Is shown when jemalloc 4.0.4 is used.
This is preventing us from testing TokuDB with Valgrind.
jemalloc 4.0.4 attached, please use:
export LD_PRELOAD=/somedir/libjemalloc.so.2
Aha. Very cool.
Maybe the warning could include something like
"TOKUDB: no jemalloc found, using default allocator. Note: TokuDB performance will be lower then when jemalloc is used."
Makes it bit more user friendly?
I vaguely recall that the horrible pattern causes swapping and eventual OOM with other allocators?
Well, that really is also THP related. IIRC it was the combination of THP and the default allocator along with the bad FT pattern that would eventually lead to severe memory fragmentation and yes, swapping and OOM in the worst case. In the more general case, jemalloc was more of a performance play with a side effect that it tended to avoid fragmentation better than malloc.
OK, what we wan to do here is to take the tokudb_check_jemalloc option and pass it through to the new flag/option described in to disable the THP check if tokudb_check_jemalloc == FALSE.
We also want to invert the default value for tokudb_check_jemalloc so that the check is NOT done by default.