[tests]: unit test TODO

[BohuTANG]

need to re-add

[jyuanmurray]

Hi Bohu,

Since nessdb shares buffered tree in design with tokudb, do you think it is a good idea to port a fraction of their tests to nessdb? For example, the tree flushing related tests and cachetable tests. Certainly we should start with util tests first.

Let me know if I can help. I would like to contribute to this issue. nessdb looks a fun project.

-Jun

[BohuTANG]

@jyuanmurray
Cool, nessDB is employing Buffered-Tree(Fractal Tree in TokuDB), and these parts have done:

1. buffered-tree index 2)MVCC in leaf 3)tree flush
   these parts are in processing:
2. recovery 2) checkpoint 3)more tests
   You are very welcome for any parts.
   Sorry for less time on it these days(busy working), but I believe I am coming back.

BohuTANG

[jyuanmurray]

@BohuTANG
Thanks. I will start looking at your packed mem array impl and think about unit tests. Or any test you can prioritize for now?

More important, I have been following your posts on douban and I always want to ask you:
" what is your ultimate goal for nessDB? how are you going to position nessDB in next 2 years?"
You do not want to simply exercise tokuDB again. Yes, a faster, scalable, resilient storage engine -- maybe Combining levelDB and tokuDB could be a direction that you can pursue, however, no matter in academia or for engineering, a proj without aiming at real world application is just a toy. I see great potential of nessDB beyond a lightweight version of tokuDB, however the concrete image is vague. Where will nessDB be best used in your mind? for example, for embeded system?

Jun

[BohuTANG]

@jyuanmurray
First, thanks for your attention, no one ever cared these like you!
IMHO, nessDB is employing buffered-tree as TokuDB with a simple way, this was my original intention to write it.
It may be a toy for a long time, I have many ideas want to try but no time(busying work).
Any suggestions?

BohuTANG

[jyuanmurray]

@BohuTANG , I do not have anything concrete in mind yet but I have a question about your packed memory array impl -- why did you choose pma? toku ppl have decided to just use FIFO queue...for accessing with a certain pattern the insetion seems to be O(log^2n).

[BohuTANG]

@jyuanmurray
the PMA is unrolled packed memory array, it's O(log^2n) and all about memmove no need to rebalance, it's very fast on small data size, you can check it.
If you use FIFO queue, you must construct a tree for searching over the FIFO.

BohuTANG

[jyuanmurray]

@BohuTANG Thanks. That makes sense. OK I will start with unit-testing the basic data structure. I have questions about your MVCC impl too. Maybe I will ping you on douban.

[BohuTANG]

@jyuanmurray
Please ping me on gtalk: [email protected]