Implement exponential back-off mechanism and enforce it at the URLRequestHttpJob level.

chrome/service/cloud_print/cloud_print_url_fetcher.h

Index: chrome/service/cloud_print/cloud_print_url_fetcher.h
===================================================================
--- chrome/service/cloud_print/cloud_print_url_fetcher.h	(revision 66188)
+++ chrome/service/cloud_print/cloud_print_url_fetcher.h	(working copy)
@@ -12,16 +12,13 @@
 
 class DictionaryValue;
 class GURL;
-class URLFetcherProtectEntry;
 class URLRequestStatus;
 
 // A wrapper around URLFetcher for CloudPrint. URLFetcher applies retry logic
 // only on HTTP response codes >= 500. In the cloud print case, we want to
 // retry on all network errors. In addition, we want to treat non-JSON responses
 // (for all CloudPrint APIs that expect JSON responses) as errors and they
-// must also be retried. Also URLFetcher uses the host name of the URL as the
-// key for applying the retry policy. In our case, we want to apply one global
-// policy for many requests (not necessarily scoped by hostname).
+// must also be retried.
 class CloudPrintURLFetcher
     : public base::RefCountedThreadSafe<CloudPrintURLFetcher>,

[sanjeevr, 2010/11/22 19:09:26]

Now that we don't post tasks to this object, perhaps it does not need to
implement RefCountedThreadSafe any longer.

[yzshen, 2010/11/22 22:59:17]

I saw the following code in CloudPrintURLFetcher::OnURLFetchComplete:

scoped_refptr<CloudPrintURLFetcher> keep_alive(this);

I guess declaring this class as ref-counted serves two purposes:
1) we want to support posted tasks;
2) by doing this, the delegate can abandon its reference to the fetcher object
within the delegate methods:

ResponseAction SomeDelegate::HandleRawData(...) {
  ...
  // request_ is scoped_refptr<CloudPrintURLFetche>,
  // the following operation is safe.
  request_ = NULL;
}

However, if we remove the base class RefCountedThreadSafe, we cannot do the same
thing anymore:

ResponseAction SomeDelegate::HandleRawData(...) {
  ...
  // request_ is scoped_ptr<CloudPrintURLFetche>,
  // the following operation will crash.
  request_.reset(NULL);
}

Since I am not very familiar with the context and this part of codebase, I would
like to leave it as it is if possible.

       public URLFetcher::Delegate {
@@ -79,11 +76,11 @@
   void StartGetRequest(const GURL& url,
                        Delegate* delegate,
                        const std::string& auth_token,
-                       const std::string& retry_policy);
+                       int max_retries);
   void StartPostRequest(const GURL& url,
                         Delegate* delegate,
                         const std::string& auth_token,
-                        const std::string& retry_policy,
+                        int max_retries,
                         const std::string& post_data_mime_type,
                         const std::string& post_data);
 
@@ -102,14 +99,13 @@
                           URLFetcher::RequestType request_type,
                           Delegate* delegate,
                           const std::string& auth_token,
-                          const std::string& retry_policy,
+                          int max_retries,
                           const std::string& post_data_mime_type,
                           const std::string& post_data);
   void StartRequestNow();
 
   scoped_ptr<URLFetcher> request_;
   Delegate* delegate_;
-  URLFetcherProtectEntry* protect_entry_;
   int num_retries_;
 };