Jun, First, in this case, one developer has a single broker running, but another was doing some work locally without a broker. This is how they found the issue.
Second, this issue is with the KafkaLog4jAppender, so from what I can tell, we have to use a broker list. I suggested that they do test this to ensure it works. Lastly, and most importantly, a workaround was created, but it involved wrapping each call to the logger in a try/catch block. This really isn't a preferred practice since it will litter the code with these. I'm not sure what the correct solution is. Ideally, it would be nice if these errors were logged somewhere on the file system while the broker is down. -----Original Message----- From: Jun Rao [mailto:[email protected]] Sent: Tuesday, December 11, 2012 12:35 PM To: [email protected] Subject: Re: Kafka Log4J appender closes all comms If he/she uses more than than 1 broker and uses either a zk-based producer or a broker list based producer with a vip, it should be able to proceed when one of the brokers is down. Thanks, Jun On Tue, Dec 11, 2012 at 6:58 AM, Sybrandy, Casey < [email protected]> wrote: > Hello, > > A coworker is having an issue where he's writing a plugin for > ElasticSearch and is using the Kafka Log4J appender to log errors to a > topic. However, if for some reason the broker is missing, the whole > server stops responding to requests. Is this related to what is > disucussed in the link below? Is this a known behaviour? Is it something > we're doing wrong? > Just wanted to ask before we reported a new bug. > > > http://mail-archives.apache.org/mod_mbox/incubator-kafka-dev/201209.mb > ox/%3C1438326217.82968.1347663007581.JavaMail.jiratomcat@arcas%3E >
