Impala, TThreadPoolServer: TServerTransport died on accept: SSL_accept: error code: 0

[#1]

Hi ,

After install a load balancer with ssl for impala, all the impala deamons are generating many messages on each log with the following description : "INFO thrift-util.cc:109 TThreadPoolServer: TServerTransport died on accept: SSL_accept: error code: 0"

we can have a lot messages per second, and it never stops.

This message is part of a health test or something like that ?

What can i do to check something wrong with my conection to the load balancer?

Log example :

Apr 6, 10:10:53.549 AM INFO thrift-util.cc:109 TThreadPoolServer: TServerTransport died on accept: SSL_accept: error code: 0
Apr 6, 10:10:53.937 AM INFO thrift-util.cc:109 TThreadPoolServer: TServerTransport died on accept: SSL_accept: error code: 0
Apr 6, 10:10:54.037 AM INFO thrift-util.cc:109 TThreadPoolServer: TServerTransport died on accept: SSL_accept: error code: 0

[#2]

This usually means that SSL is configured badly - is your load balancer and / or client configured to make an SSL connection to Impala?

The repeated connections are a mystery - they're coming from the client RPC port. So maybe something is using your load balancer repeatedly? Or is the LB itself attempting to create a set of connections (maybe to warm up its connection pool)?

[#3]

Hi Henry,

Thanks for your suppport and interest.

These are the answers for your questions:

1. Yes, we are trying to make an SSL connection to Impala, however , we don't know if we need to change something. Do you know some test or steps to check this ?

2. We don't now if a client are trying to check the connection, so, how i can check this ? I can't find an other log with more information or clues to identify the problem.

3.The LB does not have any health test or something like that, only manage the conections from/to CDH.

[#4]

You will need to tell your client (is it shell, JDBC or something else?) to use SSL. Have you read http://www.cloudera.com/documentation/archive/impala/2-x/2-0-x/topics/impala_ssl.html? The instructions for the shell are at the bottom of that pa

[#5 · Accepted solution]

Hi,

We reviewed the configuration of the Impala service and removed an extraneous role group containing only a single Impala daemon node. After restarting the Impala service and re-deploying the client configurations, ODBC clients connected successfully through the HAProxy load balancer .

 Cloudera support  help me with this issue reported on the case 91228.

Thanks