helix-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Zhen Zhang <zzh...@linkedin.com>
Subject RE: race condition in Helix message handler factory registration
Date Fri, 14 Nov 2014 19:02:16 GMT
Hi Vlad, thanks. I created a jira to track this:


From: vlad.gm@gmail.com [vlad.gm@gmail.com]
Sent: Thursday, November 13, 2014 5:17 PM
To: user@helix.apache.org
Subject: race condition in Helix message handler factory registration

I have a small issue when running a Helix cluster in Docker (all on the same machine, very
small RTT):

I register two message handler factories, similarly to the way it is done in BootstrapProcess.java
example. The manager is not yet connected. My understanding is that this places the handlers
in a hashmap from which they will be read and registered later, upon connect.

Upon connect, I see that the DefaultMessagingService.onConnected function is called and Helix
starts registering the message handlers. However, messages start arriving already and HelixTaskExecutor.createMessageHandler
is called, before all handlers have been registered. The result is that I get NullPointerExceptions
for the missing handlers in the following line of HelixTaskExecutor.createMessageHandler:

MessageHandlerFactory handlerFactory = item.factory();

due to the fact that there is no item (MessageHandlerFactoryRegistryItem) for that message
type registered yet.

I came upon this bug after updating the git code from an old snapshot of 0.7.1. I saw that
the relevant code changes are from around the summer.


View raw message