Daniel Templeton created YARN-4394:
--------------------------------------
Summary: ClientServiceDelegate doesn't handle retries during AM
restart as intended
Key: YARN-4394
URL: https://issues.apache.org/jira/browse/YARN-4394
Project: Hadoop YARN
Issue Type: Bug
Reporter: Daniel Templeton
Assignee: Daniel Templeton
In the {{invoke()}} method, I found the following code:
{code}
private AtomicBoolean usingAMProxy = new AtomicBoolean(false);
...
// if it's AM shut down, do not decrement maxClientRetry as we wait for
// AM to be restarted.
if (!usingAMProxy.get()) {
maxClientRetry--;
}
usingAMProxy.set(false);
{code}
When we create the AM proxy, we set the flag to true. If we fail to connect,
the impact of the flag being true is that the code will try one extra time,
giving it 400ms instead of just 300ms. I can't imagine that's the intended
behavior. After any failure, the flag will forever more be false, but
fortunately (?!?) the flag is otherwise unused.
Looks like I need to do some archeology to figure out how we ended up here.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
