[WIP] Maintenance tasks in service

[wilbaker]

This is extremely WIP.

Resolves #112

TODOs:

• Wire up maintenance scheduler on Windows
• Add logging where appropriate
• Update unit tests
• Manually test changes on Windows and Mac
• Move string computations to the constructor (or start/init method) where appropriate
• Decide if maintenance tasks should only run for the logged in user on Windows (@derrickstolee and I agreed we can start with that approach)
• Cleanup commit history

Follow up work (outside the scope of this PR):

• [Mount Removal] Move repo registration from mount verb to clone verb #111 [Mount Removal] Move repo registration from mount verb to clone verb
• [Mount Removal] Remove 'mount' and 'unmount' verbs, and remove the service verb options for mounting and unmounting #118 [Mount Removal] Remove 'mount' and 'unmount verbs'
• [Mount Removal] Delete Scalar.Mount.* projects #132 [Mount Removal] Delete Scalar.Mount.* projects
• [Mount Removal] The 'fetch-commits-and-trees' task should check if a cache server is configured when run by the service #174 [Mount Removal] The 'fetch-commits-and-trees' task should check if a cache server is configured when run by the service
• Stop creating so many maintenance log files #177 Stop creating so many maintenance log files
• [Mount Removal] Scalar.Service should provide a way to kill an active maintenance task, and it should be used when upgrading #185 [Mount Removal] Scalar.Service should provide a way to kill an active maintenance task, and it should be used when upgrading
• [Mount Removal] The maintenance tasks need a replacement for 'EnlistmentRootReady' #186 [Mount Removal] The maintenance tasks need a replacement for 'EnlistmentRootReady'
• Improve Scalar.Service's handling of concurrently logged in users and tokens #187 Improve Scalar.Service's handling of concurrently logged in users

[derrickstolee]

I like the direction. Look at the GitMaintenanceQueue for directions for some of your TODOs.

[derrickstolee]

To allow stopping, you'll want to create a Process object and store it as a private member. A Stop() method would then allow killing the process, creating an exception through the CallMaintenance() method.

[wilbaker]

@derrickstolee I plan to tackle this work as a follow up in #185

On the Mac side I think it should be pretty easy/straightforward as we're already using Process objects.

However, on the Windows side we have a big mess of native calls that ultimately end with a call to CreateProcessAsUser:

scalar/Scalar.Platform.Windows/CurrentUser.cs

Line 140 in 9018727

if (CreateProcessAsUser(

And now that I take another look we're not even waiting for that process to finish ☹️.

We might end up needing an interface that will internally use Process objects on Mac, and store PROCESS_INFORMATION on Windows (and then we can PInvoke to TerminateProcess to stop the process).

I will also do some research and see if there is a way to create a Process object for the process we're spawning on Windows. (GetProcessById might be an option, but we'd have to make sure we're not introducing any races if the process exits quickly).

[derrickstolee]

I don't know why, but when I tried to build your tip commit, my build failed because ScalarService.Windows.cs referred to ScalarMountProcess. But when I use the pull request merge commit, everything builds fine.

Anyway, I'm installing on my laptop to get some extra checks that things are working.

[wilbaker]

@derrickstolee

I don't know why, but when I tried to build your tip commit, my build failed

That might be on me, I did not have a chance to confirm the latest commit built on Windows before I headed out. I'll check that first thing when I get in tomorrow.

Also, I'm going to switch back to using a queue for the tasks (like we had before). The queue approach is battle tested, and it should be pretty easy to adjust the existing unit tests to work with a new queue. Plus as we discussed offline the edge case of "task taking too long" should not be a common occurrence.

[wilbaker]

to get some extra checks that things are working

FYI - I haven't had a chance to do any manual tests yet and so things might not be working correctly yet.

My plan for tomorrow:

• Switch back to queue and bring back its unit tests
• Run some manual tests
• Create a new non-WIP PR with a cleaned up branch

I plan to defer the work to have the functional tests not register when cloning (at least on Mac) until we make the changes to register in clone. If any of the maintenance tests get flaky before then we can have them unmount after cloning (if they aren't already).

[wilbaker]

Closing out this PR, it will be replaced with a non-WIP PR.