The New NASA Calendar

Why Cassandra is Unfit for Production

Cassandra is a tragedy.

Database replication should be a queue, or otherwise kept strict track of. If a datum should be replicated from X to Y and Z, then if it hasn't gotten to Z yet, then it should eventually.  The database is allowed to fail to replicate at first, but it is not allowed to just give up.

Cassandra views 1000 replication events from X to Z as a flock of birds who set off toward their destination. Most should arrive. If some don't ... well, no problem, you still have at least 950, right?

Also, there is absolutely no backpressure to throttle a writing process that would really like to write as fast is it can but no faster. This makes replication failure much harder to avoid.

To deal with the consequences, Cassandra depends on read repair (which is ludicrous) and a manual process that amounts to read repair of everything all at once.

It's pathetic that the Cassandra guys won't fix these problems, or even acknowledge that they are problems.

The Toolsmith's Dilemma

When I am using tools is when I most want to fix them. But that is when I least can.  Right now, for instance, I am trying to write some fricking PERL APPLICATIONS, but what do I have on my mind? The ways the tools should be better.  I've even patched some of them:

• namespace::autoclean needs -except (patched on github)
• namespace::autoclean should support Mouse (patched on github)
• namespace::autoclean wipes out overload functions (patched on github)
• B::Hooks::EndOfScope calls the same callback more than once. sometimes (worked around on github, but I have no idea why that happens, and really we need real core feature for stuff like this, which are fairly obvious in implementation)
• Mouse introspection - method detection is lame (should work like autoclean)
• Moose inner/augment doesn't work if more than one function in a class uses it (lame)
• Method::Signatures::Simple has no invocant-less variant (obvious fix)
• Method::Signatures invocant-less variant causes assertion failures on Perl 5.12.1 when there is any syntax error elsewhere (wtf)
• NYTProf with Coro (crash and/or nonsense results)
• perl -d with Coro (crashola, so back to printf debugging)
• how cool a modern Perl debugger could be

Oh, and let's not even talk about the incredible panoply of core features and redesigns just asking to be written! First, for example, a real working is_num and is_str that I know how to do, just haven't been able to get to.

But I have actual programs to write, don't I?

Lesson learned: it's too late for Perl, but if I want to keep writing C++ productively, I conclude that I must never peek inside another C++ compiler for the rest of my life.  And probably not the Boost libraries, either.

Child's Play

When I was a mere sprout I learned to play 3D chess. It wasn't the Star Trek kind; it was three full boards vertically aligned. The details of the rules variant I learned don't matter... the point is, if you think the solution space of normal chess is big, just imagine that when you think you have the enemy king cornered, bloop it goes one level UP.

During this period, my stepfather uncharacteristically invited me to play chess with him. And when the game began, I was astounded at how simple it was. Only ONE plane! Child's play, literally.

That's what it feels like to switch gears from C++ to Perl.

Retrofitting Modern Perl, Part Dieux

As a first step to introducing modern Perl to the Topsy code base, I've created a Local::Std module that provides all of the basic features I think no Perl program should be without.  It's all the same idea of what ToolSet does, but with a bit more control.

Of course Moose support is necessary when writing OO code.  There is much of Moose that annoys me, but using it is definitely better than not using it.  Then again, for many cases, Mouse is adequate.  So we use Any::Moose to allow loading either.

Also standard for classes are my own MooseX::Locked (when using real Moose) plus MooseX::StrictConstructor (which ought to be the Moose default, really).

I also provide easy access to MooseX::ParameterizedRole.  When you need a PR, you really need it.

The interface to all this is very simple, an absolutely vital design point:

• use Local::Std - only language features (described in a future post)
• use Local::Class - Std + Any::Moose + namespace clean
• use Local::Role - Std + Any::Moose('Role') + namespace clean
• use Local::RoleParam - Std + Moose + MooseX::ParameterizedRole + namespace clean

Next post: The language features in Local::Std.  Lots of goodies.

PS: I have ported autoclean to Mouse.  I hope the patches are accepted upstream.  Meanwhile, when using Mouse, we fall back to namespace::clean.  This seems ... less than ideal.