On 6/4/18 12:16 AM, Stack wrote:
On Sun, Jun 3, 2018 at 8:54 PM, 张铎(Duo Zhang)<[email protected]> wrote:
What;s your plan sir? Branch branch-2.1 from branch-2.0?
Its a suggestion.
I like Andrew's notion that we left-shift how we have been thinking about
version numbers; that we releases tend toward minor increments rather than
patch increments as we have been doing up to this.
If we are going to act on Andrew's suggestion, now is the time to do it.
2.0.0 was rough. 2.1.0 could be branched from branch-2.0 being 2.0.0 but
with 100+ bug and perf fixes. I could even see folks deploying a 2.1.0 in
production. Perhaps there'll be a 2.2.0 and a 2.3.0. They'll be boring bug
and perf improvements only.
We already have enough to define a substantial 3.0.0 IMO what with serial
replication and HBASE-20312 CCSMap.
I'm trying to avoid 3.0.0 being like 2.0.0 where it takes years for it to
ship. Meantime we accumulate a mountain of testing, perf, and whatever else
tech debt.
What do folks think?
mmmm, that's a good point. We keep saying that we aren't going to fall
into the same trap over and over, yet here we are setting ourselves up
to do it again :)
I would have no complaints against 2.0.0+ becoming 2.1.0, but I feel
like that would make Duo's RM life harder too (assuming branch-2 is more
stable than master).
I honestly don't know how to balance that. Someone eventually has to
bite the bullet :\
