Elasticsearch按节点角色的滚动升级顺序咨询（5.x版本内升级）

Hey there! Great question about rolling upgrades for your Elasticsearch cluster—let me walk you through the optimal order and best practices tailored to your node roles:

Recommended Upgrade Sequence #

Follow this order to minimize downtime and cluster disruption:

1. Master-eligible nodes (3 nodes)

   • This is the correct starting point, but you must upgrade them one at a time. Master nodes manage cluster state, so you need to maintain a quorum (at least 2 healthy master nodes) throughout the process to avoid cluster instability.
   • Steps for each master node:
     • Stop the Elasticsearch service on the node.
     • Install the new 5.x version.
     • Start the service and wait for the node to rejoin the cluster.
     • Verify cluster health is green (run curl -XGET '_cluster/health?pretty') and the node is listed in curl -XGET '_cat/nodes?v' before moving to the next master node.

2. Client nodes (2 nodes)

   • Client nodes are stateless and handle user request routing, so they’re safe to upgrade next. Stick to upgrading one at a time:
     • Stop the node, install the new version, then restart it.
     • Since client nodes don’t store data or manage cluster state, they’ll quickly rejoin the cluster and resume handling traffic with minimal impact.

3. Ingest nodes (4 nodes)

   • Ingest nodes handle data pre-indexing processing and are also stateless. Upgrade them one by one:
     • When you take an ingest node down, pending ingest tasks will automatically be routed to other healthy ingest nodes in the cluster.
     • After restarting the upgraded node, confirm it’s ready to handle ingest requests before moving to the next one.

4. Data nodes (4 nodes)

   • Data nodes store your indexed data, so this is the most critical step—take it slow and verify after each upgrade:
     • Upgrade one data node at a time. Stop the node, install the new version, then restart it.
     • Wait for the node to rejoin the cluster and for the cluster health to return to green (this ensures any shard replicas are reallocated if needed).
     • Avoid heavy indexing or maintenance tasks during this phase to reduce cluster load.

Key Pre-Upgrade & During-Upgrade Tips #

• Backup first: Even for minor 5.x to 5.x upgrades, take a snapshot of your important indexes using the snapshot API—better safe than sorry.
• Check plugin compatibility: Run elasticsearch-plugin list on each node before upgrading to ensure all installed plugins are compatible with the target 5.x version (most minor version updates maintain plugin compatibility, but double-check).
• Monitor closely: Keep an eye on cluster health and node status throughout the process using Kibana (if available) or the Elasticsearch API endpoints mentioned earlier.

内容的提问来源于stack exchange，提问作者alesnav