You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@drill.apache.org by "ASF GitHub Bot (JIRA)" <ji...@apache.org> on 2017/11/27 16:43:00 UTC

[jira] [Commented] (DRILL-5963) Canceling a query hung in planning state, leaves the query in ENQUEUED state for ever.

    [ https://issues.apache.org/jira/browse/DRILL-5963?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16267025#comment-16267025 ] 

ASF GitHub Bot commented on DRILL-5963:
---------------------------------------

GitHub user arina-ielchiieva opened a pull request:

    https://github.com/apache/drill/pull/1051

    DRILL-5963: Query state process improvements

    1. Added two new query states: PREPARING (when foreman is initialized) and PLANNING (includes logical and / or physical planning).
    2. Process of query enqueuing was made non-blocking. Once query was enqueued, fragments runner is called to submit fragments locally and remotely.
    3. Ability to cancel query during planning and enqueued states was added.
    4. Logic for submitting fragments was moved from Foreman to new class FragmentsRunner.
    5. Major type in DrillFuncHolderExpr was cached for better performance.
    
    Other details in [DRILL-5963](https://issues.apache.org/jira/browse/DRILL-5963)

You can merge this pull request into a Git repository by running:

    $ git pull https://github.com/arina-ielchiieva/drill DRILL-5963

Alternatively you can review and apply these changes as the patch at:

    https://github.com/apache/drill/pull/1051.patch

To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:

    This closes #1051
    
----
commit 1df913951ae9cadf3813836c4a03e66645d68965
Author: Arina Ielchiieva <ar...@gmail.com>
Date:   2017-11-24T15:59:42Z

    DRILL-5963: Query state process improvements
    
    1. Added two new query states: PREPARING (when foreman is initialized) and PLANNING (includes logical and / or physical planning).
    2. Process of query enqueuing was made non-blocking. Once query was enqueued, fragments runner is called to submit fragments locally and remotely.
    3. Ability to cancel query during planning and enqueued states was added.
    4. Logic for submitting fragments was moved from Foreman to new class FragmentsRunner.
    5. Major type in DrillFuncHolderExpr was cached for better performance.

----


> Canceling a query hung in planning state, leaves the query in ENQUEUED state for ever.
> --------------------------------------------------------------------------------------
>
>                 Key: DRILL-5963
>                 URL: https://issues.apache.org/jira/browse/DRILL-5963
>             Project: Apache Drill
>          Issue Type: Bug
>          Components: Execution - Flow
>    Affects Versions: 1.12.0
>         Environment: Drill 1.12.0-SNAPSHOT, commit: 4a718a0bd728ae02b502ac93620d132f0f6e1b6c
>            Reporter: Khurram Faraaz
>            Assignee: Arina Ielchiieva
>            Priority: Critical
>             Fix For: 1.13.0
>
>         Attachments: enqueued-2.png
>
>
> Canceling the below query that is hung in planning state, leaves the query in ENQUEUED state for ever.
> Here is the query that is hung in planning state
> {noformat}
> 0: jdbc:drill:schema=dfs.tmp> select 1 || ',' || 2 || ',' || 3 || ',' || 4 || ',' || 5 || ',' || 6 || ',' || 7 || ',' || 8 || ',' || 9 || ',' || 0 || ',' AS CSV_DATA from (values(1));
> +--+
> |  |
> +--+
> +--+
> No rows selected (304.291 seconds)
> {noformat}
> Explain plan for that query also just hangs.
> {noformat}
> explain plan for select 1 || ',' || 2 || ',' || 3 || ',' || 4 || ',' || 5 || ',' || 6 || ',' || 7 || ',' || 8 || ',' || 9 || ',' || 0 || ',' AS CSV_DATA from (values(1));
> ...
> {noformat}
> The above issues show the following problems:
> *1. Simple query with reasonable number of concat functions hangs.*
> In reality query does not hang it just take lots of time to execute. The root cause is that during planning time DrillFuncHolderExpr return type is extensively used to determine matching function, matching type etc. Though this type is retrieved via [getter|https://github.com/apache/drill/blob/master/exec/java-exec/src/main/java/org/apache/drill/exec/expr/DrillFuncHolderExpr.java#L41] in reality complex logic is executed beaneath it. For example for [concat function|https://github.com/apache/drill/blob/master/exec/java-exec/src/main/java/org/apache/drill/exec/expr/fn/output/ConcatReturnTypeInference.java#L47]. Since function return type can not be changes during DrillFuncHolderExpr life time, it is safe to cache it.
> *2. No mechanism to cancel query during ENQUEUED state.*
> Currently Drill does not have mechanism to cancel query before STARTING / RUNNING states. Plus ENQUEUED state includes two PLANNING and ENQUEUED.
> Also submitting mechanism for submitting query to the queue is blocking, making foreman wait till enqueueing is done Making it non-blocking will prevent consuming threads that just sit idle in a busy system and also is important when we move to a real admission control solution.
> The following changes were made to address above issues:
> a. two new states were added: PREPARING (when foreman is initialized) and PLANNING (includes logical and / or physical planning).
> b. process of query enqueuing was made non-blocking. Once query was enqueued, fragments runner is called to submit fragments locally and remotely.
> c. ability to cancel query during planning and enqueued states was added.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)