test: vinyl/errinj.test.lua fails under high load

[avtikhon]

Tarantool version:
All

OS version:
All

Bug description:
[009] --- vinyl/errinj.result Tue Apr 16 16:43:36 2019
[009] +++ vinyl/errinj.reject Wed Apr 17 09:42:36 2019
[009] @@ -530,7 +530,7 @@
[009] ...
[009] t2 - t1 < 1
[009] ---
[009] -- true
[009] +- false
[009] ...
[009] test_run:cmd("cleanup server test")
[009] ---
[009]

Checked under highload the delay and found that it is longer than checked 1 second and in resl near to 1.5 sec:
[029] --- vinyl/errinj.result Wed Apr 17 10:18:47 2019
[029] +++ vinyl/errinj.reject Wed Apr 17 10:20:18 2019
[029] @@ -528,9 +528,9 @@
[029] t2 = fiber.time()
[029] ---
[029] ...
[029] -t2 - t1 < 1
[029] ----
[029] -- true
[029] +t2 - t1 < 1 or t2 - t1
[029] +---
[029] +- 1.6016597747803
[029] ...
[029] test_run:cmd("cleanup server test")
[029] ---
[029]

in 100 parallel runs the failed delays were:
[002] +- 1.4104716777802
[022] +- 1.3933029174805
[044] +- 1.4296517372131
[033] +- 1.6380662918091
[001] +- 1.9799520969391
[027] +- 1.7067711353302
[043] +- 1.3778221607208
[034] +- 1.3820221424103
[032] +- 1.3820221424103
[020] +- 1.6275615692139
[050] +- 1.6275615692139
[048] +- 1.1880359649658

Suggested fix is to increase the awaiting delay up to 3 or even 5 secs.

Test "check that all dump/compaction tasks that are in progress at
the time when the server stops are aborted immediately.":

s = box.schema.space.create('test', {engine = 'vinyl'})
_ = s:create_index('i1', {parts = {1, 'unsigned'}})
_ = s:create_index('i2', {parts = {2, 'unsigned'}})
box.error.injection.set('ERRINJ_VY_RUN_WRITE_STMT_TIMEOUT', 0.01)
for i = 1, 1000 do s:replace{i, i} end
_ = fiber.create(function() box.snapshot() end)
fiber.sleep(0.01)
test_run:cmd('switch default')
t1 = fiber.time()
test_run:cmd("stop server test")
t2 = fiber.time()
t2 - t1 < 1

Steps to reproduce:
./test-run.py -j 50 --force vinyl/

Optional (but very desirable):

• coredump
• backtrace
• netstat