[
https://issues.apache.org/jira/browse/SLING-8187?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16725828#comment-16725828
]
Julian Sedding commented on SLING-8187:
---------------------------------------
I created a PR. While I am confident that it improves the status-quo, I am not
100% happy due to the following draw-back:
* SlingServletResolver has a mandatory reference to SlingServletContext, thus
deferring registration of SlingServletContext causes registration of
SlingServletResolver
* SlingRequestProcessor (via SlingMainServlet) has a dynamic optional
reference to SlingServletResolver, but throws an exception causing a 503
response while no SlingServletResolver is bound
--> thus the time-window for this 503 response increases with this fix - with
the fix applied, I see it frequently if there are requests to my system during
startup; without the fix I could not observe any.
> Deadlock in SlingMainServlet after SLING-8051
> ---------------------------------------------
>
> Key: SLING-8187
> URL: https://issues.apache.org/jira/browse/SLING-8187
> Project: Sling
> Issue Type: Improvement
> Components: Engine
> Affects Versions: Engine 2.6.16
> Reporter: Julian Sedding
> Assignee: Julian Sedding
> Priority: Major
> Time Spent: 10m
> Remaining Estimate: 0h
>
> The fix for SLING-8051 moved registration of the {{SlingServletContext}}
> instance from {{SlingMainServlet#activate()}} to {{SlingMainServlet#init()}}.
> It turns out that this can cause a deadlock due to Felix' global lock and a
> synchronized block in Felix HTTP WhiteboardManager implementation.
> This issue is about finding a way to fix the issue originally described in
> SLING-8051 without trisking a deadlock.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
