Cause is:

1. The symlink for jemalloc is the devel package for both CentOS and Ubuntu. In my tests the symlink is installed but in production it wouldn't be
2. If jemalloc won't load there is a shell error when ColumnStore processes execute commands
3. If there are any shell errors in logs the replication commands think they have failed

To fix this:

1. We need to detect at compile time if libjemalloc.so.1 or libjemalloc.so.2 is used (get symlink location?)
2. Test if library exists / is installed before LD_PRELOAD is set in columnstore_run.sh

On to the next problems:

1. server_id and log_bin. Don't know how to solve this in an automated yet as this can be in any .cnf file now. May need to write an entire .cnf parser. Or maybe recursively find the .cnf files and regex each one?
2. the error message is missing a '/' in it.

Workaround:

1. Install jemalloc-devel / libjemalloc-dev package
2. Add log_bin to /etc/my.cnf.d/columnstore.cnf

At that point this function should work as long as server_id isn't set in any other .cnf file.

Made the run script figure out which to use of libjemalloc.so, libjemalloc.so.1, and libjemalloc.so.2. Didn't address the log_bin or server_id issue. Todd reports that it fixes the Big Problem, and that he has a workaround for setting those 2 vars.

Made the PR, is in Andrew's queue.

As a side note, the real problem here is that one process or another is searching for strings in a log file to decide whether a cmd or script worked or not. It sees the error msgs from ld.so and thinks it failed when it didn't. By itself, ld.so not finding a lib in LD_PRELOAD isn't fatal.

Added fixes for everything else.

LGTM; merged it.