Error occurring in failing update is not reported in the log

[pfrenssen]

This has been split off from #3303.

@claudiu-cristea reported in #3303 that if an error occurs during a database update this is not logged correctly. It appears this is only working correctly for exceptions being thrown in hook_update_N(), but not when a post-update fails.

We should fix this and extend the test coverage for post-updates.

[claudiu-cristea]

@pfrenssen, I guess the problem is that messages logged in subsequent processes are not shown (have no explanation why for the moment). As hook_update_N() are running first, they manage to run in the main process, while post updates are spawned in subsequent processes. Just a guess.

[pfrenssen]

I did some tests on the branch which is exhibiting the problems (ec-europa/joinup-dev#1045):

• The error which is being thrown is Exception: The theme implementations may not be rendered until all modules are loaded. in Drupal\Core\Theme\ThemeManager->render() (line 139 of web/core/lib/Drupal/Core/Theme/ThemeManager.php). This is strange, why aren't the modules loaded?
• The error only occurs when running hook_update_N() updates together with a post-update. If the hook_update_N() updates are run first, and then the post-update separately the exception is not thrown and the update finishes successfully.

[pfrenssen]

I have fixed the missing log output in #3310.

I have also provided a test case for the The theme implementations may not be rendered until all modules are loaded exception in #3311, but I didn't have the time yet to investigate this.

[pfrenssen]

I found the cause why the modules are not being loaded, this has been introduced in 978893e

In this commit the container is being rebuilt after calling drupal_flush_all_caches(), but look what this does:

public function rebuildContainer() {
  // Empty module properties and for them to be reloaded from scratch.
  $this->moduleList = NULL;
  [...]
}

[pfrenssen]

OK this took me a long time to find since I was completely on the wrong track. I thought it was caused by the kernel juggling that was introduced recently in the UpdateKernel PR, but it turns out it was this container rebuild. I have a green test now:

$ UNISH_NO_TIMEOUTS=1 UNISH_TMP=/home/pieter/v/drush-sut php unish.phpunit.php tests/UpdateDBTest.php --stop-on-failure -vvv
Executing: '/home/pieter/v/drush-sut/drush-sut/vendor/bin/phpunit' --configuration /home/pieter/v/drush-sut/drush-sut/vendor/drush/drush/tests tests/UpdateDBTest.php --stop-on-failure -vvv
PHPUnit 4.8.36 by Sebastian Bergmann and contributors.

Runtime:	PHP 7.1.12
Configuration:	/home/pieter/v/drush/tests/phpunit.xml.dist

.....

Time: 1.3 minutes, Memory: 4.00MB

OK (5 tests, 11 assertions)

Posting fix in #3311.