Delay asynchronous work from the metrics clients until after startup.

chrome/browser/metrics/chrome_metrics_service_client.cc

Index: chrome/browser/metrics/chrome_metrics_service_client.cc
diff --git a/chrome/browser/metrics/chrome_metrics_service_client.cc b/chrome/browser/metrics/chrome_metrics_service_client.cc
index da34086e4d4f1db29e5757b68223a22ba1a5188d..c94f55f7887c412b7b639c590b081521a69693e5 100644
--- a/chrome/browser/metrics/chrome_metrics_service_client.cc
+++ b/chrome/browser/metrics/chrome_metrics_service_client.cc
@@ -22,6 +22,7 @@
 #include "base/strings/string16.h"
 #include "base/threading/platform_thread.h"
 #include "build/build_config.h"
+#include "chrome/browser/after_startup_task_utils.h"
 #include "chrome/browser/browser_process.h"
 #include "chrome/browser/chrome_notification_types.h"
 #include "chrome/browser/google/google_brand.h"
@@ -138,9 +139,10 @@ CreateInstallerFileMetricsProvider(bool metrics_reporting_enabled) {
   // Fetch a worker-pool for performing I/O tasks that are not allowed on
   // the main UI thread.
   scoped_refptr<base::TaskRunner> task_runner =
-      content::BrowserThread::GetBlockingPool()
-          ->GetTaskRunnerWithShutdownBehavior(
-              base::SequencedWorkerPool::CONTINUE_ON_SHUTDOWN);
+      new AfterStartupTaskUtils::Runner(
+          content::BrowserThread::GetBlockingPool()
+              ->GetTaskRunnerWithShutdownBehavior(
+                  base::SequencedWorkerPool::CONTINUE_ON_SHUTDOWN));
 
   // Create an object to monitor files of metrics and include them in reports.
   std::unique_ptr<metrics::FileMetricsProvider> file_metrics_provider(
@@ -419,8 +421,8 @@ void ChromeMetricsServiceClient::Initialize() {
 #endif
   metrics_service_->RegisterMetricsProvider(
       std::unique_ptr<metrics::MetricsProvider>(
-          new metrics::NetworkMetricsProvider(
-              content::BrowserThread::GetBlockingPool())));
+          new metrics::NetworkMetricsProvider(new AfterStartupTaskUtils::Runner(
+              content::BrowserThread::GetBlockingPool()))));

[Alexei Svitkine (slow), 2016/08/12 18:28:18]

Similar to AntiVirusMetricsProvider, this could delay getting the data in time.

 
   // Currently, we configure OmniboxMetricsProvider to not log events to UMA
   // if there is a single incognito session visible. In the future, it may
@@ -480,12 +482,14 @@ void ChromeMetricsServiceClient::Initialize() {
       std::unique_ptr<metrics::MetricsProvider>(
           new browser_watcher::WatcherMetricsProviderWin(
               chrome::GetBrowserExitCodesRegistryPath(),
-              content::BrowserThread::GetBlockingPool())));
-
-  antivirus_metrics_provider_ = new AntiVirusMetricsProvider(
-      content::BrowserThread::GetBlockingPool()
-          ->GetTaskRunnerWithShutdownBehavior(
-              base::SequencedWorkerPool::CONTINUE_ON_SHUTDOWN));
+              new AfterStartupTaskUtils::Runner(
+                  content::BrowserThread::GetBlockingPool()))));

[Alexei Svitkine (slow), 2016/08/12 18:28:18]

This seems fine. Used to clean up registry data when metrics is off.

+
+  antivirus_metrics_provider_ =
+      new AntiVirusMetricsProvider(new AfterStartupTaskUtils::Runner(

[Alexei Svitkine (slow), 2016/08/12 18:28:18]

This will delay running GetAntiVirusProductsOnFileThread(). (Which is not well
named, as an aside.)

If it runs too late, ProvideSystemProfileMetrics() will not have AV info in
time.

Seems like this could affect behavior. I would want us to verify things or add a
histogram to measure the impact of this, rather than just blindly landing this.

[gab, 2016/08/15 19:39:56]

For startup's sake on slow machines (forgetting the initial re-ordering premise
of this CL for a second), should we not delay the first UMA ping until this
information has been gathered instead of rushing to acquire it while the machine
is under heavy load on startup? This doesn't appear critical to painting the
first frame and hence we should be happy to delay it.

Most users will complete startup in the first 30 seconds and then would schedule
these tasks (with a random 0-10s delay, see after_startup_task_utils.cc). So for
most users we should have this information in time (1 minute IIUC), for others,
gathering it is probably hurting anyways and delaying the first UMA ping (or
flagging it as missing the info) sounds better to me.

From offline discussion : delaying regular ping is okay but not stability ping
(I'm not sure about details here).

+          content::BrowserThread::GetBlockingPool()
+              ->GetTaskRunnerWithShutdownBehavior(
+                  base::SequencedWorkerPool::CONTINUE_ON_SHUTDOWN)));
 
   metrics_service_->RegisterMetricsProvider(
       std::unique_ptr<metrics::MetricsProvider>(antivirus_metrics_provider_));