I think there's definitely (at least?) two categories of public API's in
play. What we have now is definitely "end-user" focused. All of the
things that we expect users to do fall into this.
However, we have this odd "super-user" category too. Iterators, custom
server-side implementations all require an additional degree of
understanding to do both safely and correctly.
My motivation is that I believe we should be striving for compatibility
on both of these categories (for both types of users). I'm sure those
who write Iterators are smart enough to understand how to update them
for some interface change; but, can't we do better than that in
Accumulo? I think this also touches on the recent maintenance tail
conversations from the JDK8 discussion: breakages here are just another
reason (excuse?) for people to avoid upgrading.
Perhaps such APIs are held to a lesser standard than our public API?
Maybe we start with some subset of semver and see how it goes? I think
these are all great points to consider. Presently, we have nothing; I
know we can do better than that.
Marc P. wrote:
I tend to think that what can be configured now is tailored to the
infrastructure engineers? With the exception of SKVI, is it short sighted
to suggest that the consumers of said APIs are generally few and far
between? Those flexible APIs exist for the John "Ain't no mountain high
enough" Vines of the world.
My point is that breaking an API into public and private makes maintenance
difficult because you have few people that understand certain portions of
the code. Furthermore, there is frustration in making changes at certain
levels. Could there be value in defining layers of API responsibility that
dictate for who or whom these APIs may be useful but also differentiate
temporal stability? i.e. scanner API is slightly more consumer focused and
is stabilized much more than tablet balancers, where changes may occur at a
greater frequency?
On Fri, Aug 26, 2016 at 1:24 PM, Sean Busbey<[email protected]> wrote:
On Fri, Aug 26, 2016 at 11:58 AM, Christopher<[email protected]> wrote:
+1 for creating supported interfaces in our public API for these.
Right now, I think all of these areas are suffering from bit
rot/technical
debt, and need to be cleaned up before (or in the process of) exposing
them
as public API.
Can we make this clean up a goal for a>=Accumulo 3.0 world? Or since
we'd be adding to the public API maybe>=Accumulo 2.y for some y>= 1?
We're long for a 2.0 release, and the sooner we get it out the door
(if only just for the java 8+ only change) the better chances we're
giving to downstream folks to move to it in an orderly manner. I'd
prefer we not delay that further for API additions. (IIRC, that's why
we didn't have a 2.0 release last year?)
On Fri, Aug 26, 2016 at 12:48 PM Josh Elser<[email protected]>
wrote:
In an>=Accumulo-2.0 world, I think it would be prudent to investigate
how we can address this problem to reduce maintenance burden on
ourselves and create supported "public API" interfaces for these. I
imagine that we could come up with a general approach that provides
"guidelines" for how we would handle cases like this in the future.
Another option would be that we expressly make these pluggable parts
"use at your own risk" internals with a weaker compatibility promise
than semver. We could, for example, label these points as something
that we won't break in a double-dot release (e.g. whatever works in
1.7.0 will work in 1.7.z) but still give ourselves the room to change
them in minor releases.
I primarily mention this because we don't have an established history
for maintaining code lines across major version releases, so I'm
skeptical of things that will force us to increment the version in the
master branch across major #s.
--
busbey
