[
https://issues.apache.org/jira/browse/IMPALA-5896?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tim Armstrong resolved IMPALA-5896.
-----------------------------------
Resolution: Won't Fix
The local catalog changes should address a lot of these problems. There hasn't
been any movement on this anyway so gonna close it out.
> Add safeguards to Catalog
> -------------------------
>
> Key: IMPALA-5896
> URL: https://issues.apache.org/jira/browse/IMPALA-5896
> Project: IMPALA
> Issue Type: Improvement
> Components: Catalog
> Affects Versions: Impala 2.5.0, Impala 2.6.0, Impala 2.7.0, Impala 2.8.0,
> Impala 2.9.0, Impala 2.10.0
> Reporter: Juan Yu
> Priority: Major
> Labels: supportability, usability
>
> Although we are actively working on improve metadata cache usage and
> efficiency. the data may grow faster. Without any safeguard to protect
> catalog server, we might hit catalog limit sooner or later.
> Here are Some suggestions:
> 1. Stop loading more metadata after loading large amount of files/blocks for
> a single table to avoid exceed 2GB java array limit.
> Note: A proper limit is hard to define. the metadata size also depends on
> #partitions and if incremental stats metadata present. but 5~6M seems a
> conservative and large enough limit for most of use cases (consider the best
> practices for DN is no more than 1M blocks per node).
> 2. Reject new metadata ops (e.g. loading metadata for a new table) if
> metadata cache heap usage is close to mem_limit setting to avoid OOM. This
> usage should not include the objects that can be GCed.
> This change will bring wrong result or table not get updated
> We need to make sure show proper warnings to user that there is a problem and
> they need to fix those metadata issue asap.
> e.g table *** doesn't have updated metadata or have incomplete metadata due
> to memory limit.
> Not enough memory to load table ***, please remove some unused tables from
> cache by "invalidate metadata <table>"
> More diagnose information that could help debugging catalog issues:
> - Which table has caused the crash?
> - Which table or how many partitions from a specific table we need to remove
> to get it back up again?
> - Given a huge topic size, what is the contribution of each of the tables to
> it?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
