hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Andrew Purtell (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-18848) meta region loading reporter
Date Wed, 04 Oct 2017 01:25:00 GMT

    [ https://issues.apache.org/jira/browse/HBASE-18848?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16190642#comment-16190642

Andrew Purtell commented on HBASE-18848:

MR or Spark jobs using TableInputFormat will pound on meta when starting up too

> meta region loading reporter
> ----------------------------
>                 Key: HBASE-18848
>                 URL: https://issues.apache.org/jira/browse/HBASE-18848
>             Project: HBase
>          Issue Type: Bug
>          Components: Operability, tooling
>            Reporter: stack
>              Labels: beginner
> We need reporting on meta table accesses.
> In quiet a few cases, it is said that our client -- I can believe it -- does too many
lookups against meta.
> In another case, we found that the hbase canary was ddos'ing the meta table but it took
a bunch of messing digging in rpc TRACE to figure out this was the case.
> A few users at scale are fingering meta ops as the bound to their being able to scale
more (especially if their readers are hosted in MR and startup is a big bang burst of meta
> The report would dump numbers on clients, incidence and type of queries. It would highlight
bad actors and pathological behaviors.
> This is a particular instance of what a general trace system might provide but we don't
have that yet. This could be a good first exercise implementing a trace or it might just use
trace engine mechanics to realize a few dump jsps. Whatever, we need some tooling for operators.

This message was sent by Atlassian JIRA

View raw message