Problem/Motivation
Hi, we are having an issue with missing route and roles cache context on SSO users that match an existing Drupal user.
Basically the tags are Ok. When that happens, and the user is redirected to its /user/ID page, the Primary/secondary Task Blocks get stuck for every other user across the site bc the Cache variation related to Context is basically useless. Even the Home Page shows that user's USER edit/activity etc for any other user.
A bit of background. Running latest (as of today) Drupal 10.5.x, PHP 8.3 (which is quite aggressive on caching) and proper READ_COMMITTED for the DB, caching managed via REDIS. Latest release 3.11 of this module. SAML config is a following
-Login/User: "Attempt to link SAML data to existing Drupal users". True for both email and name, we had a few Roles there (except Admin)
- User Roles: we had basically a complete "unassign" for every role, a single Unconditional Role, and then SAML attribute based value Conversions (all of the working)
This does not happen all the time and it is hard to reproduce, but we had it twice this week. The Redis report shows that the Tasks are properly variating... (means no external code is removing context ... which is not the same as "context" value was present)
"entity_view:block:themename_primary_local_tasks: 1876 (route, url.query_args, user.roles, user, user.roles:authenticated)"
Clearing caches, or re-saving Block layout solves the problem
I suspect a race condition (DB?), when the user gets the Roles Unassigned and then re-assigned and when the /user landing page is rendered there are missing context values. But it might be also about a missing route during the cache assignment process?
Others (not for this module) have reported similar issues here
https://www.drupal.org/forum/support/module-development-and-code-questio... →
Steps to reproduce
Very hard. And once it triggers it is very hard to actually debug cache ability with our actually clearing caches (no hook will run until you do so)...
Proposed resolution
I have none yet. Silly ideas: One could maybe force, for the redirect page, a one time "no cache" (cache kill switch) ?
Basically e.g alter via adding an extra context (but on large site with hundreds of users that would be really silly)
Or maybe check there, if the current route is set, and all the extra data is present and only act on e.g setting caching time to 0? (once?)
yourmodule_menu_local_tasks_alter(&$data, $route_name, RefinableCacheableDependencyInterface &$cacheability) {
$cacheability->addCacheContexts(['user']);
}
has anyone else seen this behavior? Thanks