hno squid-2.2.DEVEL3.icp_dead_detection.patch:

Squid-2.2.DEVEL3: Improved ICP dead peer detection

Some minor adjustments of ICP dead peer detection to make it behave well
on startup and low load servers.

diff --git a/ChangeLog b/ChangeLog
index d3bc328743eab4d78a15d0495c73d6ee209fabbc..b0790bd704fdb23847cc854229b0ef31816a3ed6 100644 (file)
--- a/ChangeLog
+++ b/ChangeLog
@@ -5,6 +5,8 @@ Changes to Squid-2.4.DEVEL3 ():
        - Added DISKD stats via cachemgr.
        - Added squid.conf options for DISKD magic constants.
        - Added --enable-auth-modules=... configure option
+       - Improved ICP dead peer detection to also work when the workload
+         is low
 
 Changes to Squid-2.4.DEVEL2 ():
 

diff --git a/src/neighbors.cc b/src/neighbors.cc
index 297bf0747193fa7b8e5bf0cf12d8bc7830f48d09..9a7e9fda5ae9b91b1afca89510abf107d5347576 100644 (file)
--- a/src/neighbors.cc
+++ b/src/neighbors.cc
@@ -1,6 +1,6 @@
 
 /*
- * $Id: neighbors.cc,v 1.278 2000/03/06 16:23:33 wessels Exp $
+ * $Id: neighbors.cc,v 1.279 2000/05/02 18:23:48 hno Exp $
  *
  * DEBUG: section 15    Neighbor Routines
  * AUTHOR: Harvest Derived
@@ -163,15 +163,17 @@ peerWouldBePinged(const peer * p, request_t * request)
        return 0;
     if (p->options.mcast_responder)
        return 0;
+    if (p->n_addresses == 0)
+       return 0;
     /* the case below seems strange, but can happen if the
      * URL host is on the other side of a firewall */
     if (p->type == PEER_SIBLING)
        if (!request->flags.hierarchical)
            return 0;
-    if (p->icp.port == echo_port)
-       if (!neighborUp(p))
-           return 0;
-    if (p->n_addresses == 0)
+    /* Ping dead peers every timeout interval */
+    if (squid_curtime - p->stats.last_query > Config.Timeout.deadPeer)
+       return 1;
+    if (!neighborUp(p))
        return 0;
     return 1;
 }
@@ -434,24 +436,6 @@ neighborsUdpPing(request_t * request,
             */
            p->stats.last_reply = squid_curtime;
            (*exprep) += p->mcast.n_replies_expected;
-       } else if (squid_curtime - p->stats.last_query > Config.Timeout.deadPeer) {
-           /*
-            * fake a recent reply if its been a long time since our
-            * last query
-            */
-           p->stats.last_reply = squid_curtime;
-           /*
-            * We used to not expect a reply in this case; we assumed
-            * the peer was DEAD if we hadn't queried it in a long
-            * time.  However, the number of people whining to
-            * squid-users that ICP is broken became unbearable.  They
-            * tried a single request which, to their amazement, was
-            * forwarded directly to the origin server, even thought
-            * they KNEW it was in a neighbor cache.  Ok, I give up, you
-            * win!
-            */
-           (*exprep)++;
-           (*timeout) += 1000;
        } else if (neighborUp(p)) {
            /* its alive, expect a reply from it */
            (*exprep)++;
@@ -467,6 +451,8 @@ neighborsUdpPing(request_t * request,
            }
        }
        p->stats.last_query = squid_curtime;
+       if (p->stats.probe_start == 0)
+           p->stats.probe_start = squid_curtime;
     }
     if ((first_ping = first_ping->next) == NULL)
        first_ping = Config.peers;
@@ -633,6 +619,7 @@ neighborAlive(peer * p, const MemObject * mem, const icp_common_t * header)
        p->stats.logged_state = PEER_ALIVE;
     }
     p->stats.last_reply = squid_curtime;
+    p->stats.probe_start = 0;
     p->stats.pings_acked++;
     if ((icp_opcode) header->opcode <= ICP_END)
        p->icp.counts[header->opcode]++;
@@ -665,6 +652,7 @@ neighborAliveHtcp(peer * p, const MemObject * mem, const htcpReplyData * htcp)
        p->stats.logged_state = PEER_ALIVE;
     }
     p->stats.last_reply = squid_curtime;
+    p->stats.probe_start = 0;
     p->stats.pings_acked++;
     p->htcp.counts[htcp->hit ? 1 : 0]++;
     p->htcp.version = htcp->version;
@@ -880,9 +868,10 @@ neighborUp(const peer * p)
 {
     if (!p->tcp_up)
        return 0;
-    if (squid_curtime - p->stats.last_query > Config.Timeout.deadPeer)
+    if (p->options.no_query)
        return 1;
-    if (p->stats.last_query - p->stats.last_reply > Config.Timeout.deadPeer)
+    if (p->stats.probe_start != 0 &&
+           squid_curtime - p->stats.probe_start > Config.Timeout.deadPeer)
        return 0;
     return 1;
 }

diff --git a/src/structs.h b/src/structs.h
index ecad3f643f15658d5662ad9a0e89a5895a1fdbfc..3bb006bce54af26c9893dadb23b61079ae281438 100644 (file)
--- a/src/structs.h
+++ b/src/structs.h
@@ -1,6 +1,6 @@
 
 /*
- * $Id: structs.h,v 1.316 2000/03/14 23:07:51 wessels Exp $
+ * $Id: structs.h,v 1.317 2000/05/02 18:23:49 hno Exp $
  *
  *
  * SQUID Internet Object Cache  http://squid.nlanr.net/Squid/
@@ -1054,6 +1054,7 @@ struct _peer {
        int ignored_replies;
        int n_keepalives_sent;
        int n_keepalives_recv;
+       time_t probe_start;
        time_t last_query;
        time_t last_reply;
        int logged_state;       /* so we can print dead/revived msgs */