Description of problem:
When using fluentd (0.12.x using ruby 2.0 in RHEL7), the memory usage (as measured by ps h -p $fluentdpid -o rss) grows steadily until it runs out of memory and crashes or is OOMkilled.  This seems mostly exacerbated when using fluentd as a listener (using the secure_forward protocol) of other fluentd collectors.

Version-Release number of selected component (if applicable):
RHEL 7 - ruby 2.0 - fluentd 0.12.x

How reproducible:
Every time

Steps to Reproduce:
1. Deploy OpenShift + logging + mux component
2. Put a load on fluentd e.g. logger -f file_with_many_lines
3. Measure rss of the fluentd mux

Actual results:


Expected results:
I would expect the memory usage to eventually plateau instead of growing constantly.

Additional info:
The mux case is the easiest and fastest way to show the memory growth.

If using OCP 3.7 or later, the fluentd is bundled with jemalloc - set the environment variable USE_JEMALLOC=false to disable jemalloc and use glibc malloc.