memory-infra: Never kill memory-infra background thread

base/trace_event/memory_dump_manager.h

Index: base/trace_event/memory_dump_manager.h
diff --git a/base/trace_event/memory_dump_manager.h b/base/trace_event/memory_dump_manager.h
index 1583c9ee6c7b5d37e104f1763015bce07454ec64..21dbaa0562e9f436dcf5e88af13fe0645d89043b 100644
--- a/base/trace_event/memory_dump_manager.h
+++ b/base/trace_event/memory_dump_manager.h
@@ -245,6 +245,9 @@ class BASE_EXPORT MemoryDumpManager {
   void FinalizeDumpAndAddToTrace(
       std::unique_ptr<ProcessMemoryDumpAsyncState> pmd_async_state);
 
+  // Lazily initializes dump_thread_ and returns its TaskRunner.
+  scoped_refptr<base::SingleThreadTaskRunner> GetTaskRunner();

[Primiano Tucci (use gerrit), 2017/04/24 15:30:58]

I'd call this GetOrCreateBgTaskRunnerLocked()
"GetOrCreate", to reflect the fact that will lazily initialize a new one.
"Locked" to reflect the locking need.
"Bg" to reflect that this is our "background" thread.

Also, just a minor documentation thing. I think we could return a
SequencedTaskRunner here.
Nothing will change in practice in the .cc file (a SingleThread task runner is
also sequenced). Is just about the recent discussions on chromium-dev.
"SingleThread" should be used only in the cases where we actually need thread
affinity (read: we rely on a TLS slot, which IIRC is not the case here).
In all other cases where we just need sequencing, STR is the way to go.

[hjd, 2017/04/24 16:27:59]

Done.

+
   // Internal, used only by MemoryDumpManagerDelegate.
   // Creates a memory dump for the current process and appends it to the trace.
   // |callback| will be invoked asynchronously upon completion on the same