Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Plain View
Accumulo, mail # user - RE: Move high ingest tablets to alternate node? [SEC=UNOFFICIAL]


+
dlmarion 2013-08-21, 02:12
Copy link to this message
-
RE: Move high ingest tablets to alternate node? [SEC=UNOFFICIAL]
Dickson, Matt MR 2013-08-21, 03:38
UNOFFICIAL

The bulk of the data is organised by date however we index several fields also.  These have an alphanumeric rowid that we allow Accumulo to arrange the splits for rather than pre-splitting.  The 3 tablets in question are split based on these rowid's, ie not the date rowids.  I'm not concerned if I lose the data in these extents so am keen to know if I can a) delete the problem tablets or b) can move the 3 tablets to alternate nodes.

Matt

________________________________
From: dlmarion [mailto:[EMAIL PROTECTED]]
Sent: Wednesday, 21 August 2013 12:13
To: [EMAIL PROTECTED]
Subject: RE: Move high ingest tablets to alternate node? [SEC=UNOFFICIAL]

You can create a custom load balancer for your table. I think you said that your tables are organized by date. If the hot spot are contiguous days, then you could write a load balancer that does a round robin assignment.
Sent via the Samsung GALAXY S(r)4, an AT&T 4G LTE smartphone

-------- Original message --------
From: "Dickson, Matt MR" <[EMAIL PROTECTED]>
Date: 08/20/2013 9:50 PM (GMT-05:00)
To: "'[EMAIL PROTECTED]'" <[EMAIL PROTECTED]>
Subject: Move high ingest tablets to alternate node? [SEC=UNOFFICIAL]

UNOFFICIAL

Hi,

While loading data we are seeing one node with 3 tablets that have higher ingest rates than all others on the cluster.  All 3 of the these tablets reside on the same node which is causing that node to struggle.  When we take this node out of the cluster the overall ingest rate jumps much higher.

Is there a way to manually move high ingest tablets to different nodes in order to seperate the 3 tablets in question?

I started looking at the tablet splits and tried to find the mapping of these splits to nodes in the !METADATA table but couldn't find the mappings.

Thanks in advance,
Matt Dickson
+
Dickson, Matt MR 2013-08-21, 01:50