[
https://issues.apache.org/jira/browse/HADOOP-3753?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12613437#action_12613437
]
Craig Macdonald commented on HADOOP-3753:
-----------------------------------------
I'm open to enlarging the requirements of this issue.
However, my 2c is that I want to have a scoreboard file that I can easily point
a script at to read (in this case collect and feed into RRD tool).
IMHO, Log rotations etc can be handled externally - e.g. see logrotate(8) on
linux. Overwriting should not loose data as the file is a scoreboard - it has
current values of the counters. The metrics should continue to emit updated
values, the client application reads the scoreboard file when it feels the need
(and possibly store the values in a time-indexed format externally).
This is how the counters etc in /proc on linux work. For example,
/proc/net/snmp is read every 5 minutes. It's one (virtual) scoreboard file,
which is kept updated by the kernel (but it could have equally been implemented
as being dumped out once a second/minute etc).
> metrics: FileContext support overwrite mode
> -------------------------------------------
>
> Key: HADOOP-3753
> URL: https://issues.apache.org/jira/browse/HADOOP-3753
> Project: Hadoop Core
> Issue Type: Improvement
> Components: metrics
> Reporter: Craig Macdonald
> Priority: Minor
>
> FileContext currently continually appends to the metrics log file(s),
> generating an ever lengthening file.
> In some scenarios, it would be useful to simply write the current statistics
> to the file once every period, then overwrite the file for the next period.
> For instance, this could be useful if an external application parsed the
> metrics output - e.g. Cacti to create realtime graphs.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
