[Android Crash Reporting] Allow uploading minidumps via the JobScheduler

components/minidump_uploader/android/java/src/org/chromium/components/minidump_uploader/MinidumpUploaderImpl.java

 // Copyright 2016 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 package org.chromium.components.minidump_uploader;
 
 import org.chromium.base.Log;
 import org.chromium.base.ThreadUtils;
 import org.chromium.base.VisibleForTesting;
 
@@ skipping 30 matching lines @@
      * The thread used for the actual work of uploading minidumps.
      */
     private Thread mWorkerThread;
 
     @VisibleForTesting
     public static final int MAX_UPLOAD_TRIES_ALLOWED = 3;
 
     @VisibleForTesting
     public MinidumpUploaderImpl(MinidumpUploaderDelegate delegate) {
         mDelegate = delegate;
-        mFileManager = createCrashFileManager(mDelegate.createCrashDir());
+        mFileManager = createCrashFileManager(mDelegate.getCacheDir());
         if (!mFileManager.ensureCrashDirExists()) {
             Log.e(TAG, "Crash directory doesn't exist!");
         }
     }
 
     /**
      * Utility method to allow tests to customize the behavior of the crash file manager.
-     * @param {crashDir} The directory in which to store crash files (i.e. minidumps).
+     * @param {cacheDir} The cache directory, which contains the crash directory, in which crash
+     *     files (i.e. minidumps) are stored.
      */
     @VisibleForTesting
-    public CrashFileManager createCrashFileManager(File crashDir) {
-        return new CrashFileManager(crashDir);
+    public CrashFileManager createCrashFileManager(File cacheDir) {
+        return new CrashFileManager(cacheDir);
     }
 
     /**
      * Utility method to allow us to test the logic of this class by injecting
      * test-specific MinidumpUploadCallables.
      */
     @VisibleForTesting
     public MinidumpUploadCallable createMinidumpUploadCallable(File minidumpFile, File logfile) {
         return new MinidumpUploadCallable(
                 minidumpFile, logfile, mDelegate.createCrashReportingPermissionManager());
     }
 
     /**
      * Runnable that upload minidumps.
      * This is where the actual uploading happens - an upload job consists of posting this Runnable
      * to the worker thread.
      */
     private class UploadRunnable implements Runnable {
         private final MinidumpUploader.UploadsFinishedCallback mUploadsFinishedCallback;
 
         public UploadRunnable(MinidumpUploader.UploadsFinishedCallback uploadsFinishedCallback) {
             mUploadsFinishedCallback = uploadsFinishedCallback;
         }
 
         @Override
         public void run() {
             File[] minidumps = mFileManager.getAllMinidumpFiles(MAX_UPLOAD_TRIES_ALLOWED);
+            Log.i(TAG, "Attempting to upload %d minidumps ", minidumps.length);
             for (File minidump : minidumps) {
+                Log.i(TAG, "Attempting to upload " + minidump.getName());
                 MinidumpUploadCallable uploadCallable = createMinidumpUploadCallable(
                         minidump, mFileManager.getCrashUploadLogFile());
                 int uploadResult = uploadCallable.call();
 
                 // Bail if the job was canceled. Note that the cancelation status is checked AFTER
                 // trying to upload a minidump. This is to ensure that the scheduler attempts to
                 // upload at least one minidump per job. Otherwise, it's possible for a crash loop
                 // to continually write files to the crash directory; each such write would
                 // reschedule the job, and therefore cancel any pending jobs. In such a scenario,
                 // it's important to make at least *some* progress uploading minidumps.
                 // Note that other likely cancelation reasons are not a concern, because the upload
                 // callable checks relevant state prior to uploading. For example, if the job is
                 // canceled because the network connection is lost, or because the user switches
                 // over to a metered connection, the callable will detect the changed network state,
                 // and not attempt an upload.
                 if (mCancelUpload) return;
 
                 // Note that if the job was canceled midway through, the attempt number is not
                 // incremented, even if the upload failed. This is because a common reason for
                 // cancelation is loss of network connectivity, which does result in a failure, but
                 // it's a transient failure rather than something non-recoverable.
                 if (uploadResult == MinidumpUploadCallable.UPLOAD_FAILURE) {
                     String newName = CrashFileManager.tryIncrementAttemptNumber(minidump);
                     if (newName == null) {
                         Log.w(TAG, "Failed to increment attempt number of " + minidump);
                     }
                 }
+
+                // TODO(isherman): Log success/failure metrics for Chrome.

[Ilya Sherman, 2017/03/13 03:12:09]

I'm actually not sure how to do this.  The upload job might run after the Chrome
process has terminated.  It seems overly expensive to spin up a bunch of
Chrome's startup code only to record metrics.

An alternate idea that I had is to record metrics about the pending crash
reports at startup -- how many are failed, how many uploaded successfully, etc. 
It doesn't give us quite the same fine-grained view, but it's in roughly the
same spirit, and is far more straightforward to implement.  WDYT?

[gsennton, 2017/03/13 17:57:17]

Sounds reasonable (but I don't know much about UMA).

[Ilya Sherman, 2017/03/14 02:18:55]

Actually, now that I realize that it's okay to use the shared prefs from the
job, I think it's simpler to just log success/failure metrics as we did before.

[gsennton, 2017/03/14 18:17:28]

It's fine as long as we do it just from the browser process I think (i.e. as
long as we don't create a separate process for uploading minidumps). SharedPrefs
are thread-safe but not process-safe (I think).

             }
 
             // Clean out old/uploaded minidumps. Note that this clean-up method is more strict than
             // our copying mechanism in the sense that it keeps fewer minidumps.
             mFileManager.cleanOutAllNonFreshMinidumpFiles();
 
             // Reschedule if there are still minidumps to upload.
             boolean reschedule =
                     mFileManager.getAllMinidumpFiles(MAX_UPLOAD_TRIES_ALLOWED).length > 0;
             mUploadsFinishedCallback.uploadsFinished(reschedule);
@@ skipping 38 matching lines @@
 
         // Reschedule if there are still minidumps to upload.
         return mFileManager.getAllMinidumpFiles(MAX_UPLOAD_TRIES_ALLOWED).length > 0;
     }
 
     @VisibleForTesting
     public void joinWorkerThreadForTesting() throws InterruptedException {
         mWorkerThread.join();
     }
 }