test: fix flaky failover/failover.test.lua

Router pings replicas and if they cannot respond in
failover_ping_timeout, it detaches connection from this replica.
The test checks, that connection is not closed, when long response is
done, even if failover_ping_timeout is small.

However, as the connection is detached, it can be garbage collected,
which leads to error. Let's stop garbage collection for this test.

NO_DOC=test

test/failover/failover.result

@@ -479,6 +479,10 @@ vshard.router.cfg(cfg)
 while not test_run:grep_log('router_1', 'Ping error from', 1000) do fiber.sleep(0.01) end
 ---
 ...
+collectgarbage("stop")
+---
+- 0
+...
 t = string.rep('a', 1024 * 1024 * 500)
 ---
 ...
@@ -508,6 +512,10 @@ err
 ---
 - 524288000
 ...
+collectgarbage("restart")
+---
+- 0
+...
 cfg.failover_ping_timeout = nil
 ---
 ...

test/failover/failover.test.lua

@@ -177,6 +177,7 @@ cfg.failover_ping_timeout = 0.0000001
 vshard.router.cfg(cfg)
 while not test_run:grep_log('router_1', 'Ping error from', 1000) do fiber.sleep(0.01) end
 
+collectgarbage("stop")
 t = string.rep('a', 1024 * 1024 * 500)
 rs = vshard.router.route(31)
 while rs.master ~= rs.replica do fiber.sleep(0.01) end
@@ -186,6 +187,7 @@ vshard.router.static.failover_fiber:wakeup()
 res, err = future:wait_result(5)
 err
 #res[1]
+collectgarbage("restart")
 cfg.failover_ping_timeout = nil
 vshard.router.cfg(cfg)