Customer has uncontrolled memory growth eventually leading to MaxScale exceeding its memory allocation and crashing.
Problem is not reproducible outside customer's lab & production container environments. Customer is also unable to reproduce the problem in virtualization or bare metal environments. Customer states problem happens in "idle" state, where "idle" means between 55 and 250qps
Successive efforts have ruled out
MXS-4100 (non-zero connection_keepalive) and made an effort to minimize memory consumption by limiting write q high water mark as well as query classifier cache size. None of these changes have made a difference.
Customer was shown how to run traces with jemalloc, added that to the container, and saw no change. two svg reports from different heap trace sessions attached.