maria-discuss team mailing list archive

[Reindl Harald <h.reindl@xxxxxxxxxxxxx>]

Am 03.07.2017 um 11:45 schrieb Sergei Golubchik:
> Hi, Reindl!
>
> On Jul 02, Reindl Harald wrote:
> > i get tired - "-DPLUGIN_INNODB=YES" leads to failures at configure time
> > - each time the same dance when the version changes
>
> Sorry for this. I know, this shouldn't happen. On the other hand, you
> use a rather detailed non-standard build configuration, there's a good
> chance that you're the only man on the Earth who uses exactly this
> combination of flags. So, sometimes you his something that nobody else
> sees.

MariaDB needs a minimal build target where just MyISAM, CSV (as it is 
used internally) and Aria are built and anything above can be added with 
configure options, in that case the buildbots could test it automated

> I've fixed altered_table->s->partition_info_str failure  in InnoDB.
>
> > well, when remove that thinking since xtradb is now dead it will be
> > default enabled leads finally still to a build-failure with not
> > mentioning really where and why - can we please have back "configure &&
> > make && make install" with "./configure --help" back for peace of mind
>
> It wouldn't help much. There still can be unused used and untested
> buggy combinations of flags that someone could occasionally hit.

it helps by InnoDB builds now

> > _______________________________
> >
> > [ 53%] Linking CXX static library libmyisammrg.a
> >
> > *why* at all when i say "-DPLUGIN_MYISAMMRG=NO" but remove it does still
> > make no difference
>
> Because myisammrg plugin is declared MANDATORY in its CMakeLists.txt,
> mandatory plugins cannot be disabled.

since WHEN is that the case given that i have 10.1 procution machines 
where it is not builtin

> > _______________________________
> >
> > cmake . \
> >    -DFEATURE_SET="large" \
> ...
> >    -DCOMPILATION_COMMENT="thelounge"
> >
> > [ 52%] Built target myisam
> > make[1]: Leaving directory '/home/builduser/rpmbuild/BUILD/mariadb-10.2.6'
> > Makefile:162: recipe for target 'all' failed
> > make: *** [all] Error 2
> > Fehler: Fehler-Status beim Beenden von /var/tmp/rpm-tmp.Qwgvnq (%build)
>
> Interesting, so make isn't localized, but rpmbuild is? :)

likely a implementation detail of rpmbuild to not trigger random 
buildbugs because of a never by upstream used locale at build time

> Anyway, I've just built with exactly your combination of options,
> fixed the failure in InnoDB (partition_info_str) and in mysqld.cc
> (check_openssl_compatibility).
>
> Now the build succeeds for me

good to hear - thanks!

the main reason why jump to 10.2 is 
https://jira.mariadb.org/browse/MDEV-6150 which sounds to solve a 
massive scalibility problem given that our CMS code and PHP build is 
that fast that you can't completly use the hardware and my 6 years old 
4-core workstation is only slightly faster than a 12 core 3.50 GHz Xeon 
when the myslqd process hangs around 100% up to sometimes 130% on a 
single core and even on a single request 10% of the complete runtime is 
connection to the database server and when that hangs on a single thread 
this is the bottenleck guilty

persistent connections are impossible in setups where different 
instances share up to 800 httpd workers which would mean instance-count 
* 800 = number of possible concurrent database connections