[
https://issues.apache.org/jira/browse/FLINK-1319?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14569312#comment-14569312
]
ASF GitHub Bot commented on FLINK-1319:
---------------------------------------
Github user uce commented on the pull request:
https://github.com/apache/flink/pull/729#issuecomment-107996871
Hey Timo,
I've tested this locally so far and it's working smoothly! Let's write a
blog post about this very soon. :-) As soon as I have access to more machines,
I will test it on a cluster. The stuff denoted with [RB] should be fixed in any
case before merging imo. The rest we can also do afterwards.
**USER FACING**
1. Disable the analysis by default. [RB]
2. This is cumbersome, but we should go for Annotation-based exclusions
instead of package based (see inline comments). [RB]
3. Currently manual annotations trump automatic ones, which is a good
thing, because people won't have unexpected results. Could you add a test for
this?
4. I would give the analyzer util the call location based function name
(String callLocation = Utils.getCallLocationName()). I think that will have
better output than just `Function 'Job$2' has been analyzed with the following
result: ...`
4. I would rename `UdfAnalyisMode` (see inline comments).
5. The hints are currently logged on a singe line w/o a whitespace after
each hint, like this:
```
Function modifies static fields. This can lead to unexpected behaviour
during runtime.Function returns 'null' values. This can lead to errors during
runtime.A need for forwarded fields annotations could not be found.
```
6. Some analysis like a wrong tuple index access or returning null lead
fail the program before submitting it, which is very nice. I would actually
like to log these problems on a log level WARN (instead of INFO) when only
hints are enabled.
7. When you have optimize enabled and run a filter, which changes the
input, the error you get is this. You can't tell what's wrong.
```
Exception in thread "main" org.apache.flink.api.java.sca.UdfErrorException:
UDF contains obvious errors.
at
org.apache.flink.api.java.sca.UdfAnalyzer.analyze(UdfAnalyzer.java:300)
...
```
For a wrong tuple index access, it is correct:
```
Exception in thread "main" org.apache.flink.api.java.sca.UdfErrorException:
UDF contains obvious errors.
at
org.apache.flink.api.java.sca.UdfAnalyzer.analyze(UdfAnalyzer.java:300)
...
Caused by: org.apache.flink.api.java.sca.UdfErrorException: Function
contains tuple accesses with invalid indexes. This can lead to errors during
runtime.
at
org.apache.flink.api.java.sca.UdfAnalyzer.addHintOrThrowException(UdfAnalyzer.java:413)
...
```
I'm not sure if the thrown Execption should be an `UdfErrorException`
or an `UdfAnalysisException`?
8. When optimizing is enabled, I would still print the inferred forwarded
fields.
9. Regarding the result messages: personally, I think the hints/messages
could be more consise, e.g. for example skip the `(should be kept to a
minimum)` in the number of object creations msg or just say `Forwarded fields:
none` instead of `A need for forwarded fields annotations could not be found.`
**INTERNALS**
1. I think the internals could use more comments. It's easy to get the
general idea, but it would be nice to also get some high-level comments about
the ASM-related stuff. I didn't have time to dig deep into it.
---
Will report back after cluster tests.
> Add static code analysis for UDFs
> ---------------------------------
>
> Key: FLINK-1319
> URL: https://issues.apache.org/jira/browse/FLINK-1319
> Project: Flink
> Issue Type: New Feature
> Components: Java API, Scala API
> Reporter: Stephan Ewen
> Assignee: Timo Walther
> Priority: Minor
>
> Flink's Optimizer takes information that tells it for UDFs which fields of
> the input elements are accessed, modified, or frwarded/copied. This
> information frequently helps to reuse partitionings, sorts, etc. It may speed
> up programs significantly, as it can frequently eliminate sorts and shuffles,
> which are costly.
> Right now, users can add lightweight annotations to UDFs to provide this
> information (such as adding {{@ConstandFields("0->3, 1, 2->1")}}.
> We worked with static code analysis of UDFs before, to determine this
> information automatically. This is an incredible feature, as it "magically"
> makes programs faster.
> For record-at-a-time operations (Map, Reduce, FlatMap, Join, Cross), this
> works surprisingly well in many cases. We used the "Soot" toolkit for the
> static code analysis. Unfortunately, Soot is LGPL licensed and thus we did
> not include any of the code so far.
> I propose to add this functionality to Flink, in the form of a drop-in
> addition, to work around the LGPL incompatibility with ALS 2.0. Users could
> simply download a special "flink-code-analysis.jar" and drop it into the
> "lib" folder to enable this functionality. We may even add a script to
> "tools" that downloads that library automatically into the lib folder. This
> should be legally fine, since we do not redistribute LGPL code and only
> dynamically link it (the incompatibility with ASL 2.0 is mainly in the
> patentability, if I remember correctly).
> Prior work on this has been done by [~aljoscha] and [~skunert], which could
> provide a code base to start with.
> *Appendix*
> Hompage to Soot static analysis toolkit: http://www.sable.mcgill.ca/soot/
> Papers on static analysis and for optimization:
> http://stratosphere.eu/assets/papers/EnablingOperatorReorderingSCA_12.pdf and
> http://stratosphere.eu/assets/papers/openingTheBlackBoxes_12.pdf
> Quick introduction to the Optimizer:
> http://stratosphere.eu/assets/papers/2014-VLDBJ_Stratosphere_Overview.pdf
> (Section 6)
> Optimizer for Iterations:
> http://stratosphere.eu/assets/papers/spinningFastIterativeDataFlows_12.pdf
> (Sections 4.3 and 5.3)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
