CODEWITHSUNDEEP
elasticsearchelasticsearch-5
user553965
user553965

Reputation: 1281

Unable to move shards using cluster.routing.allocation

I have an ElasticSearch cluster with several indices on 2 data nodes (es-data-0 & es-data-1) and want to move all shards off of node es-data-1 before decommissioning it.

Moving shards 1 at a time works well. The following command takes several seconds to move the shard. 

POST /_cluster/reroute
{
    "commands": [
        {
            "move": {
                "index" : "index_operations_log",
                "shard" : 0,
                "from_node" : "es-data-1",
                "to_node" : "es-data-0"
            }
        }
    ]
}

But if I try to do cluster-level shard allocation filtering, it does not affect. For example, the following has no apparent effect on shard status:

PUT /_cluster/settings
{
    "transient" : {
        "cluster.routing.rebalance.enable": "none"
    }
}

PUT /_cluster/settings
{
    "transient": {
        "cluster.routing.allocation.exclude._name": "es-data-1"
    }
}

even though it returns

{
    "acknowledged": true,
    "persistent": {},
    "transient": {
        "cluster": {
            "routing": {
                "allocation": {
                    "exclude": {
                        "_name": "es-data-1"
                    }
                }
            }
        }
    }
}

What am I missing?

Upvotes: 1

Views: 1478

Answers (1)

user553965
user553965

Reputation: 1281

Figured it out. I ran

GET /_cluster/settings

and saw that I had set some cluster.routing.allocation settings from earlier that conflicted with these new ones. I cleared the conflicting rules by setting their values to "" and the shards started moving over.

In general, the

PUT /_cluster/settings
{
    "transient" : {
        "cluster.routing.allocation.require": "..."
    }
}

call doesn't report errors, so I've found the only way to troubleshoot issues like above is - if the shards aren't moving as expected, try moving the shards 1 at a time using POST /_cluster/reroute. This reports detailed errors. Then, if you're able to move individual shards with POST /_cluster/reroute but cluster- or index-level shard moving still isn't working, use GET /_cluster/settings and GET /*/settings
to check for other existing routing allocation rules that conflict. If they exist, they can be reset by doing PUT /_cluster/settings with their values = "".

Upvotes: 2

Related Questions