Couchbase indexes

Couchbase Server
1

Hello,
I have created an index with a replica and distributed the index on all index nodes. But I see that the index size is not evenly distributed. I see the index is occupying 90% od disk on 1 node and on other 0 bytes even though I specified num partitions 8 and all index nodes on index definition.

Thanks
Bhanu

2

index definition like below.

CREATE INDEX count_query ON ` (*****) WITH { “defer_build”:true, “nodes”:[ “node1.com:8091”,“node2.com:8091”,node3.com:8091",“node4.com:8091” ], “num_replica”:1, “num_partition”:8 }

3

@banumyname,

From your index definition, I do not see any partition by clause. So there will only be one partition. As “num_replica” is specified as “1”, there will be two indexes in the system. From the problem description, I understand that one index is using 90% of disk while the other index is not using any disk.

Which version of couchbase-server are you using? Can you kindly share the indexer logs from both the nodes where the indexes reside so that we can analyse the issue further

Thanks,
Varun

4
5

Hello Varun, thanks for replying back.

I am using partitioning by hash. I missed to post it while editing the details earlier.

PARTITION BY hash(entry_type) WHERE .

We are using couchbase 6.0.3

Thanks
Bhanu

6

It depends on values of entry_type. You can also try multiple keys PARTITION BY hash( entry_type , META().id)

7

Hi, It shouldnt matter on which key i am partitioning right ?

8

image
image.png905×159 10.3 KB

Even if i partition by different key, looks like it is still uneven. this causing the sizing constraints.

9

It shouldnt matter on which key i am partitioning right

It depends on your data distribution. If let’s say your entry_type has more documents on type “user”, then they all go to one partition and can result in larger data size.

Can you collect the partition level stats using curl -u <username:password> http://localhost:9102/stats?partition=true from each of the indexer nodes and share the details?

Thanks,
Varun