Thanks, Umesh. Seems like you're saying it's not a problem now, but
you're not sure if it would become a problem. Regardless of that, it's a
goal to not be version-specific (and thus, we can have generic hbck-v1
and hbck-v2 tools). LMK if I misread, please :)
One more thought, it would be nice to name this repository as
"operator-tools" or similar (instead of hbck). A separate repo on its
own release cadence is a nice vehicle for random sorts of recovery,
slice-and-dice, one-off tools. I think HBCK is one example of
administrator/operator tooling we provide (certainly, the most used),
but we have the capacity to provide more than just that.
On 7/24/18 5:55 PM, Umesh Agashe wrote:
Thanks Stack, Josh and Andrew for your suggestions and concerns.
I share Stack's suggestions. This would be similar to hbase-thirdparty. The
new repo could be hbase-hbck/hbase-hbck2. As this tool will be used by
hbase users/ developers, hbase JIRA can be used for hbck issues.
bq. How often does HBCK need to re-use methods and constants from code
in hbase-common, hbase-server, etc?
bq. Is it a goal to firm up API stability around this shared code.
bq. If we do this can we also move out hbck version 1?
As HBCK2 tool will be freshly written, we can try to achieve this goal. I
think its great idea to move hbck1 to new repo as well. Though I think its
more involved with hbck1 as the existing code already uses what it can from
hbase-common and hbase-server etc. modules.
bq. How often does HBCK make decisions on how to implement a correction
based on some known functionality (e.g. a bug) in a specific version(s)
of HBase. Concretely, would HBCK need to make corrections to an HBase
installation that are specific to a subset of HBase 2.x.y versions that
may not be valid for other 2.x.y versions?
I see if this happens too often, compatibility metrics will be complicated.
Thanks,
Umesh
On Tue, Jul 24, 2018 at 10:27 AM Andrew Purtell <apurt...@apache.org> wrote:
If we do this can we also move out hbck version 1? It would be really weird
in my opinion to have v2 in a separate repo but v1 shipping with the 1.x
releases. That would be a source of understandable confusion.
I believe our compatibility guidelines allow us to upgrade interface
annotations from private to LP or Public and from LP to Public. These are
not changes that impact source or binary compatibility. They only change
the promises we make going forward about their stability. I believe we can
allow these in new minors, so we could potentially move hbck out in a
1.5.0.
On Mon, Jul 23, 2018 at 4:46 PM Stack <st...@duboce.net> wrote:
On Thu, Jul 19, 2018 at 2:09 PM Umesh Agashe
<uaga...@cloudera.com.invalid
wrote:
Hi,
I've had the opportunity to talk about HBCK2 with a few of you. One of
the
suggestions is to to have a separate git repository for HBCK2. Lets
discuss
about it.
In the past when bugs were found in hbck, there is no easy way to
release
patched version of just hbck (without patching HBase). If HBCK2 has a
separate git repo, HBCK2 versions will not be tightly related to HBase
versions. Fixing and releasing hbck2, may not require patching HBase.
Though tight coupling will be somewhat loosened, HBCK2 will still
depend
on
HBase APIs/ code. Caution will be required going forward regarding
compatibility.
What you all think?
I think this the way to go.
We'd make a new hbase-hbck2 repo as we did for hbase-thirdparty?
We'd use the hbase JIRA for hbase-hbck2 issues?
We'd make hbase-hbck2 releases on occasion that the PMC voted on?
Sounds great!
St.Ack
Thanks,
Umesh
JIRA: https://issues.apache.org/jira/browse/HBASE-19121.
Doc:
https://docs.google.com/document/d/1NxSFu4TKQ6lY-9J5qsCcJb9kZOnkfX66KMYsiVxBy0Y/edit?usp=sharing
--
Best regards,
Andrew
Words like orphans lost among the crosstalk, meaning torn from truth's
decrepit hands
- A23, Crosstalk
