We had a problem in a client site that resulted in a fatal error in Relay:
Fatal error: Uncaught Error: Call to a member function hMset() on bool in /app/docroot/modules/contrib/redis/src/Cache/Relay.php:112
With a backtrace like this:
Stack trace:
#0 /app/docroot/modules/contrib/redis/src/Cache/CacheBase.php(143): Drupal\redis\Cache\Relay->set('system.logging', Array, -1, Array) #1 /app/docroot/core/lib/Drupal/Core/Config/CachedStorage.php(105): Drupal\redis\Cache\CacheBase->setMultiple(Array)
#2 /app/docroot/core/lib/Drupal/Core/Config/ConfigFactory.php(165): Drupal\Core\Config\CachedStorage->readMultiple(Array)
#3 /app/docroot/core/lib/Drupal/Core/Config/ConfigFactory.php(104): Drupal\Core\Config\ConfigFactory->doLoadMultiple(Array, true)
#4 /app/docroot/core/lib/Drupal/Core/Config/ConfigFactory.php(89): Drupal\Core\Config\ConfigFactory->doGet('system.logging')
#5 /app/docroot/core/lib/Drupal.php(411): Drupal\Core\Config\ConfigFactory->get('system.logging')
#6 /app/docroot/core/includes/errors.inc(323): Drupal::config('system.logging')
#7 /app/docroot/core/includes/errors.inc(123): _drupal_get_error_level()
#8 /app/docroot/core/includes/bootstrap.inc(385): error_displayable()
#9 /app/docroot/core/includes/bootstrap.inc(369): _drupal_exception_handler_additional(Object(Error), Object(Error))
#10 [internal function]: _drupal_exception_handler(Object(Error))
So there was an exception, we don't know which one and then it checked system logging configuration, apparently had a cache miss on that, and then failed to write that.
According to the documentation $pipe = $this->client->multi(); can fail, although I've never seen that happen with any of the other integrations.
As a minimal fix, check that we could initialize the pipeline, silently skip the cache write if not or maybe do a trigger_error().
Relay supports throwing exceptions as well, maybe support that, possibly as a setting, that might be useful to track this down. There is also a setting redis.loglevel that could be set to to get useful messages out, that might be easier.
Fixed
1.0
Code
This is the minimal change to avoid a fatal error. It might make sense to update the other clients accordingly, but I've never heard of this happening there.
Automatically closed - issue fixed for 2 weeks with no activity.