[
https://issues.apache.org/jira/browse/FLINK-11964?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jingsong Lee updated FLINK-11964:
---------------------------------
Description:
In HybridHashTable, first select the corresponding partition according to
hashCode, and then select the bucket in the partition according to hashCode,
using the same hashCode can easily cause hash collision.
Consider doing some mix to hashCode when choosing bucket.
Like JDK HashMap, we can just XOR some shifted bits in the cheapest possible
way to reduce systematic lossage, as well as to incorporate impact of the
highest bits that would otherwise never be used in index calculations because
of table bounds. (bucket use power-of-two masking). Just like: (hash ^ (hash
>>> 16))
In some cases, if a lot of conflicts occurred, this will lead to job hang,
because hash join will degenerate to nested loop join.
was:
In HybridHashTable, first select the corresponding partition according to
hashCode, and then select the bucket in the partition according to hashCode,
using the same hashCode can easily cause hash collision.
Consider doing some mix to hashCode when choosing bucket.
Like JDK HashMap, we can just XOR some shifted bits in the cheapest possible
way to reduce systematic lossage, as well as to incorporate impact of the
highest bits that would otherwise never be used in index calculations because
of table bounds. (bucket use power-of-two masking). Just like: (hash ^ (hash
>>> 16))
In some cases, if a lot of conflicts occurred, this will lead to job hang,
because hash join will regression to nested loop join.
> Avoid hash collision of partition and bucket in HybridHashTable in Blink SQL
> ----------------------------------------------------------------------------
>
> Key: FLINK-11964
> URL: https://issues.apache.org/jira/browse/FLINK-11964
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Task
> Reporter: Jingsong Lee
> Priority: Major
> Fix For: 1.10.0
>
>
> In HybridHashTable, first select the corresponding partition according to
> hashCode, and then select the bucket in the partition according to hashCode,
> using the same hashCode can easily cause hash collision.
> Consider doing some mix to hashCode when choosing bucket.
> Like JDK HashMap, we can just XOR some shifted bits in the cheapest possible
> way to reduce systematic lossage, as well as to incorporate impact of the
> highest bits that would otherwise never be used in index calculations because
> of table bounds. (bucket use power-of-two masking). Just like: (hash ^
> (hash >>> 16))
> In some cases, if a lot of conflicts occurred, this will lead to job hang,
> because hash join will degenerate to nested loop join.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
