Multiprocess sharing on .NET Core

[nblumhardt]

Adds a new configuration for netcoreapp1.0, as although Mutex is available on netstandard1.3 I think it's unlikely to work on all platforms targeted by that TFM (UWP doesn't like System.Threading references, for instance).

Performance is about 10x slower than atomic append, but otherwise appears to work as far as my limited tests have covered.

Required for serilog/serilog-sinks-rollingfile#37.

[nblumhardt]

I'll remove this from the example, since it's possible people will copy it into their projects and sacrifice performance for no reason.

[nblumhardt]

On this and the other WaitOne() call below, should we add a timeout and fail if the mutex can't be acquired? E.g. a 10s timeout is highly unlikely to be hit, but would at least keep things alive.

[nblumhardt]

Ignore the comment about UWP: it's System.Threading.Thread it rejects. Will update the PR.

[nblumhardt]

Alrighty, I think this is ready to go.

[nblumhardt]

(Does also need a test on Linux. Assumed Travis was churning away as it does for serilog/serilog but not so. Might do that from dev along with macOS.)

[skomis-mm]

I was working on the same thing :-) Hope it helps.
Also tests should be corrected at place where
const string InvalidPath = "/\\"; to pass on Linux

In my tests with this perf is about 1.5x slower than atomic_append. On Ubuntu its about 3x times slower, but it also ~3x times slower with shared=false

[skomis-mm]

I think its better to name it with the full path name to ensure uniqueness and better for crossplat, something like

Path.GetFullPath(path).Replace(Path.DirectorySeparatorChar, ':') + MutexNameSuffix

[nblumhardt]

Good point, will do.

[skomis-mm]

I believe you should also handle AbandonedMutexException in case of another process crash and also protect against async exceptions with help of return value of WaitOne():

bool ownsMutex = false;
try
{
    try
    {
        ownsMutex = _syncRoot.WaitOne(MutexWaitTimeout);
    }
    catch (System.Threading.AbandonedMutexException)
    {
        ownsMutex = true;
    }

    // ...
}
finally
 {
      if (ownsMutex)
      {
          _syncRoot.ReleaseMutex();
      }
 }

[nblumhardt]

Thanks, I'll look into it 👍

[nblumhardt]

Sorry about that, I should have made a note here - I just picked it up this afternoon when I realised how much file-related work was queued up :-)

My vague "10x" was with two processes writing on Windows. When there's contention it really seems to slow down. Probably not noticeable in practice though.

Thanks for checking the code, I'll run back through it in the morning.

[nblumhardt]

I think everything should be covered now.

[skomis-mm]

Looks good 👍