2009-06-03 15:02:14 |
Thomas Herve |
description |
The watchdog automatically restarts the broker if it dies for some reason. In this case, the client appears to work (some exchange happens, it pings the server correctly), but in fact the state of the system is pretty wrong:
* manager mostly doesn't work
* monitoring doesn't work
* there is no package information upload.
What happens is that some calls (register_message, register_plugin) are bound the current running broker and are only done are start up: if the broker is restarted, they are not done again.
The easy solution I can think is to restart the client completely if the broker dies. We may come with some smart solution but it looks like it would be tricky to think about all the cases. |
The watchdog automatically restarts the broker if it dies for some reason. In this case, the client appears to work (some exchange happens, it pings the server correctly), but in fact the state of the system is pretty wrong:
* manager mostly doesn't work
* monitoring doesn't work
* there is no package information upload.
What happens is that some calls (register_message, register_plugin) are bound the current running broker and are only done are start up: if the broker is restarted, they are not done again.
We need to tell the monitor and manager plugins that the broker has restarted, so that they registered themselves and the message types again. |
|