replication: do not ignore replication_connect_quorum.

On bootstrap and after replication reconfiguration
replication_connect_quorum was ignored. The instance tried to connect to
every replica listed in replication parameter, and errored if it wasn't
possible.
The patch alters this behaviour. The instance still tries to connect to
every node listed in replication, but does not raise an error if it was
able to connect to at least replication_connect_quorum instances.

Closes #3428

@TarantoolBot document
Title: replication_connect_quorum is not ignored.
Now on replica set bootstrap instance tries to connect
not to every replica, listed in box.cfg.replication, but
only to replication_connect_quorum replicas. If after
replication_connect_timeout seconds instance is not connected to
at least replication_connect_quorum other instances, we
throw an error.

src/box/box.cc

@@ -595,7 +595,7 @@ cfg_get_replication(int *p_count)
  * don't start appliers.
  */
 static void
-box_sync_replication(double timeout, bool connect_all)
+box_sync_replication(bool connect_quorum)
 {
 	int count = 0;
 	struct applier **appliers = cfg_get_replication(&count);
@@ -607,7 +607,7 @@ box_sync_replication(double timeout, bool connect_all)
 			applier_delete(appliers[i]); /* doesn't affect diag */
 	});
 
-	replicaset_connect(appliers, count, timeout, connect_all);
+	replicaset_connect(appliers, count, connect_quorum);
 
 	guard.is_active = false;
 }
@@ -626,7 +626,7 @@ box_set_replication(void)
 
 	box_check_replication();
 	/* Try to connect to all replicas within the timeout period */
-	box_sync_replication(replication_connect_timeout, true);
+	box_sync_replication(true);
 	/* Follow replica */
 	replicaset_follow();
 }
@@ -1866,7 +1866,7 @@ box_cfg_xc(void)
 		title("orphan");
 
 		/* Wait for the cluster to start up */
-		box_sync_replication(replication_connect_timeout, false);
+		box_sync_replication(false);
 	} else {
 		if (!tt_uuid_is_nil(&instance_uuid))
 			INSTANCE_UUID = instance_uuid;
@@ -1888,7 +1888,7 @@ box_cfg_xc(void)
 		 * receive the same replica set UUID when a new cluster
 		 * is deployed.
 		 */
-		box_sync_replication(TIMEOUT_INFINITY, true);
+		box_sync_replication(true);
 		/* Bootstrap a new master */
 		bootstrap(&replicaset_uuid, &is_bootstrap_leader);
 	}

src/box/replication.cc

@@ -46,7 +46,7 @@ struct tt_uuid INSTANCE_UUID;
 struct tt_uuid REPLICASET_UUID;
 
 double replication_timeout = 1.0; /* seconds */
-double replication_connect_timeout = 4.0; /* seconds */
+double replication_connect_timeout = 30.0; /* seconds */
 int replication_connect_quorum = REPLICATION_CONNECT_QUORUM_ALL;
 double replication_sync_lag = 10.0; /* seconds */
 
@@ -540,7 +540,7 @@ applier_on_connect_f(struct trigger *trigger, void *event)
 
 void
 replicaset_connect(struct applier **appliers, int count,
-		   double timeout, bool connect_all)
+		   bool connect_quorum)
 {
 	if (count == 0) {
 		/* Cleanup the replica set. */
@@ -557,7 +557,7 @@ replicaset_connect(struct applier **appliers, int count,
 	 * - register a trigger in each applier to wake up our
 	 *   fiber via this channel when the remote peer becomes
 	 *   connected and a UUID is received;
-	 * - wait up to CONNECT_TIMEOUT seconds for `count` messages;
+	 * - wait up to REPLICATION_CONNECT_TIMEOUT seconds for `count` messages;
 	 * - on timeout, raise a CFG error, cancel and destroy
 	 *   the freshly created appliers (done in a guard);
 	 * - an success, unregister the trigger, check the UUID set
@@ -571,6 +571,8 @@ replicaset_connect(struct applier **appliers, int count,
 	state.connected = state.failed = 0;
 	fiber_cond_create(&state.wakeup);
 
+	double timeout = replication_connect_timeout;
+
 	/* Add triggers and start simulations connection to remote peers */
 	for (int i = 0; i < count; i++) {
 		struct applier *applier = appliers[i];
@@ -587,15 +589,14 @@ replicaset_connect(struct applier **appliers, int count,
 		double wait_start = ev_monotonic_now(loop());
 		if (fiber_cond_wait_timeout(&state.wakeup, timeout) != 0)
 			break;
-		if (state.failed > 0 && connect_all)
-			break;
 		timeout -= ev_monotonic_now(loop()) - wait_start;
 	}
 	if (state.connected < count) {
 		say_crit("failed to connect to %d out of %d replicas",
 			 count - state.connected, count);
 		/* Timeout or connection failure. */
-		if (connect_all)
+		if (connect_quorum && state.connected <
+		    MIN(count, replication_connect_quorum))
 			goto error;
 	} else {
 		say_verbose("connected to %d replicas", state.connected);

src/box/replication.h

@@ -356,13 +356,13 @@ replicaset_add(uint32_t replica_id, const struct tt_uuid *instance_uuid);
  *
  * \param appliers the array of appliers
  * \param count size of appliers array
- * \param timeout connection timeout
- * \param connect_all if this flag is set, fail unless all
- *                    appliers have successfully connected
+ * \param connect_quorum if this flag is set, fail unless at
+ *		       least replication_connect_quorum
+ *		       appliers have successfully connected.
  */
 void
 replicaset_connect(struct applier **appliers, int count,
-		   double timeout, bool connect_all);
+		   bool connect_quorum);
 
 /**
  * Resume all appliers registered with the replica set.

test/replication-py/init_storage.test.py

@@ -57,7 +57,7 @@
 
 server.stop()
 replica = TarantoolServer(server.ini)
-replica.script = 'replication/replica.lua'
+replica.script = 'replication-py/replica.lua'
 replica.vardir = server.vardir #os.path.join(server.vardir, 'replica')
 replica.rpl_master = master
 replica.deploy(wait=False)

test/replication-py/master.lua

@@ -3,6 +3,8 @@ os = require('os')
 box.cfg({
     listen              = os.getenv("LISTEN"),
     memtx_memory        = 107374182,
+    replication_connect_timeout = 1.0,
+    replication_timeout = 0.3
 })
 
 require('console').listen(os.getenv('ADMIN'))

test/replication-py/replica.lua

@@ -7,6 +7,8 @@ box.cfg({
     listen              = os.getenv("LISTEN"),
     replication         = os.getenv("MASTER"),
     memtx_memory        = 107374182,
+    replication_connect_timeout = 1.0,
+    replication_timeout = 0.3
 })
 
 box_cfg_done = true

test/replication/autobootstrap.lua

@@ -21,7 +21,8 @@ box.cfg({
         USER..':'..PASSWORD..'@'..instance_uri(2);
         USER..':'..PASSWORD..'@'..instance_uri(3);
     };
-    replication_connect_timeout = 0.5,
+    replication_connect_timeout = 3.0;
+    replication_timeout = 0.5;
 })
 
 box.once("bootstrap", function()

test/replication/autobootstrap_guest.lua

@@ -20,7 +20,7 @@ box.cfg({
         instance_uri(2);
         instance_uri(3);
     };
-    replication_connect_timeout = 0.5,
+    replication_connect_timeout = 5,
 })
 
 box.once("bootstrap", function()

test/replication/ddl.lua

@@ -22,7 +22,8 @@ box.cfg({
         USER..':'..PASSWORD..'@'..instance_uri(3);
         USER..':'..PASSWORD..'@'..instance_uri(4);
     };
-    replication_connect_timeout = 0.5,
+    replication_timeout = 0.1,
+    replication_connect_timeout = 2.0,
 })
 
 box.once("bootstrap", function()

test/replication/errinj.result

@@ -418,7 +418,7 @@ test_run:cmd("create server replica_timeout with rpl_master=default, script='rep
 ---
 - true
 ...
-test_run:cmd("start server replica_timeout with args='0.01'")
+test_run:cmd("start server replica_timeout with args='0.1, 0.5'")
 ---
 - true
 ...
@@ -474,7 +474,7 @@ errinj.set("ERRINJ_RELAY_REPORT_INTERVAL", 0)
 ...
 -- Check replica's ACKs don't prevent the master from sending
 -- heartbeat messages (gh-3160).
-test_run:cmd("start server replica_timeout with args='0.009'")
+test_run:cmd("start server replica_timeout with args='0.009, 0.5'")
 ---
 - true
 ...
@@ -522,7 +522,7 @@ for i = 0, 9999 do box.space.test:replace({i, 4, 5, 'test'}) end
 -- during the join stage, i.e. a replica with a minuscule
 -- timeout successfully bootstraps and breaks connection only
 -- after subscribe.
-test_run:cmd("start server replica_timeout with args='0.00001'")
+test_run:cmd("start server replica_timeout with args='0.00001, 0.5'")
 ---
 - true
 ...

test/replication/errinj.test.lua

@@ -173,7 +173,7 @@ errinj.set("ERRINJ_RELAY_EXIT_DELAY", 0)
 box.cfg{replication_timeout = 0.01}
 
 test_run:cmd("create server replica_timeout with rpl_master=default, script='replication/replica_timeout.lua'")
-test_run:cmd("start server replica_timeout with args='0.01'")
+test_run:cmd("start server replica_timeout with args='0.1, 0.5'")
 test_run:cmd("switch replica_timeout")
 
 fiber = require('fiber')
@@ -199,7 +199,7 @@ errinj.set("ERRINJ_RELAY_REPORT_INTERVAL", 0)
 -- Check replica's ACKs don't prevent the master from sending
 -- heartbeat messages (gh-3160).
 
-test_run:cmd("start server replica_timeout with args='0.009'")
+test_run:cmd("start server replica_timeout with args='0.009, 0.5'")
 test_run:cmd("switch replica_timeout")
 
 fiber = require('fiber')
@@ -219,7 +219,7 @@ for i = 0, 9999 do box.space.test:replace({i, 4, 5, 'test'}) end
 -- during the join stage, i.e. a replica with a minuscule
 -- timeout successfully bootstraps and breaks connection only
 -- after subscribe.
-test_run:cmd("start server replica_timeout with args='0.00001'")
+test_run:cmd("start server replica_timeout with args='0.00001, 0.5'")
 test_run:cmd("switch replica_timeout")
 fiber = require('fiber')
 while box.info.replication[1].upstream.message ~= 'timed out' do fiber.sleep(0.0001) end

test/replication/master.lua

@@ -4,6 +4,7 @@ box.cfg({
     listen              = os.getenv("LISTEN"),
     memtx_memory        = 107374182,
     replication_connect_timeout = 0.5,
+    replication_timeout = 0.1
 })
 
 require('console').listen(os.getenv('ADMIN'))

test/replication/master_quorum.lua

@@ -20,7 +20,8 @@ box.cfg({
         instance_uri(2);
     };
     replication_connect_quorum = 0;
-    replication_connect_timeout = 0.1;
+    replication_timeout = 0.5;
+    replication_connect_timeout = 2.0;
 })
 
 test_run = require('test_run').new()

test/replication/on_replace.lua

@@ -20,7 +20,8 @@ box.cfg({
         USER..':'..PASSWORD..'@'..instance_uri(1);
         USER..':'..PASSWORD..'@'..instance_uri(2);
     };
-    replication_connect_timeout = 0.5,
+    replication_timeout = 0.5,
+    replication_connect_timeout = 1.0,
 })
 
 env = require('test_run')

test/replication/quorum.lua

@@ -15,8 +15,8 @@ require('console').listen(os.getenv('ADMIN'))
 box.cfg({
     listen = instance_uri(INSTANCE_ID);
     replication_timeout = 0.05;
-    replication_sync_lag = 0.01;
-    replication_connect_timeout = 0.1;
+    replication_sync_lag = 0.1;
+    replication_connect_timeout = 3.0;
     replication_connect_quorum = 3;
     replication = {
         instance_uri(1);

test/replication/rebootstrap.lua

@@ -15,7 +15,7 @@ box.cfg({
     listen = instance_uri(INSTANCE_ID),
     instance_uuid = '12345678-abcd-1234-abcd-123456789ef' .. INSTANCE_ID,
     replication_timeout = 0.1,
-    replication_connect_timeout = 0.5,
+    replication_connect_timeout = 2.0,
     replication = {
         instance_uri(1);
         instance_uri(2);

test/replication/replica_no_quorum.lua

@@ -5,7 +5,8 @@ box.cfg({
     replication         = os.getenv("MASTER"),
     memtx_memory        = 107374182,
     replication_connect_quorum = 0,
-    replication_connect_timeout = 0.1,
+    replication_timeout = 0.1,
+    replication_connect_timeout = 0.5,
 })
 
 require('console').listen(os.getenv('ADMIN'))

test/replication/replica_timeout.lua

@@ -1,13 +1,14 @@
 #!/usr/bin/env tarantool
 
 local TIMEOUT = tonumber(arg[1])
+local CON_TIMEOUT = arg[2] and tonumber(arg[2]) or TIMEOUT * 3
 
 box.cfg({
     listen              = os.getenv("LISTEN"),
     replication         = os.getenv("MASTER"),
     memtx_memory        = 107374182,
     replication_timeout = TIMEOUT,
-    replication_connect_timeout = TIMEOUT * 3,
+    replication_connect_timeout = CON_TIMEOUT,
 })
 
 require('console').listen(os.getenv('ADMIN'))

test/replication/replica_uuid_ro.lua

@@ -22,7 +22,7 @@ box.cfg({
         USER..':'..PASSWORD..'@'..instance_uri(2);
     };
     read_only = (INSTANCE_ID ~= '1' and true or false);
-    replication_connect_timeout = 0.5,
+    replication_connect_timeout = 5,
 })
 
 box.once("bootstrap", function()