[ 
https://issues.apache.org/jira/browse/IMPALA-9118?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17779702#comment-17779702
 ] 

ASF subversion and git services commented on IMPALA-9118:
---------------------------------------------------------

Commit b6bdf4c525acfe8b35928d43155cf3bea9be26cb in impala's branch 
refs/heads/master from stiga-huang
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=b6bdf4c52 ]

IMPALA-9118: Show catalog operation details in catalogd webUI

This patch extends the /operations page in catalogd WebUI to show the
in-flight and finished catalog operations. The following fields are
shown for each operation:
 - Thread ID
 - Query ID
 - Client IP
 - Coordinator
 - Type
 - Target
 - User
 - Start Time
 - End Time (only shown for finished operations)
 - Duration
 - Status
 - Details

Finished operation records are currently kept in memory and limited by
the catalog_operation_log_size flag (defaults to 100).

To collect the above fields, this patch extends
TCatalogServiceRequestHeader to contain the coordinator hostname. Also
fix some catalog RPCs that didn't fill these fields.

Tests:
 - Add e2e test in custom_cluster/test_web_pages.py
 - Manually verify the web pages when running a GVO job

Change-Id: I3cf3f0da2be2be79e546762a8083d4de338ff6aa
Reviewed-on: http://gerrit.cloudera.org:8080/20428
Reviewed-by: Impala Public Jenkins <[email protected]>
Tested-by: Impala Public Jenkins <[email protected]>


> Add debug page for in-flight DDLs in catalogd
> ---------------------------------------------
>
>                 Key: IMPALA-9118
>                 URL: https://issues.apache.org/jira/browse/IMPALA-9118
>             Project: IMPALA
>          Issue Type: New Feature
>          Components: Catalog
>            Reporter: Quanlong Huang
>            Assignee: Quanlong Huang
>            Priority: Critical
>              Labels: observability, supportability
>         Attachments: Selection_082.png
>
>
> In a busy cluster, it's possible that many DDL/DML queries keep in the 
> CREATED state for several minutes. Especially when using with sync_ddl=true, 
> tens of minutes are also possible. They may be waiting for the ExecDdl RPC to 
> catalogd to finish.
> It'd be helpful for debugging DDL/DML hangs if we can show the in-flight DDLs 
> in catalogd. I think the following fields are important:
>  * thread id
>  * coordinator
>  * db name / table name
>  * ddl type, e.g. AddPartition, DropTable, CreateTable, etc. More types 
> [here|https://github.com/apache/impala/blob/3.3.0/common/thrift/JniCatalog.thrift#L31].
>  * last event, e.g. waiting for table lock, got table lock, loading file 
> metadata, waiting for sync ddl version etc.
>  * start time
>  * time elapsed
>  * (optional) params link to show the TDdlExecRequest in json format
> It'd be better to also include running REFRESH/INVALIDATE METADATA commands 



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to