Using MariaDB with TCMalloc or jemalloc
Read the Profiling Memory Usage page for more information on how to debug high memory consumption.
Using tcmalloc or jemalloc
TCMalloc is a malloc replacement library optimized for multi-threaded usage. It also features a built-in heap debugger and profiler.
Another malloc replacement that may speed up MariaDB is jemalloc.
The procedures to use one of these libraries with MariaDB are the same. Many other malloc replacement libraries (as well as heap debuggers and profilers) can be used with MariaDB in a similar fashion.
Checking the malloc Implementation in Use
If you are unsure which malloc implementation is in use, or if you used one of the procedures explained in this page and you want to verify if it succeeded, you can run this query:
SHOW GLOBAL VARIABLES LIKE 'version_malloc_library';
A value of "system" indicates the system default, which is normally malloc. If another library is used, this query will return the name and version of the library.
Building MariaDB with an alternative to malloc
To build MariaDB 5.5 with TCMalloc
, you need to use the following command
cmake -DCMAKE_EXE_LINKER_FLAGS='-ltcmalloc' -DWITH_SAFEMALLOC=OFF
To use jemalloc, the option should be -ljemalloc
.
Starting mariadbd-safe with an alternative to malloc
If you want to do this only one time, as a test, you can also start a standard MariaDB server with TCmalloc
with:
/usr/sbin/mariadbd-safe --malloc-lib=tcmalloc
If you want to configure mariadbd-safe to use tcmalloc or jemalloc, edit your configuration file, in the [server]
or [mariadbd]
group:
malloc-lib=tcmalloc
Starting mariadbd with an alternative to malloc
First, locate the library file that needs to be used:
# jemalloc find /usr/lib -name "libjemalloc.so.*" # tcmalloc find /usr/lib -name "libtcmalloc.so.*"
Now pass it to mariadbd
using the LD_PRELOAD
variable:
LD_PRELOAD=/path/to/library mariadbd
For example, on OpenSuse 15.4 one would do:
Configuring systemd
If you use systemd to run MariaDB, first locate the library as explained above. The locate the service configuration file:
systemctl status mariadb |grep Loaded
Now edit the mariadb.service
file by adding a line to the [Service]
group:
Environment=LD_PRELOAD=<path-to-library>
For example:
[Service] Environment=LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so.2
Now you should reload the configuration, so that the news setting will take effect, and restart MariaDB:
systemctl daemon-reload systemctl restart mariadb
Dockerfile
If you run MariaDB on Docker and use an image from a Dockerfile that is publicly available, most probably you have an entrypoint that is a Bash script, which starts mariadbd
directly. You can edit this Bash script as explained above. Or you can set the LD_PRELOAD
variable from the Dockerfile:
ENV LD_PRELOAD=<path-to-library>
To find the library file you can run one of these commands while the container is running:
# jemalloc docker exec -ti <container-name> find /usr/lib -name "libjemalloc.so.*" # tcmalloc docker exec -ti <container-name> find /usr/lib -name "libtcmalloc.so.*"
Vagrantfile
Usually Vagrant is used to start a complete system in a virtual machine. If this is your case, you can use one of the methods above, for example you can modify your Vagrantfile to copy a modified version of the mariadb.service
file into the guest system to configure systemd.
If you use Vagrant with the Docker provider, you can follow the instructions above to modify the Dockerfile.
Finding memory leaks with jemalloc
jemalloc can provide a report of memory leaks at program exit:
MALLOC_CONF=prof_leak:true,lg_prof_sample:0,prof_final:true \ LD_PRELOAD=${JEMALLOC_PATH}/lib/libjemalloc.so.2 path-to-mariadbd
This will produce something like:
<jemalloc>: Leak summary: 267184 bytes, 473 objects, 20 contexts <jemalloc>: Run jeprof on "jeprof.19678.0.f.heap" for leak detail
You can learn more about the memory leaks with jeprof, that is included with jemalloc:
jeprof --show_bytes path-to-mariadbd jeprof.19678.0.f.heap
You can also generate a PDF call graph of the leak:
jeprof --show_bytes --pdf path-to-mariadbd jeprof.19678.0.f.heap > /tmp/mariadbd.pdf