PlzNavigate: implement process reuse for ServiceWorkers

content/browser/site_instance_impl.h

 // Copyright (c) 2012 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.
 
 #ifndef CONTENT_BROWSER_SITE_INSTANCE_IMPL_H_
 #define CONTENT_BROWSER_SITE_INSTANCE_IMPL_H_
 
 #include <stddef.h>
 #include <stdint.h>
 
@@ skipping 46 matching lines @@
   // process-per-site and ServiceWorkers proactively join a process that already
   // contains the site.
   enum class ProcessReusePolicy {
     // In this mode, all instances of the site will be hosted in the same
     // RenderProcessHost.
     PROCESS_PER_SITE,
 
     // In this mode, subframes will be hosted in a designated RenderProcessHost.
     USE_DEFAULT_SUBFRAME_PROCESS,
 
+    // In this mode, the site will be rendered in a suitable RenderProcessHost
+    // in priority, and a new RenderProcessHost will be created if none exist. A
+    // suitable RenderProcessHost is one that has a frame rendering the site, or
+    // that is expecting a navigation to the site.
+    REUSE_CHECKING_FRAMES_AND_NAVIGATIONS,

[Charlie Reis, 2017/05/15 03:41:52]

I'll suggest REUSE_PENDING_OR_COMMITTED_SITE.  That way "site" is in the name
(as the criteria for reuse), and pending vs committed seems like a more direct
comparison than "frames" vs "navigations" (which are independent concepts).

The comment can be slightly rephrased to convey this (and since "in priority"
wasn't explained):

In this mode, the site will be rendered in a RenderProcessHost that is already
in use for the site, either for a pending navigation or a committed navigation. 
If none exists, a new process will be created.  If multiple such processes
exist, ones that have foreground frames are given priority, and otherwise one is
selected randomly.

(As a side note, I think we might actually be able to use this exact behavior
for PROCESS_PER_SITE.  The main difference is using an existing process if a
navigation is pending, and I think that may actually fix a race in
process-per-site rather than cause problems.  That's for a followup CL, though,
and we should think it through.)

[clamy, 2017/05/16 14:50:46]

Done.

+
     // By default, a new RenderProcessHost will be created unless the process
     // limit has been reached. The RenderProcessHost reused will be chosen
     // randomly and not based on the site.
     DEFAULT,
   };
 
   void set_process_reuse_policy(ProcessReusePolicy policy) {
     process_reuse_policy_ = policy;
   }
   ProcessReusePolicy process_reuse_policy() const {
@@ skipping 123 matching lines @@
   ProcessReusePolicy process_reuse_policy_;
 
   base::ObserverList<Observer, true> observers_;
 
   DISALLOW_COPY_AND_ASSIGN(SiteInstanceImpl);
 };
 
 }  // namespace content
 
 #endif  // CONTENT_BROWSER_SITE_INSTANCE_IMPL_H_