[ 
https://issues.apache.org/jira/browse/MESOS-3348?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Joseph Wu updated MESOS-3348:
-----------------------------
    Description: 
Tasks currently log their output (i.e. stdout/stderr) to files (the "sandbox") 
on an agent's disk.  In some cases, the accumulation of these logs can 
completely fill up the agent's disk and thereby kill the task or machine.

To prevent this, we should either implement a log rotation mechanism or 
capped-size logging.  This would be used by executors to control the amount of 
logs they keep.  

Master/agent logs will not be affected.

We will first scope out several possible approaches for log rotation/capping in 
a design document (see [MESOS-3356]).  Once an approach is chosen, this story 
will be broken down into some corresponding issues.

  was:
Tasks currently log their output (stdout/stderr) to files on an agent's disk.  
In some cases, the accumulation of these logs can completely fill up the 
agent's disk and thereby kill the task or machine.

To prevent this, we should either implement a log rotation mechanism or 
capped-size logging.

We will first scope out several possible approaches for log rotation/capping in 
a design document (see [MESOS-3356]).  Once an approach is chosen, this story 
will be broken down into some corresponding issues.


> Add either log rotation or capped-size logging (for tasks)
> ----------------------------------------------------------
>
>                 Key: MESOS-3348
>                 URL: https://issues.apache.org/jira/browse/MESOS-3348
>             Project: Mesos
>          Issue Type: Story
>    Affects Versions: 0.23.0
>            Reporter: Joseph Wu
>            Assignee: Joseph Wu
>              Labels: mesosphere
>
> Tasks currently log their output (i.e. stdout/stderr) to files (the 
> "sandbox") on an agent's disk.  In some cases, the accumulation of these logs 
> can completely fill up the agent's disk and thereby kill the task or machine.
> To prevent this, we should either implement a log rotation mechanism or 
> capped-size logging.  This would be used by executors to control the amount 
> of logs they keep.  
> Master/agent logs will not be affected.
> We will first scope out several possible approaches for log rotation/capping 
> in a design document (see [MESOS-3356]).  Once an approach is chosen, this 
> story will be broken down into some corresponding issues.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to