Zookeeper clients should wait longer for master

Description

I'm setting up Tachyon (v0.8.2-rc1) on top of Mesos and the clients don't wait long enough for the master, increasing MAX_TRY in common/src/main/java/tachyon/LeaderInquireClient.java from 10 to 30 seems to help with this.

However the stack trace if you don't do this isn't ideal (see below) - it'd be good if it behaved similarly to if the master wasn't responding (a more common case I suspect?) - i.e. retries (forever?)

{{2015-11-09 12:37:17,401 INFO WORKER_LOGGER (ClientBase.java:connect) - Tachyon client (version ${project.version}) is trying to connect with BlockMaster master @ null
2015-11-09 12:37:17,403 ERROR WORKER_LOGGER (TachyonWorker.java:main) - Failed to initialize the block worker, exiting.
java.lang.NullPointerException
at tachyon.util.network.NetworkAddressUtils.getFqdnHost(NetworkAddressUtils.java:468)
at tachyon.security.authentication.AuthenticationUtils.createTSocket(AuthenticationUtils.java:106)
at tachyon.security.authentication.AuthenticationUtils.getClientTransport(AuthenticationUtils.java:81)
at tachyon.ClientBase.connect(ClientBase.java:115)
at tachyon.client.WorkerBlockMasterClient.getId(WorkerBlockMasterClient.java:101)
at tachyon.worker.WorkerIdRegistry.registerWithBlockMaster(WorkerIdRegistry.java:59)
at tachyon.worker.block.BlockWorker.<init>(BlockWorker.java:200)
at tachyon.worker.TachyonWorker.main(TachyonWorker.java:42)
at tachyon.mesos.TachyonWorkerExecutor$1.run(TachyonWorkerExecutor.java:71) }}

Environment

0.8.2-rc1

Assignee

Unassigned

Reporter

Adrian Bridgett

Labels

Components

Affects versions

Priority

Minor