13880bd8c1
When a node dies that carries a watcher shard or a shard is relocated to another node, then watcher needs not only trigger a reload on the node where the shard relocation happened, but also on other nodes where copies of this shard, as different watches may need to be loaded. This commit takes the change of remote nodes into account by not only storing the local shard allocation ids in the WatcherLifeCycleService, but storing a list of ShardRoutings based on the local active shards. This also fixes some tests, which had a wrong assumption. Using `TestShardRouting.newShardRouting` in our tests for cluster state creation led to the issue of always creating new allocation ids which implicitely lead to a reload. |
||
|---|---|---|
| .. | ||
| licenses | ||
| src | ||
| build.gradle | ||