accumulo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Keith Turner <>
Subject Re: Assignment Overview Doc
Date Thu, 13 Aug 2015 18:37:11 GMT
On Thu, Aug 13, 2015 at 12:03 PM, Josh Elser <> wrote:

> Yes, I omitted the session ID intentionally due to confusion of "what the
> heck is that random string". I thought I had mentioned that there is a
> uniqueness string on there to identify instances of servers on the same
> host and port, but maybe I dropped that out in the end.
> Ultimately, it's not really relevant for the assignment discussion,
> assignment just cares about a unique instance of a server. I'll see if I
> can maybe clarify that.
Its important in that its used to know when a future or current location is
a dead tserver and needs to be reassigned.

> Thanks for taking the time to read and send back some feedback!
> Keith Turner wrote:
>> i noticed the document uses<server>:<port>   to identify tablet servers
>> some places.   Something more specific is used in the code
>> :<server>:<port>:<zk session id>.   This last dimension allows
the master
>> to distinguish between two different instances of a tserver running at the
>> same host and port at different times.
>> On Wed, Aug 12, 2015 at 12:39 PM, Josh Elser<>
>> wrote:
>> I had some HBase friends ask me for a design document on how Tablet
>>> assignment in Accumulo works. Their assignment logic continues to be a
>>> pain
>>> point and there's been a lot of confusion about wanting to use FATE for
>>> assignments (even moreso when I would tell people: no, that's
>>> unnecessary).
>>> Since we didn't have a good reference document (I'm guessing because it
>>> was designed prior to the initial import of code), I took the time to
>>> write
>>> one up based on my understanding of the code and some pointers from Eric.
>>> Think this deserves a place in our documentation somewhere? Seems like
>>> it's a good reference sheet to have lying around, but I'm not sure where
>>> best it'd be stored.
>>> Huge thanks to Eric for his pointers and editing.
>>> - Josh

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