[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [jgit-dev] Dropping storage.dht package
- From: Shawn Pearce <spearce@xxxxxxxxxxx>
- Date: Tue, 14 Aug 2012 16:20:31 -0700
- Delivered-to: email@example.com
On Tue, Aug 14, 2012 at 3:36 PM, Chris Aniszczyk <caniszczyk@xxxxxxxxx> wrote:
> On Tue, Aug 14, 2012 at 1:55 PM, Matthias Sohn
> <matthias.sohn@xxxxxxxxxxxxxx> wrote:
>> 2012/8/14 Shawn Pearce <spearce@xxxxxxxxxxx>
>>> I am considering deleting the DHT storage backend for JGit. Its
>>> horribly complex and doesn't perform well compared to the local
>>> filesystem and DFS backends. The DFS backend is in heavy use at
>>> Google, and we are hoping we can find the time to port it to open
>>> source systems like HDFS. If we do want to keep a pure a database sort
>>> of backend, it shouldn't be what the DHT one is given how complex it
>>> is, and how slow it is. We should instead have a simpler schema, and
>>> accept whatever performance tradeoff that makes.
>>> Unfortunately the DHT storage backend might be viewed as part of the
>>> 2.0 API. So deleting it means a breaking API change. But if nobody is
>>> actually using the DHT backend, nobody would notice such a deletion.
>> +1, if it's complex, slow and hard to maintain we should remove it
>> the implementations of this API I am aware of are:
>> - your cassandra implementation
>> - your hbase impementation
>> - Philpp's JDBC implementation prototype
> +1, there's also nothing saying that we can't just bump the major version
> since this would be considered "breaking" if we considered DHT a public API.
OK. I will prepare a change within the next day or two to delete the
DHT code. We may have to rev to 3.0 as a result of that change going
in, but I really don't think anyone would really notice.