LIBEVENT_TARGET:
$(MAKE) -C libevent/make clean

LazyAndSet does not seem to work with BasicSet. Could be an issue in either, will need to write some unit tests to make sure. Mostly likely bug is in BasicSet ...

https://gist.github.com/665437

./src/server.cpp refers to some header files in a subdirectory "event2" which are not included.

A couple of the makefiles use g++ as the compiler. I think it is better to let the user define the compiler (e.g., by using $(CXX)).

Do we really need this?

https://github.com/javasoze/kamikaze/tree/master/src/test/java/com/kamikaze/test

Make compile time faster - you can leave the header only files as a separate copy and we can change the source to inline it for performance in a production build (if it actually makes a difference).

Need to save on RAM!

might also need to change leveldb allocator

The following non-source files are included in the repository:
leveldb/build_config.mk
leveldb/libleveldb.dylib
leveldb/libleveldb.dylib.1
leveldb/libleveldb.dylib.1.16

They should be removed + ignored.

"total+area+of+9.7%A0square+miles" does not convert back correctly

see http://urldecode.it/url/characters/

Also how should we handle empty fields?

Or simply use some library to escape xml entities before returning it ...

This look very good
http://highlandsun.com/hyc/mdb/microbench/

Will probably need to implement swap as we can't return the abstract Set from the current CompressedSet::removeDoc() function ...

It seems to work but it breaks the tokenizer ...

We need to collect for each blocking operation typ

• Wait count
• total wait-time
• AVG. Wait time
• Wait / transaction

The getString functions are fragile. I am getting errors about
one of them trying to override another.

It feels like it would be better and simpler to do:

            template <class T>
            static string getString(T v)
            {
                    std::stringstream ss;
                    ss << v;
                    return ss.str();
            }

(Arguably, it is maybe not the fastest way to do these conversions.)

Basically in case you're trying to run two instances of the server. We could consider adding parameters for db files etc, but this might make the startup have too much overhead.

We can reserve id 0 to mark the start of iteration, a docID of 0 is invalid ... they will always start with 1

we should something portable like leveldb::port::Mutex
see InvertedIndexBatch.hpp for an example

Seems to blow up with rapid xml :(

recreate the xml files but using a proper library ... or just dump text with the title as the file name

Maybe push to second release?

Need to add support in the Engine, test, add support in the server and the API. We can perhaps leave this for a second release - let's focus on creating a release and the website for now :)

https://github.com/linkedin/kamikaze/tree/master/src/test/java/com/kamikaze/test/perf