as the num of plugins grown up,we need to restart the cluster the install new plugins or replace new version of the plugins,it is not cool,if we can do this on the fly,that will be great.
may be we can install plugin like this:
curl -XPOST http://localhost:9200/_plugin/install/medcl/analysis-string2int/1.0
so,after that,it is suggested that the plugin will be installed all over the cluster-wide,no need to setup one by one,and without shard relocation.
great!
+1
Although we understand how this could make things easier, we are a bit concerned that it would add a lot of complexity (dynamic classloading, how to deal with nodes that are down or have different version, etc.). We would rather recommend on using orchestration tools like chef or puppet to ensure consistent installation of plugins across all nodes of the cluster.
Perhaps plugin list should distinguish between installed and loaded plugins and those which were not loaded yet.
+1
Can I install a plugin without restarting the cluster? The cost of a cluster restart is too high.
No. You need to restart the cluster. See rolling restart in docs.
Can I install a plugin without restarting the cluster? The cost of a cluster restart is too high.
Installing a plugin is a rare event.
Most helpful comment
Perhaps
plugin listshould distinguish between installed and loaded plugins and those which were not loaded yet.