[QUESTION]mdadm4.1 upgrade to mdadm4.2,mdmonitor services failed to start if no raid in environment

[QUESTION]mdadm4.1 upgrade to mdadm4.2,mdmonitor services failed to start if no raid in environment

[miaoguanqin]

Hi
  Here we meet a question. After we upgrade mdadm from 4.1 to 4.2, we 
execute :
     systemctl start mdmonitor
the mdmonitor service failed to start when no raid device in 
environment. error message are as follows:

mdmonitor.service - MD array monitor
      Loaded: loaded (/usr/lib/systemd/system/mdmonitor.service; 
enabled; vendor preset: enabled)
      Active: failed (Result: protocol) since Thu 2023-05-11 10:52:32 
CST; 5s ago
     Process: 999741 ExecStartPre=mkdir -p /run/mdadm (code=exited, 
status=0/SUCCESS)
     Process: 999743 ExecStart=/sbin/mdadm --monitor $MDADM_MONITOR_ARGS 
-f --pid-file=/run/mdadm/mdadm.pid (c>

May 11 10:52:32 localhost.localdomain systemd[1]: Starting MD array 
monitor...
May 11 10:52:32 localhost.localdomain systemd[1]: mdmonitor.service: 
Can't open PID file /run/mdadm/mdadm.pid>
May 11 10:52:32 localhost.localdomain systemd[1]: mdmonitor.service: 
Failed with result 'protocol'.
May 11 10:52:32 localhost.localdomain systemd[1]: Failed to start MD 
array monitor.

In the mdmonitor service file, type is set to forking and the PIDFILE 
field is set. The systemd detection process is as follows:
(1)when the parent process exits, a signal is sent to systemd
(2)systemd wakes up and checks whether pidfile exists by PIDFILE field.
(3)If the pidfile file does not exist, the service status is set to fail.
In function Monitor code logic, after the parent process creates a 
pidfile, before systemd detects the pidfile, the pidfile is deleted from 
the child process. As a result, the systemd cannot detect the pidfile 
and sets the service status to fail.

It is a problem for user, because the mdmonitor service status is fail.
If there is no RAID device in the environment, We want the service 
status is expected to be inactive after the service is started. Can you 
have any advice for this problem ?

Re: [QUESTION]mdadm4.1 upgrade to mdadm4.2,mdmonitor services failed to start if no raid in environment

[Blazej Kucman]

On Thu, 11 May 2023 15:56:31 +0800
miaoguanqin <miaoguanqin@huawei.com> wrote:

> Hi
>   Here we meet a question. After we upgrade mdadm from 4.1 to 4.2, we 
> execute :
>      systemctl start mdmonitor
> the mdmonitor service failed to start when no raid device in 
> environment. error message are as follows:
> 
> mdmonitor.service - MD array monitor
>       Loaded: loaded (/usr/lib/systemd/system/mdmonitor.service; 
> enabled; vendor preset: enabled)
>       Active: failed (Result: protocol) since Thu 2023-05-11 10:52:32 
> CST; 5s ago
>      Process: 999741 ExecStartPre=mkdir -p /run/mdadm (code=exited, 
> status=0/SUCCESS)
>      Process: 999743 ExecStart=/sbin/mdadm --monitor
> $MDADM_MONITOR_ARGS -f --pid-file=/run/mdadm/mdadm.pid (c>
> 
> May 11 10:52:32 localhost.localdomain systemd[1]: Starting MD array 
> monitor...
> May 11 10:52:32 localhost.localdomain systemd[1]: mdmonitor.service: 
> Can't open PID file /run/mdadm/mdadm.pid>
> May 11 10:52:32 localhost.localdomain systemd[1]: mdmonitor.service: 
> Failed with result 'protocol'.
> May 11 10:52:32 localhost.localdomain systemd[1]: Failed to start MD 
> array monitor.
> 
> In the mdmonitor service file, type is set to forking and the PIDFILE 
> field is set. The systemd detection process is as follows:
> (1)when the parent process exits, a signal is sent to systemd
> (2)systemd wakes up and checks whether pidfile exists by PIDFILE
> field. (3)If the pidfile file does not exist, the service status is
> set to fail. In function Monitor code logic, after the parent process
> creates a pidfile, before systemd detects the pidfile, the pidfile is
> deleted from the child process. As a result, the systemd cannot
> detect the pidfile and sets the service status to fail.
> 
> It is a problem for user, because the mdmonitor service status is
> fail. If there is no RAID device in the environment, We want the
> service status is expected to be inactive after the service is
> started. Can you have any advice for this problem ?

Hi,

I do not know if the service should be in this state for such a case,
but the start of the mdmonitor service is also forced by udev rule
udev-md-raid-arrays.rules

# ENV{MD_LEVEL}=="raid[1-9]*", ENV{SYSTEMD_WANTS}+="mdmonitor.service"

so if the service is failed, it will be started anyway after creating
a new RAID.

Regards,
Blazej