hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Lukáš Drbal <lukas.dr...@gmail.com>
Subject Re: Rowkey design and presplit table
Date Thu, 07 Mar 2013 22:32:27 GMT
Hello guys,

sorry for my longest response, iam working on cluster update from 0.94.1 to
0.94.5.

Ted:
yes, i'll post my solution after import data into production cluster

Asaf:
"Why do you need to use prefix split policy?"
Maybe i don't need it. I want distribute "unknown" keys to all nodes, avoid
hotspoting and split hell in import time. I know only one about this ids -
grow in time (very fast). We have this data now in key-value storage
implemented in mysql. For all this ids we use bigint.

Rowkey with Bytes.toBytes(long) are very interested. But i still don't know
how presplit table for this.

Maybe i can use something like <salt>_<articleId>_<categoryId>_<commentId>
where:
<salt> = last x (for example 2) bytes from articleId
<articleId>, <categoryId> and <commentId> = long values.

I have now one more usecase - read comments sorted desc by created time.
Its next point for this design - latest comment, article and category has
bigger id, so i can use reverse long for commentId.

Thats sound fine, i'll take a look on this rowkey and make some tests in
dev cluster. Thank you!

James:
I must have access on this data from node.js and php (via thrift) and i
don't want install something new on production cluster. But thanks, iam
looking on Phoenix after first announce here. Looks very cool!


Regards
--
Lukas Drbal

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