Bugzilla

Some time in the past mhowell suggested using a semaphore to determine whether there are other instances running. This would allow us to lessen the issue where launching an additional instance of the same installation will perform an update and thereby require the other instance to restart. Below is a very simple flow of how I think this could work.

1. Create or open the semaphore on startup.
2. Use the Install hash for the name.
3. Prefix with Global\ on Windows so it is available to all users
4 The security of the semaphore will need to allow other users to work with it (need to figure out which users should be able to). This would allow other processes to DOS the update.
5. Permissions will need to be set on the Linux and OS X semaphores files so other users can use it. This would allow other processes to DOS the update.
6. For exits and crashes it will need to release the semaphore and it should close the handle when the instance count is 0. Does anyone know if this will be an issue in our crash handling code?
7. After X amount of time has passed (channel specific) based on a user pref (prevents other user processes from preventing this) without updating due to the semaphore force the update. This will likely be done with a new update status and this will mitigate the ability to DOS the update.
8. UI will most likely be shown when it is going to force the update.
9. Change the update restart UI or possibly display a warning so the user is informed that clicking the button will force the update.

Note: Some people have had issues with users creating semaphores on Windows Terminal server so it should ignore the semaphore when it can't be created or opened.

Some time in the past mhowell suggested using a semaphore to determine whether there are other instances running. This would allow us to lessen the issue where launching an additional instance of the same installation will perform an update and thereby require the other instance to restart. Below is a very simple flow of how I think this could work.

1. Create or open the semaphore on startup.
2. Use the Install hash for the name.
3. Prefix with Global\ on Windows so it is available to all users
4 The security of the semaphore will need to allow other users to work with it (need to figure out which users should be able to). This would allow other processes to DOS the update.
5. Permissions will need to be set on the Linux and OS X semaphores files so other users can use it. This would allow other processes to DOS the update.
6. For exits and crashes it will need to release the semaphore and it should close the handle when the instance count is 0. Does anyone know if this will be an issue in our crash handling code?
7. After X amount of time has passed (channel specific) based on a user pref (prevents other user processes from preventing this) without updating due to the semaphore force the update. This will likely be done with a new update status and this will mitigate the ability to DOS the update.
8. UI will most likely be shown when it is going to force the update.
9. Change the update restart UI or possibly display a warning so the user is informed that clicking the button will force the update.

Note: Some people have had issues with users creating semaphores on Windows Terminal server so it should ignore the semaphore when it can't be created or opened.

I *think* this method should also be sufficient for the Update Agent to determine if there is a running instance for an update.

Some time in the past mhowell suggested using a semaphore to determine whether there are other instances running. This would allow us to lessen the issue where launching an additional instance of the same installation will perform an update and thereby require the other instance to restart. Below is a very simple flow of how I think this could work.

1. Create or open the semaphore on startup.
2. Use the Install hash for the name.
3. Prefix with Global\ on Windows so it is available to all users
4. The security of the semaphore will need to allow other users to work with it (need to figure out which users should be able to). This would allow other processes to DOS the update.
5. Permissions will need to be set on the Linux and OS X semaphores files so other users can use it. This would allow other processes to DOS the update.
6. For exits and crashes it will need to release the semaphore and it should close the handle when the instance count is 0. Does anyone know if this will be an issue in our crash handling code?
7. After X amount of time has passed (channel specific) based on a user pref (prevents other user processes from preventing this) without updating due to the semaphore force the update. This will likely be done with a new update status and this will mitigate the ability to DOS the update.
8. UI will most likely be shown when it is going to force the update.
9. Change the update restart UI or possibly display a warning so the user is informed that clicking the button will force the update.

Note: Some people have had issues with users creating semaphores on Windows Terminal server so it should ignore the semaphore when it can't be created or opened.

I *think* this method should also be sufficient for the Update Agent to determine if there is a running instance for an update.