[
https://issues.apache.org/jira/browse/IMPALA-8792?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16892978#comment-16892978
]
Tim Armstrong commented on IMPALA-8792:
---------------------------------------
I was thinking that this would be a more generic extension mechanism. I can
imagine us having a plugin interface that calls into a user-provided JAR with
some information about the query. We'd need to think carefully about the
information provided so that it was future-proof, e.g. providing the full AST
is risk because our internal data structures might change.
I've personally been mostly opposed to adding predefined rejection rules based
on compile-time information, e.g. "reject based on estimated bytes scanned"
because of the false positive/negative problem - queries get rejected because
of a bad estimate or threshold and there's no workaround except for completely
disabling the rule. IMPALA-6032 has a couple of proposals like that, and we've
looked at them enough to know that they're hard to do in a robust way.
If we have a generic extension mechanism then users are at least empowered to
fix that themselves. Or at least users will be aware of the potential to shoot
themselves in a foot if they are doing something that feels advanced like
writing a plugin instead of configuring a builtin option.
> User Defined Admission Rules
> ----------------------------
>
> Key: IMPALA-8792
> URL: https://issues.apache.org/jira/browse/IMPALA-8792
> Project: IMPALA
> Issue Type: New Feature
> Reporter: Quanlong Huang
> Priority: Major
>
> Impala cluster may be shared by many products. Admin may want to ban some
> queries as a work around for some bugs or known limitations, to avoid
> problematic queries impact the whole cluster.
> The rules can base on user names, query text, AST after parsing, or the plan
> after analysis. It'd be better if admin can add/delete/modify rules without
> restarting the coordinators.
> Example rules:
> * Ban all queries submitted from user idiot_product_name
> * Ban all queries containing text "application-name=xxx" (usually added as
> hint, used to recognize a product)
> * Ban all queries with text length larger than a threshold
> * Ban all queries that joins more than 20 tables (require AST)
> * Ban all queries that joins two specify fact tables (require AST)
> * Ban all queries using specify wrong join path (require AST)
> * Ban all queries that contain too complex expressions, e.g. #nodes in the
> expression tree exceeding a threshold (require AST)
>
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscr...@impala.apache.org
For additional commands, e-mail: issues-all-h...@impala.apache.org
