Index Replication ends up in a cluster with yellow state


(Niko) #1

Hi,
we have created an index with version 0.17.1. After finishing the indexing
process the cluster was in green state and then we tried to create more
replicas (4) with the following call:

curl -XPUT 'localhost:9200/my_index/_settings' -d '
{
"index" : {
"number_of_replicas" : 4
}
}

From this moment the cluster changed to yellow state and all of the new
created shards remain unassigned. The index is still questionable but has
none replica.

Do you have any idea what happened?

Thanks in advance,

Niko


(Shay Banon) #2

How many nodes are you running? If you increase the number of replicas, then
you need to have enough nodes for them to be allocated on (as a shard and
its replica will not be allocated on the same node).

As a side note, when you increase the replica count, and you do have enough
nodes for them to be allocated, the cluster will still be in a yellow state,
until those replicas will recover the data and move to active mode.Yellow
state means all primary shards are allocated, but not all replicas are
allocated.

On Fri, Aug 5, 2011 at 1:46 PM, Niko Gross niko.gross@googlemail.comwrote:

Hi,
we have created an index with version 0.17.1. After finishing the indexing
process the cluster was in green state and then we tried to create more
replicas (4) with the following call:

curl -XPUT 'localhost:9200/my_index/_settings' -d '
{
"index" : {
"number_of_replicas" : 4
}
}

From this moment the cluster changed to yellow state and all of the new
created shards remain unassigned. The index is still questionable but has
none replica.

Do you have any idea what happened?

Thanks in advance,

Niko


(Niko) #3

We're running the cluster with 21 nodes, so there should be enough nodes. But how long is the time the replicas need to recover the data and move to active mode. I thought this would be an almost instant process.

Am 05.08.2011 um 21:42 schrieb Shay Banon:

How many nodes are you running? If you increase the number of replicas, then you need to have enough nodes for them to be allocated on (as a shard and its replica will not be allocated on the same node).

As a side note, when you increase the replica count, and you do have enough nodes for them to be allocated, the cluster will still be in a yellow state, until those replicas will recover the data and move to active mode.Yellow state means all primary shards are allocated, but not all replicas are allocated.

On Fri, Aug 5, 2011 at 1:46 PM, Niko Gross niko.gross@googlemail.com wrote:
Hi,
we have created an index with version 0.17.1. After finishing the indexing process the cluster was in green state and then we tried to create more replicas (4) with the following call:

curl -XPUT 'localhost:9200/my_index/_settings' -d '
{
"index" : {
"number_of_replicas" : 4
}
}

From this moment the cluster changed to yellow state and all of the new created shards remain unassigned. The index is still questionable but has none replica.

Do you have any idea what happened?

Thanks in advance,

Niko


(Shay Banon) #4

New replicas require replicating the data (per shard). That date needs to be
transferred to the new replica shards, and the time it takes is the time it
will take to transfer that data, and depends mainly on the size of the
shards.

On Sat, Aug 6, 2011 at 4:20 PM, Nikolas Gross niko.gross@googlemail.comwrote:

We're running the cluster with 21 nodes, so there should be enough nodes.
But how long is the time the replicas need to recover the data and move to
active mode. I thought this would be an almost instant process.

Am 05.08.2011 um 21:42 schrieb Shay Banon:

How many nodes are you running? If you increase the number of replicas,
then you need to have enough nodes for them to be allocated on (as a shard
and its replica will not be allocated on the same node).

As a side note, when you increase the replica count, and you do have enough
nodes for them to be allocated, the cluster will still be in a yellow state,
until those replicas will recover the data and move to active mode.Yellow
state means all primary shards are allocated, but not all replicas are
allocated.

On Fri, Aug 5, 2011 at 1:46 PM, Niko Gross niko.gross@googlemail.comwrote:

Hi,
we have created an index with version 0.17.1. After finishing the indexing
process the cluster was in green state and then we tried to create more
replicas (4) with the following call:

curl -XPUT 'localhost:9200/my_index/_settings' -d '
{
"index" : {
"number_of_replicas" : 4
}
}

From this moment the cluster changed to yellow state and all of the new
created shards remain unassigned. The index is still questionable but has
none replica.

Do you have any idea what happened?

Thanks in advance,

Niko


(system) #5