Briefly, this error occurs when Elasticsearch’s machine learning feature fails to distribute its anomaly detection models evenly across the available nodes. This could be due to insufficient resources, network issues, or configuration problems. To resolve this, you can try increasing the resources (CPU, memory) of your nodes, check your network for any connectivity issues, or review your Elasticsearch configuration to ensure it allows for proper rebalancing. Additionally, you can manually move models between nodes if necessary.
This guide will help you check for common problems that cause the log ” failed to rebalance models ” to appear. To understand the issues related to this log, read the explanation below about the following Elasticsearch concepts: plugin, rebalance.
Overview
Cluster rebalancing is the process by which an Elasticsearch cluster distributes data across the nodes. Specifically, it refers to the movement of existing data shards to another node to improve the balance across the nodes (as opposed to the allocation of new shards to nodes). Usually, it is a completely automatic process that requires no outside intervention. However, there are a number of parameters Elasticsearch uses to regulate this process.
Examples
The command below will establish the cluster settings to enable automatic cluster rebalancing. It is not necessary to run the command (the values used are in fact the defaults).
PUT /_cluster/settings?flat_settings=true
{
"transient" : {
"cluster.routing.rebalance.enable": "all",
"cluster.routing.allocation.allow_rebalance": "indices_all_active" ,
"cluster.routing.allocation.cluster_concurrent_rebalance":"2"
}
}Notes and good things to know
In general, the cluster rebalance settings have sensible defaults. It is generally not advisable to disable cluster rebalancing. It is usually most sensible to wait until indices are all active before rebalancing since we consider the highest priority to be recovering the indices rather than moving them around. Finally, it is recommended to limit the number of concurrent rebalances to 2 (the default) since having a large number of shards moving around at a given time can use a lot of resources resource and cause instability. Increasing this number would only make sense on large clusters.
You can consider the “rebalance” process to be a tendency to spread the total number of shards across all nodes in the cluster, and also to spread the total number of shards for a given index as evenly as possible across the cluster. The rebalance is a “soft” algorithm, and will be overruled by other “hard” factors such as disk-based or shard allocation awareness.
If you think your cluster is not rebalancing as it should first check the “hard” limits you have on shard allocation awareness or disk-based shard allocation before tweaking the rebalance parameters.
Manual rebalancing
It is also possible to rebalance manually using a command like this:
POST /_cluster/reroute?dry_run=true
{
"commands" : [
{
"move" : {
"index" : "test", "shard" : 0,
"from_node" : "node1", "to_node" : "node2"
}
}
]
}It is advisable to include the dry_run parameter to check the result of your action, and if everything is in order then repeat the command with dry_run=false.
Bear in mind that if you rebalance manually, Elasticsearch may move the same (or another shard) back automatically, compensating for your previous action. Similarly, there may be constraints that will prevent your reallocation from being accepted by the cluster.
Log Context
Log “failed to rebalance models” classname is TrainedModelAssignmentClusterService.java.
We extracted the following from Elasticsearch source code for those seeking an in-depth context :
rebalanceReason.get();
ActionListener.wrap(
newMetadata -> logger.debug(
() -> format("rebalanced model assignments [%s]"; Strings.toString(newMetadata; false; true))
);
e -> logger.warn("failed to rebalance models"; e)
)
);
}
}
