N1QL index not found error code 5000

Hi,

When running our integration tests against a single Couchbase server 4.0 GA with the latest JAVA client 2.0.0, I will have some random failure of these tests with the following message:
[[{“msg”:“index is deleted or node hosting index is down - cause: Index not found”,“code”:5000}]]

The index is there and has not been deleted (other integration tests do use the same index) - in my case this is the primary index for my query.

So I am wondering if there are some communications issues between the different processes (query and index and data) under load which are exposed through this error message (node hosting index is down). Is there any configuration possible to allow better communication timeout?

I have configured my java client to use timeout of 5 minutes during integration tests, but can see this error returning in only few seconds!

I have already changed running my integration tests single threaded to avoid loading the single node but I still get this error…

Anyone has seen this issue? Any idea?

Many thanks.

Could you send out the “select * from system:indexes;”. Are you also able to see the index reported under the index tab on admin console?
is this a transient error (happens sometimes but goes away) or is this a permanent error?
thanks
-cihan

Hi,

This is a transient error that appears when running our integration tests against a one node couchbase node having the 3 roles, query, data and index. So this will happen at different point during running the tests, sometimes all the tests will pass and not fail!

This is why I was wondering of it was due to load and the indexer node being detected as down transiently - and this is leading me to ask if there are nobs to turn to test this assumption?

The indexes are there, already created before running the tests, the N1QL query and the UI return the same information about the indexes all marked at 100% available.

Regards.
Thanks.

if you are hitting the limits of the computational capacity of the node, that may very well delay the indexer response and make it appear as down. cb_collecinfo would be best but if not, Could you share the machine specs and settings form the settings tab in the admin console?
thanks
-cihan

Sorry for taking so long to reply…

I will spend sometime gathering the information needed…

But I have new issues on our build server where we run integration tests… We are deleting all the buckets (which deletes all indexes associated) and then recreating them before running the tests, and doing this lead to a lot of failure in integration tests with the number of elements returned using N1QL queries and indexes incorrect so random integration tests failures…

Is there a better approach or a way to make sure indexes will be in the right state before we start running integration tests?