[
https://issues.apache.org/jira/browse/EAGLE-2?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15011485#comment-15011485
]
Libin, Sun commented on EAGLE-2:
--------------------------------
[~ralphsu], thanks for pointing it out, I think you are right, we should use
API rather than run shell,
But when I try the example of SimpleConsumer from
https://cwiki.apache.org/confluence/display/KAFKA/0.8.0+SimpleConsumer+Example
The API return much different result from shell command result for some
partitions(For other partition, the results are close)
For example parition_0 of hdfs_audit_log topic:
shell comamnd return: 621021574
simple consumer api return 583080920
the difference is 621021534-583080920=37940614,
checked The consumer offset from zk is 621021574, which matches shell command
result
Not sure if the consumer api is not accurate or I didn't use it correctly, will
do some investigation on it
> watch message process backlog in Eagle UI
> -----------------------------------------
>
> Key: EAGLE-2
> URL: https://issues.apache.org/jira/browse/EAGLE-2
> Project: Eagle
> Issue Type: Improvement
> Environment: production
> Reporter: Edward Zhang
> Assignee: Hao Chen
> Original Estimate: 96h
> Remaining Estimate: 96h
>
> Message latency is a key factor for Eagle to enable realtime security
> monitoring. For hdfs audit log monitoring, kafka is used as datasource. So
> there is always some gap between current max offset in kafka and processed
> offset in eagle. The gap is the backlog which eagle should consume quickly as
> much as quickly. If the gap can be sampled for every minute or 20 seconds,
> then we understand if eagle is catching up or is lagging behind more.
> The command to get current max offset in kafka is
> bin/kafka-run-class.sh kafka.tools.GetOffsetShell --broker-list xxxx --topic
> hdfs_audit_log --time -1
> and Storm-kafka spout would store processed offset in zookeeper, in the
> following znode:
> /consumers/hdfs_audit_log/eagle.hdfsaudit.consumer/partition_0
> So technically we can get the gap and write that to eagle service then in UI
> we can watch the backlog
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
