simplify election handling

make sure we read and update the flags from all remote nodes before we
reach the first codepath that can call do_recovery()
since during do_recovery() we need to know what the flags are.

(This used to be ctdb commit e85f3806483ea420559d449e0e4d81bec996740f)

diff --git a/ctdb/server/ctdb_recoverd.c b/ctdb/server/ctdb_recoverd.c
index f47ec94ad8c8695811b49c178d5e8ffadca7715e..eb87b128225253db002f189ad08960a53ee26e00 100644 (file)
--- a/ctdb/server/ctdb_recoverd.c
+++ b/ctdb/server/ctdb_recoverd.c
@@ -1017,26 +1017,14 @@ static bool ctdb_election_win(struct ctdb_recoverd *rec, struct election_message
 
        ctdb_election_data(rec, &myem);
 
-       /* try to use a unbanned node */
-       if ((em->node_flags & NODE_FLAGS_BANNED) &&
-           !(myem.node_flags & NODE_FLAGS_BANNED)) {
-               cmp = 1;
-       }
-       if (!(em->node_flags & NODE_FLAGS_BANNED) &&
-           (myem.node_flags & NODE_FLAGS_BANNED)) {
-               cmp = -1;
-       }
+       /* we cant win if we are banned */
+       if (rec->node_flags & NODE_FLAGS_BANNED) {
+               return 0;
+       }       
 
-       /* try to use a healthy node */
-       if (cmp == 0) {
-               if ((em->node_flags & NODE_FLAGS_UNHEALTHY) &&
-                   !(myem.node_flags & NODE_FLAGS_UNHEALTHY)) {
-                       cmp = 1;
-               }
-               if (!(em->node_flags & NODE_FLAGS_UNHEALTHY) &&
-                   (myem.node_flags & NODE_FLAGS_UNHEALTHY)) {
-                       cmp = -1;
-               }
+       /* we will automatically win if the other node is banned */
+       if (em->node_flags & NODE_FLAGS_BANNED) {
+               return 1;
        }
 
        /* try to use the most connected node */
@@ -1066,7 +1054,7 @@ static int send_election_request(struct ctdb_recoverd *rec, TALLOC_CTX *mem_ctx,
        struct election_message emsg;
        uint64_t srvid;
        struct ctdb_context *ctdb = rec->ctdb;
-       
+
        srvid = CTDB_SRVID_RECOVERY;
 
        ctdb_election_data(rec, &emsg);
@@ -1580,7 +1568,17 @@ again:
                goto again;
        }
 
-       if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
+       /* grap the nodemap from the recovery master */
+       ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn, 
+                                  mem_ctx, &remote_nodemap);
+       if (ret != 0) {
+               DEBUG(0, (__location__ " Unable to get nodemap from recovery master %u\n", 
+                         nodemap->nodes[j].pnn));
+               goto again;
+       }
+
+
+       if (remote_nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
                DEBUG(0, ("Recmaster node %u no longer available. Force reelection\n", nodemap->nodes[j].pnn));
                force_election(rec, mem_ctx, pnn, nodemap);
                goto again;
@@ -1639,6 +1637,37 @@ again:
                goto again;
        }
 
+
+       /* we are recovery master, go through the list of all connected nodes
+          and get the nodeflags from them and update our copy of nodeflags
+        */
+       for (j=0; j<nodemap->num; j++) {
+               if (nodemap->nodes[j].flags & NODE_FLAGS_DISCONNECTED) {
+                       continue;
+               }
+               if (nodemap->nodes[j].pnn == pnn) {
+                       continue;
+               }
+
+               ret = ctdb_ctrl_getnodemap(ctdb, CONTROL_TIMEOUT(), nodemap->nodes[j].pnn, 
+                                          mem_ctx, &remote_nodemap);
+               if (ret != 0) {
+                       DEBUG(0, (__location__ " Unable to get nodemap from remote node %u\n", 
+                                 nodemap->nodes[j].pnn));
+                       goto again;
+               }
+
+               /* update our nodemap flags according to the other
+                  server - this gets the NODE_FLAGS_DISABLED
+                  flag. Note that the remote node is authoritative
+                  for its flags (except CONNECTED, which we know
+                  matches in this code) */
+               if (nodemap->nodes[j].flags != remote_nodemap->nodes[j].flags) {
+                       nodemap->nodes[j].flags = remote_nodemap->nodes[j].flags;
+                       rec->need_takeover_run = true;
+               }
+       }
+
        /* update the list of public ips that a node can handle for
           all connected nodes
        */
@@ -1767,15 +1796,6 @@ again:
                        }
                }
 
-               /* update our nodemap flags according to the other
-                  server - this gets the NODE_FLAGS_DISABLED
-                  flag. Note that the remote node is authoritative
-                  for its flags (except CONNECTED, which we know
-                  matches in this code) */
-               if (nodemap->nodes[j].flags != remote_nodemap->nodes[j].flags) {
-                       nodemap->nodes[j].flags = remote_nodemap->nodes[j].flags;
-                       rec->need_takeover_run = true;
-               }
        }