accumulo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sean Busbey <bus...@cloudera.com>
Subject Re: Start scripts and address/hostname
Date Fri, 26 Jun 2015 18:32:10 GMT
the rewritten scripts in Hadoop are much improved and HBase is also looking
to pull from them.

FYI, they're also slated to be part of the move of some common tooling to a
new TLP. So it's likely that in a month or two (depending on release
schedules) we could just transition to using them directly.

On Fri, Jun 26, 2015 at 1:18 PM, Josh Elser <josh.elser@gmail.com> wrote:

> Yeah, you're right, I believe AW did a big rehash on then. I'll take a
> gander.
>
> Thanks for the feedback, Eric. The situation that you outline I hope to
> make easier (w/ 100% less SSH workarounds than start-here.sh).
>
>
> John Vines wrote:
>
>> This may be tangential, but I heard the scripts for hadoop 3 had a massive
>> rewrite. Perhaps they can be consulted for desired behavior?
>>
>> On Fri, Jun 26, 2015 at 2:03 PM Eric Newton<eric.newton@gmail.com>
>> wrote:
>>
>>  Our ops people use the "start-here.sh" scripts to bring services back up
>>> after failures.  That's a great convenience: they don't have to remember
>>> which hosts are supposed to run the which service.
>>>
>>> In sympathy with your hostname troubles: the inconsistent use of hostname
>>> determination causes those tservers started with start-all.sh and
>>> start-here.sh to have different hostnames (shortname and fqdn,
>>> respectively). This has something to do with how our DNS is set-up (or
>>> hardcoded) because I cannot reproduce the effect in my development
>>> environment.
>>>
>>> As a consequence of this, the quoting hell of ssh, the limitations of
>>> writing code in Bash, I'm avoiding The Scripts as much as possible.  I am
>>> happy you are taking this on.
>>>
>>> -Eric
>>>
>>> On Thu, Jun 25, 2015 at 1:24 PM, Josh Elser<josh.elser@gmail.com>
>>> wrote:
>>>
>>>  I've been on a tear within our scripts in the last day. I've been moving
>>>> towards getting an accumulo-daemon.sh with some reasonable start, stop,
>>>>
>>> etc
>>>
>>>> semantics (ala Hadoop). This can also be done without affecting the
>>>> existing start-server.sh, start-here.sh, etc scripts.
>>>>
>>>> This hypothetical accumulo-daemon.sh script is a close feel to what an
>>>> init.d script would do. It alters the state of a server process on the
>>>> local node. One thing I'm struggling to wrangle is the current ability
>>>>
>>> the
>>>
>>>> scripts/configs provide to control the interface that the server
>>>>
>>> processes
>>>
>>>> bind to.
>>>>
>>>> For example, 127.0.0.1 in the `slaves` file will result in a
>>>> TabletServer
>>>> that processes external to the local node cannot talk to. I know there
>>>>
>>> are
>>>
>>>> likely fringe cases (multiple NICs, bonded interface) which I don't
>>>> fully
>>>> understand to ensure proper support.
>>>>
>>>> Is anyone an expert here and could give some advice about the kinds of
>>>> configuration that the scripts should provide to lets users run Accumulo
>>>> how they want to? I would like to move away from having to pass the
>>>> hostname/IP to scripts locally (e.g. `accumulo-daemon.sh start tserver`
>>>> would start a tserver locally), but I don't want break an existing
>>>> deployment.
>>>>
>>>> - Josh
>>>>
>>>>
>>


-- 
Sean

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