Mailing List Archive

Author: phk
Date: 2006-09-15 10:48:13 +0200 (Fri, 15 Sep 2006)
New Revision: 986

Modified:
trunk/varnish-cache/bin/varnishd/cache_session.c
Log:
Add XXX comment


Modified: trunk/varnish-cache/bin/varnishd/cache_session.c
===================================================================
--- trunk/varnish-cache/bin/varnishd/cache_session.c 2006-09-15 08:43:39 UTC (rev 985)
+++ trunk/varnish-cache/bin/varnishd/cache_session.c 2006-09-15 08:48:13 UTC (rev 986)
@@ -10,6 +10,15 @@
*
* We identify srcaddrs instead of full addr+port because the desired level
* of granularity is "whois is abuse@ or tech-c@ in the RIPE database.
+ *
+ * XXX: The two-list session management is actually not a good idea
+ * XXX: come to think of it, because we want the sessions reused in
+ * XXX: Most Recently Used order.
+ * XXX: Another and maybe more interesting option would be to cache
+ * XXX: free sessions in the worker threads and postpone session
+ * XXX: allocation until then. This does not quite implment MRU order
+ * XXX: but it does save some locking, although not that much because
+ * XXX: we still have to do the source-addr lookup.
*/

#include <stdlib.h>