#542 Add first draft for chronos-multithreading docs

docs/examples/multithreading.nim

@@ -0,0 +1,34 @@
+import std/[os]
+import chronos
+import chronos/threadsync
+
+proc cleanupThread*() {.gcsafe, raises: [].}=
+  {.cast(gcsafe).}:
+    try:
+      `=destroy`(getThreadDispatcher())
+      when defined(gcOrc):
+        GC_fullCollect()
+    except Exception as e:
+      echo "Exception during cleanup: ", e[]
+
+proc processAsync() {.async.} =
+  await sleepAsync(1000) # Waiting on an HTTP Request or the like
+  echo "Done with async work"
+
+proc threadProc(threadSignal: ThreadSignalPtr) {.thread.} =
+  block:
+    asyncSpawn processAsync() # Start some async work
+
+    waitFor threadSignal.wait() # Do async work and then go to sleep until main thread wakes you up
+
+  cleanupThread() # Clean up thread local variables, e.g. dispatcher
+
+var thread: Thread[ThreadSignalPtr]
+let signal = new(ThreadSignalPtr)[]
+
+createThread(thread, threadProc, signal)
+sleep(10)
+discard signal.fireSync()
+
+joinThread(thread)
+discard signal.close()

docs/src/multithreading_async.md

@@ -0,0 +1,10 @@
+# Multithreading with async event-loops
+
+At times you may want to run asynchronous work on another thread for whatever reason,
+then wait while working through the asynchronous work until you receive a signal to continue.
+
+This kind of setup can be facilitated by using `threadSync/ThreadSignalPtr`
+
+```nim
+{{#include ../examples/multithreading.nim}}
+```

docs/src/threads.md

@@ -16,3 +16,21 @@ to notify other threads of progress from within an async procedure.
 ```nim
 {{#include ../examples/signalling.nim}}
 ```
+
+### Memory management in multithreading scenarios
+
+Keep in mind that nim _does not clean up thread-variables_ when the thread exits.
+This is problematic for thread variables that are ref-types or contain ref-types, as those will not be cleaned up and thus start causing memory leaks.
+
+This means, that the dispatcher, which is a thread variable and a ref-type, needs to be cleaned up manually. The same is true for the thread-local variables `localInstance` and `utcInstance` of std/times, which will be used if you use any logging library such as chronicles.
+Not just that, you will also need to invoke ORC's cycle-collector manually, as chronos' data-structures are at times cyclical and thus need to be collected as well.
+
+You can do so by calling `=destroy` on it and forcing ORC to do a collect of cyclical data-structures like so:
+
+```nim
+`=destroy`(times.localInstance)
+`=destroy`(times.utcInstance)
+`=destroy`(getThreadDispatcher())
+when defined(gcOrc):
+  GC_fullCollect()
+```