[
https://issues.apache.org/jira/browse/BIGTOP-1754?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14358991#comment-14358991
]
Konstantin Boudnik commented on BIGTOP-1754:
--------------------------------------------
bq. I am a linux admin noob
Sorry dude - didn't mean to sound harsh. Every single one of us was and is a
noob in anything we start doing just yet. Appreciate your feedback, really!
It's hard to call it standard, but there are price to pay when {{/var}} is just
a directory under {{/}} instead of its own file system. The most common you
just pointed out: once the logs fill the all the space, your root directory
becomes 100% utilized which causes all sorts of problems for the system. For
the same reason, {{/tmp}} shouldn't be IMO co-located with {{/}}. As the
standard goes, IIRC most of the modern Linux distros will _nonetheless_
co-locate {{/var}} and {{/}} if you choose automatic partitioning option. And
it might work well for most of the desktop installations or even some server
deployments.
Another option is to limit the size of the logs and number of rotations Bigtop
components are dumping down. Which has different side-effect where you might
potentially miss an error message you'd catch otherwise.
It's always a balance... Sigh..
> Optional global for root log directory
> --------------------------------------
>
> Key: BIGTOP-1754
> URL: https://issues.apache.org/jira/browse/BIGTOP-1754
> Project: Bigtop
> Issue Type: Improvement
> Components: debian, rpm
> Affects Versions: 0.8.0
> Reporter: Rob Leidle
> Priority: Minor
> Fix For: backlog
>
> Original Estimate: 72h
> Remaining Estimate: 72h
>
> Right now most, if not all, Bigtop RPM packages are setup to log to
> /var/log/app where app is something like hue, hadoop-yarn, or hive. Often
> times systems are configured with small root partitions that can't handle
> this large degree of logging. However, it is very useful to have the logs
> always be in /var/log/hadoop-hdfs as this is an easy path to remember.
> The proposal is that we add a global variable for defining a logging root. So
> for example it could be defined as /mnt/log. If this variable is set then the
> directory will be created and a symlink in var log will be created. So in the
> case of hadoop-yarn:
> /var/log/hadoop-yarn -> /mnt/log/hadoop-yarn
> would be the symlink created. All RPM specs would be updated to support this
> global root logging location. For example in the hue RPM to define the log
> directory variable, the definition and symlink creation would be something
> like this:
> LOG_DIR=$LOG_ROOT/hue
> ln -sf /var/log/hue $LOG_DIR
> And of course if no LOG_ROOT was defined then the log directory /var/log/hue
> would be simply created as things currently work.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
