Purging
phil/http-cache-tags-bundleInvalidator::invalidate('menu') gathers tags to purge; they are purged once, each tag once, at the end of:
| Where | When |
|---|---|
| A request | kernel.terminate, once the response is sent |
| A command | console.terminate |
| A Messenger worker's message | WorkerMessageHandledEvent or WorkerMessageFailedEvent: a page an asynchronous job changes, a publication phil/scheduled-transitions-bundle runs from Symfony Scheduler, is purged as soon as the message is done, not when the worker stops |
When an entity changes
A CacheTagResolver names the tags of the responses that show an entity; with autoconfiguration, the bundle finds
it. With invalidate_entities, every entity a Doctrine flush inserts, updates or deletes is passed to the resolvers.
final class PageTags implements CacheTagResolver
{
public function tagsFor(object $entity): iterable
{
return $entity instanceof Page ? ['page:'.$entity->id, 'page-list'] : [];
}
}Invalidator::invalidateEntity($entity) does the same for an object that is not flushed.
The caches
| Purger | What it sends |
|---|---|
VarnishPurger | A PURGE request to each Varnish, the tags in xkey-purge, separated by spaces, for the xkey module |
CloudflarePurger | POST /zones/{zone}/purge_cache with {"tags": […]}, a hundred tags per call, with the API token |
Every Varnish is purged, and every Cloudflare call made, even when one fails. An answer of 400 or more, or a cache
that cannot be reached, is then one PurgeFailed that names each failure, the transport error as its previous one:
those responses stay stale until they expire. InvalidationFlushListener logs it as an error, on the http_cache
channel, with the container's logger, so that a response already sent or a worker keeps going; without a logger, it
is thrown.