves (ADR-0011). They need the package's newest migrations: publish them and run php artisan migrate.

php artisan dynamic-throttle:bypass on --for=30m   # every request goes through, unlimited
php artisan dynamic-throttle:scale 2 --for=2h      # every limit doubles; 0.5 halves them
php artisan dynamic-throttle:status                # what servers apply, and until when
php artisan dynamic-throttle:bypass off
php artisan dynamic-throttle:scale off

• End times. --for takes 90s, 30m, 2h or 1d, up to 365 days. Without it, a change ends after one hour, and --forever sets no end time. Servers stop applying a setting at its end time, by their own clock, even when nothing compiles in between.
• Scaling. The factor goes from 0.1 to 10, with at most two decimals. A rule's limit becomes max(1, ceil(max_attempts × factor)), a dry-run rule's too, and the counts made before the change still count.
• One bad rule. php artisan dynamic-throttle:list shows every rule with its id. dynamic-throttle:disable-rule 12 switches rule 12 off, and enable-rule 12 brings it back.
• Confirmation. In production, turning bypass on, a factor below 1 and disabling an enforcing rule ask first. Scripts pass --force.
• Propagation. Each change compiles the manifest, and servers apply it within manifest.refresh_seconds. If the manifest cannot be written, the change stays saved, the command exits 1, and servers keep the previous manifest until a compile succeeds.
• The audit log. Each change writes one append-only entry to dynamic_throttle_audit_log: the actor (cli: and the system user), the action, the target, and the state before and after.
• In deploy scripts. dynamic-throttle:status exits 1 when the config is invalid, such as an unknown driver, an undefined store or an invalid prefix. dynamic-throttle:health --strict also exits 1 for a setting left on with no end time, or with one more than a day away (the emergency check).
• The last resort. When the database and the cache are both down, set DYNAMIC_THROTTLE_ENABLED=false and reload the config: the middleware then does nothing.