2.4.37 (CentOS 8): event MPM: child process blocked out with Async connections stuck in the closing state

[<hostalp@xxxxxxx>]

Hello,

we've been performing some load tests (using Apache JMeter) and noticed that one HTTP Server child process become effectively blocked out (accepting = no) with 67 Async connections stuck in the closing state (see the server status output below).

It's been like that for several hours already and hasn't cleared yet (I intentionally left it in that state to see if it would eventually clear out). At the OS level there are no TCP connections (to this web server) visible at all, everything has been fully closed.

Does anyone have any idea what may cause that - whether it's a HTTP Server bug or some weird misconfiguration issue?

OS is CentOS 8 fully updated, with httpd installed from the standard repository.

Server status:

Server Version: Apache/2.4.37 (centos)

Server MPM: event
Server Built: Dec 23 2019 20:45:34

Current Time: Wednesday, 29-Apr-2020 19:24:53 CEST
Restart Time: Wednesday, 29-Apr-2020 01:01:20 CEST
Parent Server Config. Generation: 1
Parent Server MPM Generation: 0
Server uptime: 18 hours 23 minutes 32 seconds
Server load: 0.04 0.26 0.18
1 requests currently being processed, 999 idle workers

Slot    PID    Stopping    Connections     Threads    Async connections
total    accepting    busy    idle    writing    keep-alive    closing
0    10301    no    0    yes    0    200    0    0    0
1    10805    no    0    yes    0    200    0    0    0
2    25559    no    67    no    0    200    0    0    67
3    26063    no    1    yes    1    199    0    0    0
4    10806    no    0    yes    0    200    0    0    0
Sum    5    0    68         1    999    0    0    67

Some  related config:

MaxKeepAliveRequests    1000
KeepAliveTimeout           5

ThreadLimit              200
ServerLimit                5
StartServers               1
MaxRequestWorkers       1000
MinSpareThreads           40
MaxSpareThreads          240
ThreadsPerChild          200
MaxConnectionsPerChild     0

AsyncRequestWorkerFactor   1

(I had to lower the AsyncRequestWorkerFactor from the default value (2) because of issues with server closing some keep-alive connections prematurely in some highly variable workload cases, causing issues with Java clients using the Apache Http Client which doesn't seem to handle this well.)

Best regards,

Petr