From 812d80e4035afe3bfc51fcb9ed76bede117a1c78 Mon Sep 17 00:00:00 2001
From: Johan Wikman <johan.wikman@mariadb.com>
Date: Mon, 3 Sep 2018 11:03:48 +0300
Subject: [PATCH] MXS-2011 Use alternative servers

If the replication fails using the current config, we retry
immediately using another config, without waiting anything at
all.

Only when we have unsuccessfully tried with all servers, will
we wait a while before starting again.
---
 server/modules/routing/binlogrouter/blr.cc    |  1 +
 server/modules/routing/binlogrouter/blr.hh    |  3 ++-
 .../routing/binlogrouter/blr_master.cc        | 26 +++++++++++++++++++
 3 files changed, 29 insertions(+), 1 deletion(-)

diff --git a/server/modules/routing/binlogrouter/blr.cc b/server/modules/routing/binlogrouter/blr.cc
index 47f461950..0bff46eba 100644
--- a/server/modules/routing/binlogrouter/blr.cc
+++ b/server/modules/routing/binlogrouter/blr.cc
@@ -817,6 +817,7 @@ static MXS_ROUTER* createInstance(SERVICE *service, MXS_CONFIG_PARAMETER* params
      * automatic master replication start phase
      */
 
+    inst->current_config = 0;
     rc = blr_file_read_master_config(inst);
 
     /*
diff --git a/server/modules/routing/binlogrouter/blr.hh b/server/modules/routing/binlogrouter/blr.hh
index c31ac5496..81cebb709 100644
--- a/server/modules/routing/binlogrouter/blr.hh
+++ b/server/modules/routing/binlogrouter/blr.hh
@@ -821,7 +821,8 @@ struct ROUTER_INSTANCE: public MXS_ROUTER
     enum binlog_storage_type   storage_type;/*< Enables hierachical binlog file storage */
     char              *set_slave_hostname;  /*< Send custom Hostname to Master */
     ROUTER_INSTANCE  *next;
-    std::vector<ChangeMasterConfig> configs;              /*< Current config. */
+    std::vector<ChangeMasterConfig> configs;/*< Available configs. */
+    int               current_config;       /*< The config currently being used. */
 };
 
 /** Master Semi-Sync capability */
diff --git a/server/modules/routing/binlogrouter/blr_master.cc b/server/modules/routing/binlogrouter/blr_master.cc
index f2baef686..d062533cc 100644
--- a/server/modules/routing/binlogrouter/blr_master.cc
+++ b/server/modules/routing/binlogrouter/blr_master.cc
@@ -377,6 +377,23 @@ blr_restart_master(ROUTER_INSTANCE *router)
         /* Force unconnected state */
         router->master_state = BLRM_UNCONNECTED;
         router->retry_count++;
+
+        int current_config = (router->current_config + 1) % router->configs.size();
+        if (current_config != router->current_config) // Will be different unless there is but one.
+        {
+            mxb_assert(current_config < static_cast<int>(router->configs.size()));
+
+            const ChangeMasterConfig& old_config = router->configs[router->current_config];
+            router->current_config = current_config;
+            const ChangeMasterConfig& new_config = router->configs[router->current_config];
+
+            blr_master_set_config(router, new_config);
+
+            MXS_NOTICE("Connection to %s:%d failed, now trying with %s:%d.",
+                       old_config.host.c_str(), old_config.port,
+                       new_config.host.c_str(), new_config.port);
+        }
+
         spinlock_release(&router->lock);
 
         blr_start_master_in_main(router, connect_retry);
@@ -3094,6 +3111,15 @@ static int blr_check_connect_retry(ROUTER_INSTANCE *router)
         return -1;
     }
 
+    mxb_assert(router->configs.size() > 0);
+    if (router->current_config < static_cast<int>(router->configs.size() - 1))
+    {
+        // We have unused configs; no need to sleep anything at all. We will
+        // sleep only when we have unsuccessfully cycled through all available
+        // servers.
+        return 0;
+    }
+
     /* Return the interval for next reconnect */
     if (router->retry_count >= router->retry_interval / BLR_MASTER_BACKOFF_TIME)
     {