Julien Riou<p>Upgrade from MySQL 8.0.33 to 8.0.41 this morning on production.</p><p>Everything went smoothly... until:</p><p>Lock timeout when upgrading an intermediate primary in a ClusterSet. It looks like <a href="https://bugs.mysql.com/bug.php?id=100242" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">bugs.mysql.com/bug.php?id=1002</span><span class="invisible">42</span></a>. I know that kill -9 has been used in the past (yes, MySQL can be challenging and it's the only solution), so let's re-initialize the host. 6TB of data, it's going to be long but we have no choice at this point.</p><p>The Percona Server version of MySQL Shell doesn't support the Javascript interface anymore. The official documentation has Javascript examples everywhere but not Python. I had to guess and learn the Python equivalent in emergency.</p><p>Then if we define a "localAddress" to "Cluster.add_instance", MySQL tries to use the deprecated "XCOM" communication stack instead of "MYSQL". If you have errors on port 33061, you should remove the localAddress and/or use the "MYSQL" communication stack <a href="https://dev.mysql.com/doc/mysql-shell/8.0/en/shell-admin-api-communication-stack.html" rel="nofollow noopener noreferrer" translate="no" target="_blank"><span class="invisible">https://</span><span class="ellipsis">dev.mysql.com/doc/mysql-shell/</span><span class="invisible">8.0/en/shell-admin-api-communication-stack.html</span></a></p><p>Every time the service is started, it takes 30 seconds to initialize the InnoDB buffer pool because the host has 1TB of RAM.</p><p>For an unknown reason, the mandb process takes multiple minutes when upgrading a package.</p><p>Time is really long when you do upgrades.</p><p><a href="https://hachyderm.io/tags/mysql" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>mysql</span></a></p>