Simulated Crash reporting for ErrAborted requests

content/browser/loader/resource_dispatcher_host_impl.cc

Index: content/browser/loader/resource_dispatcher_host_impl.cc
diff --git a/content/browser/loader/resource_dispatcher_host_impl.cc b/content/browser/loader/resource_dispatcher_host_impl.cc
index f342602626ab81b27875ce55c0effbbb83afdc6c..2b3b9b278d573201e25d3c509e57da0fff415706 100644
--- a/content/browser/loader/resource_dispatcher_host_impl.cc
+++ b/content/browser/loader/resource_dispatcher_host_impl.cc
@@ -15,6 +15,9 @@
 #include "base/command_line.h"
 #include "base/compiler_specific.h"
 #include "base/debug/alias.h"
+#include "base/debug/crash_logging.h"
+#include "base/debug/dump_without_crashing.h"
+#include "base/debug/stack_trace.h"
 #include "base/logging.h"
 #include "base/memory/scoped_ptr.h"
 #include "base/memory/shared_memory.h"
@@ -906,6 +909,18 @@ void ResourceDispatcherHostImpl::DidFinishLoading(ResourceLoader* loader) {
       case net::ERR_ABORTED:
         UMA_HISTOGRAM_LONG_TIMES(
             "Net.RequestTime2.ErrAborted", request_loading_time);
+        // 100ms is typically "instantaneous" in human perception.
+        if (request_loading_time.InMilliseconds() < 100) {
+          const char kFastErrAbortTrace[] = "fast-err-abort-trace";
+          const base::debug::StackTrace* stack_trace =
+              loader->request()->stack_trace();
+          if (stack_trace) {
+            base::debug::SetCrashKeyToStackTrace(kFastErrAbortTrace,

[eroman, 2015/11/16 22:56:11]

Instead of adding a crash key to the report, I think it will be easier to  just
alias a local variable:

base::Debug::StackTrace strack_trace_copy = *loader->request()->stack_trace();
base::Debug::Alias(&stack_trace_copy);
..
<dumpwithoutcrashing>


The advantage of this approach is you can easily symbolicate it for Windows
crash dumps using something like:
Also note that getting stack traces is somewhat unpredictable; it doesn't always
work. It is better now that we no longer omit the frame pointer, but I have done
this plenty of times and gotten back an unwalkable callstack, so don't hold your
breath :(

+                                                 *stack_trace);
+            base::debug::DumpWithoutCrashing();

[eroman, 2015/11/16 22:56:11]

Do you have any estimates on how often this will be called? Can you suggest some
limits or bounds on this?

As a safety-net, how about limiting this to either a total number per session
(say 1 or 2 for the lifetime of the browser), or maybe a time based threshold,
like 1 report at most every hour?

Also, put a link to the bug report by this line, so crash triagers will know
what it corresponds to.

+            base::debug::ClearCrashKey(kFastErrAbortTrace);
+          }
+        }
         break;
       case net::ERR_CONNECTION_RESET:
         UMA_HISTOGRAM_LONG_TIMES(