[hbase] Separate HBase v0.9x and v1.x bindings

[cmccoy]

See #272.
This change breaks HBaseClient10 into a separate module, and downgrades the HBase version used by the hbase binding to v0.98.13.

Note that the HBase 1.0 binding is now accessed via hbase1, instead of hbase-10 so that the runner script maps to the correct binding.

I tested against HBase v0.98.13 (with hbase) and v1.0.1.1 (with hbase1). The naming is a bit confusing...does anyone have an opinion on this naming shceme versus an hbase09 and hbase1 binding?

[busbey]

I'd like us to support HBase 0.94, and that won't work with the HBase 0.98 jars.

Do we want to have per-major-client-version bindings?

that would mean:

• HBase 0.94
• HBase 0.98
• HBase 1.y

That would mean we couldn't, e.g. take advantage of client-side optimizations in minor versions (atleast not without gymnastics). I certainly like this better than the alternative of per-minor-client-version bindings.

Unrelated, does keeping the 1.y API stuff in a module allow that module to build against the HBase 2.0.0-SNAPSHOT? Just wondering if we can close out #379 via this.

[busbey]

wrt naming, what if we introduce a means for bindings to have submodules and a way to tell us which class corresponds to a version request?

Essentially, something that would let us do:

$ ycsb load hbase --binding-version 1.0

or

$ ycsb load hbase --binding-version 0.94

instead of having it tied to the binding name directly.

[busbey]

the travis failure is because the distribution assembly pom needs to list the hbase1 binding as a dependency

[cmccoy]

Fixed the travis failure, thanks. I'll try to see if this addresses #379 tomorrow.

Re: the --binding-version flag...how would we handle new (and backwards compatible) versions? Would there be some mapping from DB version to binding version?

[cmccoy]

OK, I separated the HBase bindings into three versions:

• HBase 1.x (hbase10)
• HBase 0.98.x (hbase098)
• HBase 0.94.x (hbase094)

The hbase094 binding has no actual code, it just pulls in hbase098 and replaces dependencies. I don't have a lot of experience with Maven; LMK if there's a tidier way of doing this.

I'd like to get this in and put off adding a --binding-version flag.

[busbey]

I'm fine waiting on the binding-version flag.

@gkamat do you have time to review this?

@cmccoy are you wanting to get this into the release this weekend?

[busbey]

as the patch currently stands, we'd have to make the next release version 0.4.0, since the "hbase" binding target goes away.

[cmccoy]

I'm happy targeting this for the next release (leaving it out of 0.3.1).

[busbey]

I think with this patch there is no "hbase" binding, right?

[busbey]

this doc error is the only fix-before-push thing I see.

[cmccoy]

Sorry for the delay. All comments addressed - thanks!

[busbey]

+1