Workaround for shutdown error #722
Conversation
This commit is a workaround for an error which seems to happen since the upgrade to Micronaut 4.6. In a nutshell, when the `/stop` endpoint is called, the thread which was started to monitor if the service is properly shutdown after a timeout was started and the application context was closed, _before_ the `stop` method would return something to the client. As a consequence, there was an error message saying that the application context wasn't open and that a bean (the message writers) weren't found. This was not quite true, since the application context _used to be_ open but wasn't. The workaround, which isn't great, is to use the task scheduler to delay the shutdown by a few hundreds of milliseconds. This gives the opportunity to send the response back to the client _before_ the application context is shutdown. Note that in any case, this wasn't a big issue, since the service would be shutdown anyway, but the error message for the user wasn't great.
|
There was a problem hiding this comment.
Does this deserve any further investigation in core? @yawkat @dstepanov @graemerocher
|
In 4.6 @graemerocher added a check to prevent accessing the bean context when it's shutdown. |
|
correct. The context used to be more forgiving in terms of allowing access to beans even after shutdown. That is no longer the case and it was a bug that it was possible. That said this seems to be related to our lack of proper support for graceful shutdown which needs to be correctly ideally in 4.7 |
|
So then, should this logic be handled with a server shutdown event listener, instead of a custom controller method? |
|
not sure how an event listener would help |
This commit is a workaround for an error which seems to happen since the upgrade to Micronaut 4.6. In a nutshell, when the
/stopendpoint is called, the thread which was started to monitor if the service is properly shutdown after a timeout was started and the application context was closed, before thestopmethod would return something to the client. As a consequence, there was an error message saying that the application context wasn't open and that a bean (the message writers) weren't found. This was not quite true, since the application context used to be open but wasn't.The workaround, which isn't great, is to use the task scheduler to delay the shutdown by a few hundreds of milliseconds. This gives the opportunity to send the response back to the client before the application context is shutdown.
Note that in any case, this wasn't a big issue, since the service would be shutdown anyway, but the error message for the user wasn't great.