Logging config dramatically impacts throughput

[Tom Zeller]

> On Sep 28, 2015, at 11:06 AM, Marvin Addison <marvin.addison at gmail.com> wrote:
> 
> Maybe we should switch the defaults to it.
> 
> +1 for IDP_PROCESS appender.

Sounds like we need a JIRA, feel free to assign it to me.

> The only downside is that it could increase likelihood of losing log messages in the worst case, e.g. VM crash before dispatch queue is drained. I opted for discardingThreshold=0 in our environment to prevent losing messages when the queue is nearing its capacity, which is the other notable situation where you can lose messages. With those caveats in mind, I felt like it was a win for IDP_PROCESS where we can tolerate losing logs under worst case. In contrast, I avoided the async appender for IDP_AUDIT where we don't want to increase risk of losing data. Also, there's not as much volume in the audit log (ours is 1/15 the size idp-process.log at INFO, much less at DEBUG), so less to gain.

Just a comment that I think the default configuration should not be lossy, in general.