Bug 1214408 - Telemetry on SessionStore:update OOM;r=ttaubert

--HG--
extra : transplant_source : %3A%DB%3Cp%8Fj%24%CF%94%E9%5C%AA%17%0E%2C%03E%23%A2%98

browser/components/sessionstore/SessionStore.jsm

@@ -72,6 +72,9 @@ const MESSAGES = [
   // A crashed tab was revived by navigating to a different page. Remove its
   // browser from the list of crashed browsers to stop ignoring its messages.
   "SessionStore:crashedTabRevived",
+
+  // The content script encountered an error.
+  "SessionStore:error",
 ];
 
 // The list of messages we accept from <xul:browser>s that have no tab
@@ -85,6 +88,9 @@ const NOTAB_MESSAGES = new Set([
 
   // For a description see above.
   "SessionStore:update",
+
+  // For a description see above.
+  "SessionStore:error",
 ]);
 
 // The list of messages we accept without an "epoch" parameter.
@@ -95,6 +101,9 @@ const NOEPOCH_MESSAGES = new Set([
 
   // For a description see above.
   "SessionStore:crashedTabRevived",
+
+  // For a description see above.
+  "SessionStore:error",
 ]);
 
 // The list of messages we want to receive even during the short period after a
@@ -106,6 +115,9 @@ const CLOSED_MESSAGES = new Set([
 
   // For a description see above.
   "SessionStore:update",
+
+  // For a description see above.
+  "SessionStore:error",
 ]);
 
 // These are tab events that we listen to.
@@ -808,6 +820,9 @@ var SessionStoreInternal = {
       case "SessionStore:crashedTabRevived":
         this._crashedBrowsers.delete(browser.permanentKey);
         break;
+      case "SessionStore:error":
+        this.reportInternalError(data);
+        break;
       default:
         throw new Error(`received unknown message '${aMessage.name}'`);
         break;
@@ -3848,6 +3863,19 @@ var SessionStoreInternal = {
    */
   resetEpoch(browser) {
     this._browserEpochs.delete(browser.permanentKey);
+  },
+
+  /**
+   * Handle an error report from a content process.
+   */
+  reportInternalError(data) {
+    // For the moment, we only report errors through Telemetry.
+    if (data.telemetry) {
+      for (let key of Object.keys(data.telemetry)) {
+        let histogram = Telemetry.getHistogramById(key);
+        histogram.add(data.telemetry[key]);
+      }
+    }
   }
 };
 

browser/components/sessionstore/content/content-sessionStore.js

@@ -29,10 +29,6 @@ XPCOMUtils.defineLazyModuleGetter(this, "SessionHistory",
 XPCOMUtils.defineLazyModuleGetter(this, "SessionStorage",
   "resource:///modules/sessionstore/SessionStorage.jsm");
 
-XPCOMUtils.defineLazyServiceGetter(this, "cpmm",
-                                   "@mozilla.org/childprocessmessagemanager;1",
-                                   "nsISyncMessageSender");
-
 Cu.import("resource:///modules/sessionstore/FrameTree.jsm", this);
 var gFrameTree = new FrameTree(this);
 
@@ -692,12 +688,21 @@ var MessageQueue = {
       FX_SESSION_RESTORE_CONTENT_COLLECT_DATA_LONGEST_OP_MS: durationMs
     }
 
-    // Send all data to the parent process.
-    sendMessage("SessionStore:update", {
-      id: this._id, data, telemetry, flushID,
-      isFinal: options.isFinal || false,
-      epoch: gCurrentEpoch
-    });
+    try {
+      // Send all data to the parent process.
+      sendMessage("SessionStore:update", {
+        id: this._id, data, telemetry, flushID,
+        isFinal: options.isFinal || false,
+        epoch: gCurrentEpoch
+      });
+    } catch (ex if ex && ex.result == Cr.NS_ERROR_OUT_OF_MEMORY) {
+      let telemetry = {
+        FX_SESSION_RESTORE_SEND_UPDATE_CAUSED_OOM: 1
+      };
+      sendMessage("SessionStore:error", {
+        telemetry
+      });
+    }
 
     // Increase our unique message ID.
     this._id++;

browser/components/sessionstore/test/browser.ini

@@ -211,3 +211,4 @@ skip-if = true
 skip-if = os == "mac"
 
 [browser_911547.js]
+[browser_send_async_message_oom.js]

browser/components/sessionstore/test/browser_send_async_message_oom.js

@@ -0,0 +1,71 @@
+/* Any copyright is dedicated to the Public Domain.
+   http://creativecommons.org/publicdomain/zero/1.0/ */
+
+const {Services} = Cu.import("resource://gre/modules/Services.jsm", {});
+
+const HISTOGRAM_NAME = "FX_SESSION_RESTORE_SEND_UPDATE_CAUSED_OOM";
+
+/**
+ * Test that an OOM in sendAsyncMessage in a framescript will be reported
+ * to Telemetry.
+ */
+
+add_task(function* init() {
+  Services.telemetry.canRecordExtended = true;
+});
+
+function frameScript() {
+  // Make send[A]syncMessage("SessionStore:update", ...) simulate OOM.
+  // Other operations are unaffected.
+  let mm = docShell.sameTypeRootTreeItem.
+    QueryInterface(Ci.nsIDocShell).
+    QueryInterface(Ci.nsIInterfaceRequestor).
+    getInterface(Ci.nsIContentFrameMessageManager);
+
+  let wrap = function(original) {
+    return function(name, ...args) {
+      if (name != "SessionStore:update") {
+        return original(name, ...args);
+      }
+      throw new Components.Exception("Simulated OOM", Cr.NS_ERROR_OUT_OF_MEMORY);
+    }
+  }
+
+  mm.sendAsyncMessage = wrap(mm.sendAsyncMessage);
+  mm.sendSyncMessage = wrap(mm.sendSyncMessage);
+}
+
+add_task(function*() {
+  // Capture original state.
+  let snapshot = Services.telemetry.getHistogramById(HISTOGRAM_NAME).snapshot();
+
+  // Open a browser, configure it to cause OOM.
+  let newTab = gBrowser.addTab("about:robots");
+  let browser = newTab.linkedBrowser;
+  yield ContentTask.spawn(browser, null, frameScript);
+
+
+  let promiseReported = new Promise(resolve => {
+    browser.messageManager.addMessageListener("SessionStore:error", resolve);
+  });
+
+  // Attempt to flush. This should fail.
+  let promiseFlushed = TabStateFlusher.flush(browser);
+  promiseFlushed.then(() => {throw new Error("Flush should have failed")});
+
+  // The frame script should report an error.
+  yield promiseReported;
+
+  // Give us some time to handle that error.
+  yield new Promise(resolve => setTimeout(resolve, 10));
+
+  // By now, Telemetry should have been updated.
+  let snapshot2 = Services.telemetry.getHistogramById(HISTOGRAM_NAME).snapshot();
+  gBrowser.removeTab(newTab);
+
+  Assert.ok(snapshot2.sum > snapshot.sum);
+});
+
+add_task(function* cleanup() {
+  Services.telemetry.canRecordExtended = false;
+});

toolkit/components/telemetry/Histograms.json

@@ -4352,6 +4352,12 @@
     "extended_statistics_ok": true,
     "description": "Session restore: Time spent blocking the main thread while restoring a window state (ms)"
   },
+  "FX_SESSION_RESTORE_SEND_UPDATE_CAUSED_OOM": {
+    "alert_emails": ["session-restore-telemetry-alerts@mozilla.com"],
+    "expires_in_version": "default",
+    "kind": "count",
+    "description": "Count of messages sent by SessionRestore from child frames to the parent and that cannot be transmitted as they eat up too much memory."
+  },
   "FX_TABLETMODE_PAGE_LOAD": {
     "expires_in_version": "47",
     "kind": "exponential",