accumulo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Christopher <ctubb...@apache.org>
Subject Re: ACCUMULO-3177 and ACCUMULO-3178
Date Wed, 26 Nov 2014 21:12:09 GMT
On Wed, Nov 26, 2014 at 3:07 PM, Sean Busbey <busbey@cloudera.com> wrote:

> I will be as timely as I can.
>
> Thanks. However, please understand that if you are unable to respond
timely, progress can/should/will proceed without you. I just want to make
sure you have opportunity.


> Please keep in mind that our community is volunteer based, people have a
> multitude of commitments, and it is a holiday week.
>
>
Sure, I just want to make sure you have ample time to review and we can
address any objections you may have (if you still have any), but I want
some time constraint, so these contributions don't sit idle indefinitely
(that's not fair to the contributor, or to those who'd like to leverage
these features). As long as the discussion is active (holidays/weekends
considered, of course), I will not attempt to push.


> I had presumed more feedback wasn't pressing since no one had mentioned on
> the tickets how my original concerns were addressed by the revamps. If
> someone could give me a summary on each it would make things go much
> faster.
>
>
Your feedback is pressing because you're the only one who expressed
concerns, and I want to make sure you have ample opportunity to be
involved. But, I don't want it to wait forever. I did believe I had
responded sufficiently to your questions/concerns... but if there's still
something outstanding that I have not answered or addressed, please raise
it again, explicitly, and in the context of the proposed changes, so I can
respond to it.

Sure, I can give you a summary as I understand it:

Overview of the issues:

ACCUMULO-3177: Adds a per-table VolumeChooser to replace the
RandomVolumeChooser as the default VolumeChooser, with the default
per-table implementation (a new per-table property is added for this) is
now the RandomVolumeChooser. This is modeled after the per-table balancer.
The intent is to enable balancing decisions for a tablet's persistent
storage across volumes in the same way we allow balancing tablet's
"compute" operations across Tablet Servers. To do this, a
"VolumeChooserEnvironment" object is created and added to the chooser API,
which exposes the environment to the chooser (such as the tableId, from
which the namespace can be retrieved, etc.)

ACCUMULO-3178: Adds an additional example/reference implementation for a
per-table VolumeChooser, called a "PreferredVolumeChooser", which allows
users to specify a specific volume (or set of volumes) in a custom
per-table property. This is used as a test case (integration test) for
ACCUMULO-3177. (Note: your original concerns about deduplication of HDFS's
HSM effort expressed in ACCUMULO-3089 seem to apply only to this issue, not
to ACCUMULO-3177; however, my hope is that you'll see that this is
sufficiently unrelated to that, that your original objections do not apply.)

Also, some background (from my perspective):

Recall: As explained on ACCUMULO-3089, your original concerns were
addressed by splitting the original issue into separate tasks, to help
clarify the scope and intentions of that issue with more narrowly defined
changesets. You were requested to explain your objections specifically in
the context of the changesets on those tasks, because I believed you to
have misunderstood the scope of the issue (as I stated on ACCUMULO-3089),
and I hoped that doing so would help clarify where your objections were, in
relation to the proposed code changes. As far as I can tell, you did do
that, with your comments on the subsequent issues.

You commented on ACCUMULO-3177, suggesting that it might be better to apply
the configuration property on a namespace instead of a table. I responded
to that comment with an explanation of how table and namespace
configurations work in this context. As such, I hold that your comment
represents a recommended "best practice" for table management
(specifically, limiting the granting of ALTER_TABLE), but not an objection
to the table/namespace configuration-based implementation/addition itself.

You have also commented on ACCUMULO-3178, with an indication that you did
not feel that the VolumeChooser interface itself was strictly "public API",
which I took to mean that you'd be more flexible in tolerating changes to
that API with a higher degree of risk. I did not see an objection to
ACCUMULO-3178, but it does depend on ACCUMULO-3177. You also suggested not
making these dependent on ACCUMULO-3176 (because of the presence of a
limited workaround by adding configuration to a namespace first), a request
which was accommodated (but ultimately unnecessary if ACCUMULO-3176 is
permitted) in a separate review which you did not comment on after
requesting it to be made.





> --
> Sean
> On Nov 26, 2014 12:27 PM, "Christopher" <ctubbsii@apache.org> wrote:
>
> > Thank you for your response. I request that you please be timely, as
> these
> > issues have sat idle for a long time, and the maintenance burden of
> keeping
> > them current with the HEAD of master is becoming excessive. I await your
> > reviews/objections. Thanks.
> >
> >
> > --
> > Christopher L Tubbs II
> > http://gravatar.com/ctubbsii
> >
> > On Wed, Nov 26, 2014 at 11:30 AM, Sean Busbey <busbey@cloudera.com>
> wrote:
> >
> > > I will follow up on both of those tickets with my objections, please do
> > not
> > > apply them.
> > >
> > > On Tue, Nov 25, 2014 at 2:08 PM, Christopher <ctubbsii@apache.org>
> > wrote:
> > >
> > > > My intention is to, sometime this week, take a thorough look at the
> > > patches
> > > > and reviews available for ACCUMULO-3177 (create a per-table
> > > VolumeChooser)
> > > > and ACCUMULO-3178 (example implementation of an alternate
> > > > VolumeChooser/test case for ACCUMULO-3177). Given the discussions
> > > > surrounding ACCUMULO-3176, and because they originated as the overall
> > > > ACCUMULO-3089 (which had some objections that may not have been
> > > resolved) I
> > > > wanted to give this notice, to ensure that there is opportunity for
> > > > objections to occur and be discussed here first.
> > > >
> > > > Presumably, the objections for ACCUMULO-3176 are different from any
> > that
> > > > might be held for 3177 and 3178, but there was limited follow-up
> > > > clarification of the objections raised after the issues were
> re-scoped
> > as
> > > > separate, more specific, JIRA sub-tasks. So, I'm not sure there are
> any
> > > > still outstanding for those. Since the reviews are still open for
> > those,
> > > I
> > > > just wanted to invite discussion either here, or (perhaps even
> better)
> > in
> > > > the specific review in ReviewBoard.
> > > >
> > > > --
> > > > Christopher L Tubbs II
> > > > http://gravatar.com/ctubbsii
> > > >
> > >
> > >
> > >
> > > --
> > > Sean
> > >
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message