[MDEV-18714] Build is not parallel for storage/rocksdb Created: 2019-02-24  Updated: 2019-03-26  Resolved: 2019-03-26

Status: Closed
Project: MariaDB Server
Component/s: Compiling
Affects Version/s: 10.3.13
Fix Version/s: N/A

Type: Bug Priority: Major
Reporter: Michal Nowak Assignee: Unassigned
Resolution: Cannot Reproduce Votes: 0
Labels: None
Environment:

OpenIndiana 2018.10 (illumos kernel). 32-bit build.


 Description   

Starting with MariaDB 10.3.13, build process in at least some parts is not utilitizing all the cores given.

I build with '-j4' but at least when compiling storage/rocksdb and storage/innobase the CPU utilization is down to 1 core. E.g. at this line:

[ 7%] Building CXX object storage/rocksdb/CMakeFiles/rocksdblib.dir/rocksdb/cache/clock_cache.cc.o

CPU utilization seems fine for other parts.

MariaDB 10.3.12 is fine in this regard.

 Comments   
Comment by Marko Mäkelä [ 2019-02-24 ]

mnowak, can you please give some more information, including the exact steps that you are executing? Which compiler are you using? If you are using make, I would suggest posting a snippet of make VERBOSE=1 for that particular file.
Could it be that compiling that single file is taking a longer time? On Debian GNU/Linux, if I build with GCC 7 or 8, compiling the code generated rom sql_yacc.yy or sql_yacc_ora.yy is taking a very long time with -O2. But I have not noticed any such slowdown when compiling InnoDB. With clang (I have been using version 7 or 8 lately) the build time is faster.

Comment by Michal Nowak [ 2019-02-26 ]

It went away after I cleaned-up my build environment. Now it compiles as exped. Thank you!

Generated at Thu Feb 08 08:46:11 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.