[jira] [Commented] (HAWQ-296) TPC-H Query 5 encounters OOM in large HAWQ cluster with None mode

["Ruilong Huo (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HAWQ-296?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15074846#comment-15074846 ] 

Ruilong Huo commented on HAWQ-296:
----------------------------------

Root cause analysis shows that: within 12G memory quota in total, it uses about 10.8G (memory_total * runaway_detector_activation_percent = 12G * 90%) as red zone and reserves 1.2G so that runaway terminator can cancel the query with high memory usage. The reserved quota is even higher, especially on servers with large physical memory. For example, for server with 32G memory, it reserves about 3.2G; for server with 64G memory, it reserves 64G; and for server with 128G memory, it reserves 12.8G.

Given the memory allocation granularity is 1G, it is reasonable to reduce runaway_detector_activation_percent so that it won't terminate queries even there are several gigabytes of memory available.


> TPC-H Query 5 encounters OOM in large HAWQ cluster with None mode
> -----------------------------------------------------------------
>
>                 Key: HAWQ-296
>                 URL: https://issues.apache.org/jira/browse/HAWQ-296
>             Project: Apache HAWQ
>          Issue Type: Bug
>          Components: Resource Manager
>    Affects Versions: 2.0.0-beta-incubating
>            Reporter: Ruilong Huo
>            Assignee: Ruilong Huo
>
> TPC-H Query 5 encounters OOM in large HAWQ cluster with None mode.
> OOM in EXPLAIN ANALYZE of TPC-H Query 5:
> {noformat}
> explain analyze select
>  n_name,
>  sum(l_extendedprice * (1 - l_discount)) as revenue
> from
>  customer,
>  orders,
>  lineitem,
>  supplier,
>  nation,
>  region
> where
>  c_custkey = o_custkey
>  and l_orderkey = o_orderkey
>  and l_suppkey = s_suppkey
>  and c_nationkey = s_nationkey
>  and s_nationkey = n_nationkey
>  and n_regionkey = r_regionkey
>  and r_name = 'ASIA'
>  and o_orderdate >= date '1997-01-01'
>  and o_orderdate < date '1997-01-01' + interval '1 year'
> group by
>  n_name
> order by
>  revenue desc;
> 532767 [2015-12-25 23:52:39]
> NOTICE:  Canceling query because of high VMEM usage. Used: 11060MB, available 1228MB, red zone: 11056MB (runaway_cleaner.c:152)  (seg969 test45.ic:40000 pid=738560)
> {noformat}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)