Creating brand new index gets index already exists exception

What I've noticed is after my cluster (0.19.12) has been running for a few
days, trying to create a brand new index fails with an index already exists
exception. This is a brand new index, with a different, unique name. The
"type" is the same. So I might have:

index_a, type book mapping
index_b, type book mapping

then I try to create index_c, type book mapping, and get an index already
exists exception. Using the "head" plugin, I see the index gets created,
but it exists in a funky state. The primary seems OK, but the replica is
"unassigned".

During this time, "head" reports the cluster as red. But queries against
existing indexes seem OK.

I manually delete the bad index_c then restart the cluster. Now my code can
create index_c. A few minutes later, it can also create index_d.

Only thing that seems consistent is that the cluster

  1. has been running for at least a few days. It's like after a few days the
    cluster has some issues?
  2. restarting the cluster solves the issue

Obviously I don't want to restart the cluster because I'm trying to
automatically create these indexes....

Any ideas?

-T

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

You might have had a split-brain scenario where different nodes in the
cluster were forming sub-clusters. How many nodes do you have and what is
your replica count for those indexes?

Cheers,

Ivan

On Sat, Aug 24, 2013 at 12:09 PM, Tinou Bao tinou.bao@gmail.com wrote:

What I've noticed is after my cluster (0.19.12) has been running for a few
days, trying to create a brand new index fails with an index already exists
exception. This is a brand new index, with a different, unique name. The
"type" is the same. So I might have:

index_a, type book mapping
index_b, type book mapping

then I try to create index_c, type book mapping, and get an index already
exists exception. Using the "head" plugin, I see the index gets created,
but it exists in a funky state. The primary seems OK, but the replica is
"unassigned".

During this time, "head" reports the cluster as red. But queries against
existing indexes seem OK.

I manually delete the bad index_c then restart the cluster. Now my code
can create index_c. A few minutes later, it can also create index_d.

Only thing that seems consistent is that the cluster

  1. has been running for at least a few days. It's like after a few days
    the cluster has some issues?
  2. restarting the cluster solves the issue

Obviously I don't want to restart the cluster because I'm trying to
automatically create these indexes....

Any ideas?

-T

--
You received this message because you are subscribed to the Google Groups
"elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

I have 2 master nodes, 2 data nodes. I set replica count to 1. I was
concerned with the split-brain issue, so I did set

discovery.zen.minimum_master_nodes: 2

in the conf files for master and data nodes. Figured if network
partitioned, then I would get some other error, like "not enough masters."

-T

On Saturday, August 24, 2013 12:15:45 PM UTC-7, Ivan Brusic wrote:

You might have had a split-brain scenario where different nodes in the
cluster were forming sub-clusters. How many nodes do you have and what is
your replica count for those indexes?

Cheers,

Ivan

On Sat, Aug 24, 2013 at 12:09 PM, Tinou Bao <tino...@gmail.com<javascript:>

wrote:

What I've noticed is after my cluster (0.19.12) has been running for a
few days, trying to create a brand new index fails with an index already
exists exception. This is a brand new index, with a different, unique name.
The "type" is the same. So I might have:

index_a, type book mapping
index_b, type book mapping

then I try to create index_c, type book mapping, and get an index already
exists exception. Using the "head" plugin, I see the index gets created,
but it exists in a funky state. The primary seems OK, but the replica is
"unassigned".

During this time, "head" reports the cluster as red. But queries against
existing indexes seem OK.

I manually delete the bad index_c then restart the cluster. Now my code
can create index_c. A few minutes later, it can also create index_d.

Only thing that seems consistent is that the cluster

  1. has been running for at least a few days. It's like after a few days
    the cluster has some issues?
  2. restarting the cluster solves the issue

Obviously I don't want to restart the cluster because I'm trying to
automatically create these indexes....

Any ideas?

-T

--
You received this message because you are subscribed to the Google Groups
"elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to elasticsearc...@googlegroups.com <javascript:>.
For more options, visit https://groups.google.com/groups/opt_out.

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

Realize that the master nodes publish to a virtual IP while the data nodes
servers are not setup with any virtual IPs. Wondering if this is a
potential issue. Does zen discovery have issues with VIPs?

-T

On Saturday, August 24, 2013 12:09:03 PM UTC-7, Tinou Bao wrote:

What I've noticed is after my cluster (0.19.12) has been running for a few
days, trying to create a brand new index fails with an index already exists
exception. This is a brand new index, with a different, unique name. The
"type" is the same. So I might have:

index_a, type book mapping
index_b, type book mapping

then I try to create index_c, type book mapping, and get an index already
exists exception. Using the "head" plugin, I see the index gets created,
but it exists in a funky state. The primary seems OK, but the replica is
"unassigned".

During this time, "head" reports the cluster as red. But queries against
existing indexes seem OK.

I manually delete the bad index_c then restart the cluster. Now my code
can create index_c. A few minutes later, it can also create index_d.

Only thing that seems consistent is that the cluster

  1. has been running for at least a few days. It's like after a few days
    the cluster has some issues?
  2. restarting the cluster solves the issue

Obviously I don't want to restart the cluster because I'm trying to
automatically create these indexes....

Any ideas?

-T

--
You received this message because you are subscribed to the Google Groups "elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email to elasticsearch+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.