[
https://issues.apache.org/jira/browse/TEZ-2774?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14735829#comment-14735829
]
Bikas Saha commented on TEZ-2774:
---------------------------------
I will reiterate my initial comments about the logs I have mentioned. Let us
keep them there at info level. Logging at debug level is not always useful
because it may be an intermittent issue or too expensive to reproduce. Some of
the removed logs like which attempt succeeded and which got killed will only be
relevant during speculation and not logged all the time. Removing the kill log
is not going to reduce general logging size but will be sorely missed if there
is a bug in speculation. The task scheduler logs for containers have been
useful when debugging why min-held or pre-warm containers were not working as
expected. The input-spec log is used when figuring out how many events should
have actually come to an input and how many actually did for debugging hung
jobs. Then correlating it with event routing log from the AM to see where the
error is. The event routing log from the AM was removed in the initial patch
while the the next version introduces it back but continues to hide the task
spec log. This seems arbitrary.
It is hard to explain all the debugging context from previous experience.
Hence, my suggestion is to accept the comments as is instead of debating them
subjectively. With time we can remove more logs. But I am quite wary of doing
such a bulk removal and not having logs to debug when I need them.
> Reduce logging in the AM, and parts of the runtime
> --------------------------------------------------
>
> Key: TEZ-2774
> URL: https://issues.apache.org/jira/browse/TEZ-2774
> Project: Apache Tez
> Issue Type: Sub-task
> Reporter: Siddharth Seth
> Assignee: Siddharth Seth
> Attachments: TEZ-2774.1.txt, TEZ-2774.2.txt
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)