LTTng bugs repository: Issueshttps://bugs.lttng.org/https://bugs.lttng.org/themes/lttng/favicon/a.ico?14249722912015-03-26T08:19:40ZLTTng bugs repository
Redmine LTTng-tools - Feature #883 (New): consumerd: Automatic reconnect to relaydhttps://bugs.lttng.org/issues/8832015-03-26T08:19:40ZJesper Derehagjderehag@hotmail.com
<p>consumerd unfortunatly does not handle reconnects to relayd properly.</p>
<p>1. If relayd is started <strong>after</strong> connection attempt from consumerd it will leave the session in disconnected state, and a restart of the session is required.<br />2. If relayd crashes (or just a spurious disconnect), it will also leave all consumerd:s in disconnected state and <strong>all</strong> sessions needs to be restarted.</p>
<p>Suggestion is to continously try to connect to relayd if in disconnected state.<br />If always trying continously to reconnect, it should handle all types of disconnects and/or out-of-order sequences between startup of processes, and should be a much more robust solution.</p>
<p>From a configuration point of view it might be good also to add (these are only suggestions!):<br />3. Configure N3-T3 timers, initially it might be sufficient to rely on the host tcp stack config.<br />4. Configure disconnected-state strategy, -> <br /> 4.1. drop-old: drop oldest events, meaning we will continue to add new events to the circular-buffer.<br /> 4.2. drop-new: drop newest events, keeping the <strong>oldest</strong> records until reconnect.<br /> 4.3. store-to-disc: If in disconnected state, fallback to store events on local disc.</p>