Auto retire with shutdown on Windows #505
Conversation
Do retire with shutdown if $MACKEREL_AUTORETIREMENT is set.
wix/wrapper/wrapper_windows.go
Outdated
| if req.Cmd == svc.Shutdown && autoRetire() { | ||
| if err := h.retire(); err != nil { | ||
| h.elog.Error(stopEid, err.Error()) | ||
| s <- svc.Status{State: svc.Running, Accepts: svc.AcceptShutdown | svc.AcceptShutdown} |
There was a problem hiding this comment.
svc.AcceptShutdown | svc.AcceptShutdown can be replaced with single svc.AcceptShutdown ? (Or typo for svc.AcceptStop | svc.AcceptShutdown ?)
There was a problem hiding this comment.
Thanks your reviewing this. The duplicates must be removed. Will do
wix/wrapper/wrapper_windows.go
Outdated
| if req.Cmd == svc.Shutdown && autoRetire() { | ||
| if err := h.retire(); err != nil { | ||
| h.elog.Error(stopEid, err.Error()) | ||
| s <- svc.Status{State: svc.Running, Accepts: svc.AcceptShutdown | svc.AcceptShutdown} |
There was a problem hiding this comment.
Since in this case mackerel-agent itself has been stopped successfully, status reported here should not be svc.Running?
There was a problem hiding this comment.
If mackerel-agent did not accept the shutdown command unintentionally, how should behave?
There was a problem hiding this comment.
FYI, This part report status of current condition. If this return Running, service manager will send svc.Shutdown again in later.
There was a problem hiding this comment.
If mackerel-agent did not accept the shutdown command unintentionally, how should behave?
It's better to retry retiring, but should not block the shutdown process of Windows too long.
If this return Running, service manager will send svc.Shutdown again in later.
I didn't know that. Thank you to inform.
Overall it looks good to set svc.Running to tell service manager to send svc.Shutdown again, but I'm wondering what happens when Windows shutdown is cancelled during this retry.
If Windows shutdown might be cancelled, mackerel-agent service will be left as "Running", but actually mackerel-agent is not running anymore. If that may occurs we should avoid, but if not it looks safe to set svc.Running. Do you know whether such situation may happen or not?
There was a problem hiding this comment.
Windows service manager has a fixed timeout value (20 seconds) to wait for shutdown. When it goes over 20 seconds, the OS forcibly terminates.
There was a problem hiding this comment.
When it goes over 20 seconds, the OS forcibly terminates.
Understand. Thank you for explanation!
From my comment #505 (comment):
If Windows shutdown might be cancelled, mackerel-agent service will be left as "Running", but actually mackerel-agent is not running anymore. If that may occurs we should avoid, but if not it looks safe to set
svc.Running. Do you know whether such situation may happen or not?
What do you think about this? I imagine following process may happen:
- Shutdown started
- mackerel-agent service tries to stop, and…
- mackerel-agent itself stopped
- mackerel-agent retire failed
- Shutdown cancelled by some reason
shutdown -a, from Task Manager, or something else
In this case, will mackerel-agent service be "running" but agent is stopped?
There was a problem hiding this comment.
Yes, there seems to be more falls possibles.
- timeout while stopping agent
- failed to spawn
mackerel-agent retire -force - shutdown was canceled (as you mentioned)
- failed to send request of retire
I would like to make these be same as UNIX's one as possible. h.stop wait 10 seconds until terminate of mackerel-agent. If timed out, mackerel-agent will be terminated forcibly. So 1 should not be considered. 2 also not.
When failed to spawn mackerel-agent retire -force or the command exited with non-zero value, systemctl return error. I'm thinking this is same behavior as UNIX's one.
The case that shutdown is canceled (3), the session can only be used during the time-out period. So not a really cancelation. So I'm thinking terminating mackerel-agent is right thinkg.
4 possibly be occured on some environment. But we don't have idea for recover.
Either way, mackerel-agent will be detected errors by server caused by regularly posts stopped.
On UNIX, mackerel-agent make the host retired when AUTO_RETIREMENT=1. This change do the same on Windows with shutdown if MACKEREL_AUTO_RETIREMENT Is set 1. You need to set MACKEREL_AUTO_RETIREMENT
, you need it on system property of my computer. This is useful to avandon the VM host on EC2.