Try to mitigate DNS failures when downloading stuff asynchronously

* url-queue.el (url-queue-setup-runners): New function that uses
`run-with-idle-timer' for extra asynchronicity.
(url-queue-remove-jobs-from-host): New function.
(url-queue-callback-function): Remove jobs from the same host if
connection failed.

diff --git a/lisp/url/ChangeLog b/lisp/url/ChangeLog
index 7c92fc3..9285961 100644 (file)
--- a/lisp/url/ChangeLog
+++ b/lisp/url/ChangeLog
@@ -1,3 +1,11 @@
+2012-02-06  Lars Ingebrigtsen  <larsi@gnus.org>
+
+       * url-queue.el (url-queue-setup-runners): New function that uses
+       `run-with-idle-timer' for extra asynchronicity.
+       (url-queue-remove-jobs-from-host): New function.
+       (url-queue-callback-function): Remove jobs from the same host if
+       connection failed.
+
 2012-01-12  Glenn Morris  <rgm@gnu.org>
 
        * url-auth.el (url-basic-auth, url-digest-auth):

diff --git a/lisp/url/url-queue.el b/lisp/url/url-queue.el
index 534c94b..976a266 100644 (file)
--- a/lisp/url/url-queue.el
+++ b/lisp/url/url-queue.el
@@ -30,6 +30,7 @@
 
 (eval-when-compile (require 'cl))
 (require 'browse-url)
+(require 'url-parse)
 
 (defcustom url-queue-parallel-processes 6
   "The number of concurrent processes."
@@ -49,7 +50,7 @@
 
 (defstruct url-queue
   url callback cbargs silentp
-  buffer start-time)
+  buffer start-time pre-triggered)
 
 ;;;###autoload
 (defun url-queue-retrieve (url callback &optional cbargs silent)
@@ -63,7 +64,30 @@ controls the level of parallelism via the
                                      :callback callback
                                      :cbargs cbargs
                                      :silentp silent))))
-  (url-queue-run-queue))
+  (url-queue-setup-runners))
+
+;; To ensure asynch behaviour, we start the required number of queue
+;; runners from `run-with-idle-timer'.  So we're basically going
+;; through the queue in two ways: 1) synchronously when a program
+;; calls `url-queue-retrieve' (which will then start the required
+;; number of queue runners), and 2) at the exit of each job, which
+;; will then not start any further threads, but just reuse the
+;; previous "slot".
+
+(defun url-queue-setup-runners ()
+  (let ((running 0)
+       waiting)
+    (dolist (entry url-queue)
+      (cond
+       ((or (url-queue-start-time entry)
+           (url-queue-pre-triggered entry))
+       (incf running))
+       ((not waiting)
+       (setq waiting entry))))
+    (when (and waiting
+              (< running url-queue-parallel-processes))
+      (setf (url-queue-pre-triggered waiting) t)
+      (run-with-idle-timer 0.01 nil 'url-queue-run-queue))))
 
 (defun url-queue-run-queue ()
   (url-queue-prune-old-entries)
@@ -81,10 +105,27 @@ controls the level of parallelism via the
       (url-queue-start-retrieve waiting))))
 
 (defun url-queue-callback-function (status job)
+  (when (and (eq (car status) :error)
+            (eq (cadr (cadr status)) 'connection-failed))
+    ;; If we get a connection error, then flush all other jobs from
+    ;; the host from the queue.  This particularly makes sense if the
+    ;; error really is a DNS resolver issue, which happens
+    ;; synchronously and totally halts Emacs.
+    (url-queue-remove-jobs-from-host
+     (plist-get (nthcdr 3 (cadr status)) :host)))
   (setq url-queue (delq job url-queue))
   (url-queue-run-queue)
   (apply (url-queue-callback job) (cons status (url-queue-cbargs job))))
 
+(defun url-queue-remove-jobs-from-host (host)
+  (let ((jobs nil))
+    (dolist (job url-queue)
+      (when (equal (url-host (url-generic-parse-url (url-queue-url job)))
+                  host)
+       (push job jobs)))
+    (dolist (job jobs)
+      (setq url-queue (delq job url-queue)))))
+  
 (defun url-queue-start-retrieve (job)
   (setf (url-queue-buffer job)
        (ignore-errors