[ 
https://issues.apache.org/jira/browse/HIVE-25140?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17389774#comment-17389774
 ] 

Zoltan Haindrich commented on HIVE-25140:
-----------------------------------------

using aspect oriented programming as well doesn't mean you can't directly use 
the api - but it will be less disturbing (and easier to add).

you say it doesn't affect a lot of code - the latest patch is 734K long and it 
only touches VectorMapOperator for some exception. I think this approach is 
simply bad - it will just miss things here and there...and the patch is just 
getting bigger and bigger...

bq.  The very nature of manually instrumenting code like Hive to do tracing is 
to start at the top of execution (e.g. BeeLine's SQL Statement) and judicially 
look for large areas of execution that would provide us benefit from a Span.

I think this approach is usable when you are looking after a concreate problem 
and not developing a profiling tool for the system - for hive we should be 
doing the latter ; we don't know what issues will we encounter in the future.

The patch also introduces "traceable" classes which will be painfull to 
maintain.

I think the annotation aspect with an online feature switch plus a big compile 
time feature disable toggle would be the best; it wouldn't affect much things - 
people may even run the whole system without the tracing code even in the 
binaries.

> Hive Distributed Tracing -- Part 1: Disabled
> --------------------------------------------
>
>                 Key: HIVE-25140
>                 URL: https://issues.apache.org/jira/browse/HIVE-25140
>             Project: Hive
>          Issue Type: Sub-task
>            Reporter: Matt McCline
>            Assignee: Matt McCline
>            Priority: Major
>         Attachments: HIVE-25140.01.patch, HIVE-25140.02.patch, 
> HIVE-25140.03.patch
>
>
> Infrastructure except exporters to Jaeger or OpenTelementry (OTL) due to 
> Thrift and protobuf version conflicts. A logging only exporter is used.
> There are Spans for BeeLine and Hive. Server 2. The code was developed on 
> branch-3.1 and porting Spans to the Hive MetaStore on master is taking more 
> time due to major metastore code refactoring.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to