ps_tokudb_admin isn't a feature

[MARK CALLAGHAN]

From reading https://www.percona.com/doc/percona-server/5.6/tokudb/tokudb_installation.html
1) requires jemalloc 
2) requires me to disable THP 
3) is delivered as a plugin

I waste a few minutes on this each time I install TokuDB to run a perf test on it. I want those minutes back. 

#1 means I get silent errors when I try to load the plugin and see this in the error log. Toku requires jemalloc but the binary isn't linked with it.

[ERROR] TokuDB is not initialized because jemalloc is not loaded

#2 means I must have root. Not everyone has root.

#3 means I have to edit my.cnf, install with InnoDB, load plugins, shutdown, change my.cnf to enable TokuDB, startup

--

Mark Callaghan
mdca...@gmail.com

[Peter Zaitsev]

Hi Mark,

I read from this what you would like TokuDB installation to be easy but I do not understand what do you suggest (especially subject line)

Right now requirements are jemalloc and THP to be disabled. 

As our goal is high compatibility with stock MySQL which does not use jemalloc by default  we're also not enforcing it. 

Due to design issues, which change behavior even when more than one transactional storage engine is installed (and not used) we decided not to compile TokuDB in at this point

here are more details

https://www.percona.com/blog/2016/03/22/tokudb-impacts-innodb-performance/

--
You received this message because you are subscribed to the Google Groups "Percona Discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to percona-discuss...@googlegroups.com.
To post to this group, send email to percona-d...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Peter Zaitsev, CEO, Percona
Tel: +1 888 401 3401 ext 7360   Skype:  peter_zaitsev

[MARK CALLAGHAN]

I forgot about the tc_mmap bug. I hope we can fix that.
https://github.com/facebook/mysql-5.6/issues/227

The other issues just make me sad. RocksDB has a similar problem, but it is nowhere near as bad as Toku. For RocksDB without tcmalloc or jemalloc there will be allocator fragmentation leading to much larger process rss. But it works fine with any good allocator. It even works with glibc if you don't mind wasting more RAM. 

Fortunately, RocksDB doesn't require me to disable THP so I don't have to be root.

These are long known bugs in Toku, many years long. Are there plans to fix them?

Mark Callaghan
mdca...@gmail.com

[Peter Zaitsev]

Hi Mark,

Thanks for filing this as internal FB bug too!

In terms of TokuDB behavior I believe by default software should either work well or fail (preferably with clear error message about what has to be fixed).  Right now jemalloc and THP are requirements for TokuDB and  behavior when these are not enabled is not something which was validated to the point we want to support it. 

"Do not mind wasting more RAM" can be dangerous proposition for users - observing things work with small test some can put software in the production where it can show very poor performance causing system swapping or crashing out of memory - not something most will appreciate. 

Now I with MySQL Plugin Interface would allow more simple installation of bunch of plugins together as some sort of "plugin package" and allow to provide output directly to the user about requirements and failures, unfortunately it does not allow it at this point, this is why we've implemented ps_tokudb_admin to check requirements for most common use case. 

I recognize you have pretty special case and requirement and I will check if we can implement "Mark Callaghan" option to bypass safety checks :)

[Peter Zaitsev]

Hi Mark,

Have you tried --tokudb-check-jemalloc=0 option ?

It should prevent jemalloc requirement  check,  however I would advice against judging TokuDB performance with this option enabled as it is not recommended for production use :)

On Sat, Apr 2, 2016 at 1:01 PM, MARK CALLAGHAN <mdca...@gmail.com> wrote: