Invalidate cache tags after transaction

RTBC - I'd like to get this in, so that people can start testing the core patch

Hi Fabian and Brandon! Thanks for working on this issue.

I think _d8cache_has_transaction_callbacks() is returning the opposite boolean value that it should.

Instead of

function _d8cache_has_transaction_callbacks($connection)
  if (method_exists($connection, 'addRootTransactionEndCallback')) {
    return FALSE;
  }
  return TRUE;
}

Shouldn't it be

function _d8cache_has_transaction_callbacks($connection)
  if (method_exists($connection, 'addRootTransactionEndCallback')) {
    return TRUE;
  }
  return FALSE;
}

or just

function _d8cache_has_transaction_callbacks($connection)
   return method_exists($connection, 'addRootTransactionEndCallback');
}

With these patches applied (and the boolean flipped) I did not get deadlocks.

To replicate the deadlocks I made this branch in Pantheon Advanced Page Cache: https://github.com/pantheon-systems/pantheon_advanced_page_cache/compare... It reuses a previous Behat step I had written to test very long headers. The step saves nodes that reference hundreds of taxonomy terms (just with the free tagging field on the article content type in the Standard profile). By invalidating cache tags for those taxonomy terms and running Behat in two window simultaneously I can reliably replicated the deadlocks.

Great catch and thanks for verifying that this indeed solves the deadlocks.

Adding that this is postponed just on one issue.

RTBC - Excellent work!

I decided to revert the PR again and support the original API idea instead.

RTBC - We decided to change this back, because we got meanwhile consensus in D8 to add the $committed parameter.

Thanks!

FYI: #2966607: Invalidating 'node_list' and other broad cache tags early in a transaction severely increases lock wait time and probability of deadlock just landed.