Steps to reproduce

1. Create a confluence page on a test instance with many inline tasks and publish it. 
2. Rapidly untick as many inline tasks as you can as quickly as you can.

Results

1. Observe The HTTP thread stuck status in the Catalina.out logs, which will rapidly grow

   27-Aug-2021 16:25:20.259 - [1] thread(s) stuck.
   27-Aug-2021 16:25:30.265 - [2] thread(s) stuck.
   27-Aug-2021 16:25:30.300 - [3] thread(s) stuck.
   27-Aug-2021 16:25:30.302 - [4] thread(s) stuck.
   .
   .
   .
   27-Aug-2021 16:27:00.662 - [44] thread(s) stuck.
   27-Aug-2021 16:27:00.664 - [45] thread(s) stuck.
   27-Aug-2021 16:27:10.669 - [46] thread(s) stuck.
   27-Aug-2021 16:27:10.671 - [47] thread(s) stuck.
   

2. Eventually all the HTTP threads will be exhausted.
3. Catalina logs will have inlinetasks related threads in a STUCK state -

   27-Aug-2021 16:35:36.650 WARNING [Catalina-utility-1] org.apache.catalina.valves.StuckThreadDetectionValve.notifyStuckThreadDetected Thread [http-nio-8080-exec-48] (id=[1331]) has been active for [61,172] milliseconds (since [8/27/21 4:34 PM]) to serve the same request for [https://localhost:8090/rest/inlinetasks/1/task/1039232143/905/] and may be stuck (configured threshold for this StuckThreadDetectionValve is [60] seconds). There is/are [48] thread(s) in total that are monitored by this Valve and may be stuck.
   	java.lang.Throwable
   		at java.base@11.0.7/jdk.internal.misc.Unsafe.park(Native Method)
                   . . .
   		at com.atlassian.confluence.plugins.merge.TDMMerger.mergeContent(TDMMerger.java:103)
   		at com.atlassian.confluence.plugins.merge.TDMMerger.mergeContent(TDMMerger.java:118)
   		at com.atlassian.confluence.util.diffs.PageLayoutAwareMerger.mergeContent(PageLayoutAwareMerger.java:53)
   		at com.atlassian.confluence.pages.DefaultDraftManager.mergeContent(DefaultDraftManager.java:149)
   .
   .
   		at org.springframework.aop.support.DelegatingIntroductionInterceptor.invoke(DelegatingIntroductionInterceptor.java:124)
   		at org.springframework.aop.framework.ReflectiveMethodInvocation.proceed(ReflectiveMethodInvocation.java:186)
   		at org.springframework.aop.framework.JdkDynamicAopProxy.invoke(JdkDynamicAopProxy.java:212)
   		at com.sun.proxy.$Proxy2032.mergeContent(Unknown Source)
   		at com.atlassian.confluence.plugins.tasklist.service.DefaultInlineTaskService.updatePage(DefaultInlineTaskService.java:361)
   		at com.atlassian.confluence.plugins.tasklist.service.DefaultInlineTaskService.setTaskStatus(DefaultInlineTaskService.j
   

This would cause a temporary outage for the node in case of DC.

Workaround

The first workaround (listed below) is easily the most effective, as it eliminates the problem completely. If it is not possible to influence user behaviour, workaround 2 can provide an impact, but the problem will still remain to some extent.

Workaround-1 (Recommended workaround)

Open the Page in EDIT mode and then try to update the tasks which will not create multiple drafts in the background which could avoid spawning of new HTTP threads.

Workaround-2

Reduce page size, which has a proven effect on inline task thread execution time. The following values were averaged over approximately 10 tests per item).

• The impact of page version count is negligible:

  Versions	Task execution time
  20	169 ms
  5800	122 ms
  17000	155 ms

• However, page size has an impact on task change execution time:

  Size	Task execution time
  1280 bytes	93 ms
  522075 bytes	1.32 sec
  1003467 bytes	1.85 sec
  1890955 bytes	4.20 sec
  2815461 bytes	6.45 sec
  4120091 bytes	8.93 sec
  4837173 bytes	11.20 sec