[jira] [Commented] (SLING-7813) SlingHttpServletResponseImpl should log when setStatus is called after it is committed

["Bertrand Delacretaz (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SLING-7813?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17414837#comment-17414837 ] 

Bertrand Delacretaz commented on SLING-7813:
--------------------------------------------

I stumbled on this while looking at SLING-10810 and I think [commit bc9696bd|https://github.com/apache/sling-org-apache-sling-engine/commit/bc9696bd4f52a3a2b50a19f2572b80a4ac2ea3a2] introduces a non-standard behavior, where setStatus just logs a message if the request is already committed, instead of throwing an exception like [HttpServletResponse.sendError|https://javaee.github.io/javaee-spec/javadocs/javax/servlet/http/HttpServletResponse.html#sendError-int-java.lang.String-] does.

Actually, IIUC that non-standard behavior only happens if setStatus(status, message) is called with a non-empty message, otherwise super.setStatus is called.

That's suboptimal IMO, but considering that these changes are more than 3 years old now there's probably no urgent need to change that back - I'll just leave that note here in case we want to revisit this in the future.

> SlingHttpServletResponseImpl should log when setStatus is called after it is committed
> --------------------------------------------------------------------------------------
>
>                 Key: SLING-7813
>                 URL: https://issues.apache.org/jira/browse/SLING-7813
>             Project: Sling
>          Issue Type: Improvement
>          Components: Engine
>    Affects Versions: Engine 2.6.12
>            Reporter: Julian Sedding
>            Assignee: Julian Sedding
>            Priority: Minor
>             Fix For: Engine 2.6.14
>
>          Time Spent: 0.5h
>  Remaining Estimate: 0h
>
> I have been debugging a scenario, where a response did not have the (in this case) expected status code {{500}} set by an error handling script, but instead the status code was {{200}}.
> It turns out that a rendering script calls {{#flushBuffer()}} on the response early on in order to optimize user experience. Later in the rendering chain a JSP causes a {{NullPointerException}}, triggering an error handler which calls {{#setStatus(500)}}. The {{#setStatus}} call is silently ignored.
> "Fixing" this problem would require buffering the entire response and ignoring any flush calls (be it {{#flushBuffer()}}, {{#getWriter().flush()}} or {{#getOutputStream().flush()}}). This would be a change in behaviour, a violation of the Servlet spec and performance issues waiting to happen. Thus I am ruling out this option.
> However, it would be helpful to improve "debuggability" of the problem. I propose to log a warning when {{#setStatus()}} is called. Additionally, if debug logging is enabled, I propose to log a stack trace to identify where the flush call originated (unless the flush was due to too many bytes written, which is not very helpful information).
> cc [~rombert]



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