[tarantool-patches] [PATCH v3] Use SIGKILL to stop server replica

[avtikhon]

Used the signal option set to SIGKILL to stop server replica
routine to be able to stop the replica immediately to imitate
the replica crash and, then, wake up.
The current case happened when we wanted to set ERRINJ_WAL_DELAY
for a tarantool instance and then stop it. By default the
SIGTERM was used and was not sufficient there, because the
main thread stil waited for the stuck WAL thread after the
signal. In that case the replica finished reading the *.xlog
file, but master server didn't know about it and saved the
previous *.xlog file for replica for its restart. When the
signal was changed from default to 9 (SIGKILL) replica didn't
have a chance to read all data from *.xlog file due to it was
killed immediately. So after replica restart it removed the previous
*.xlog file after its reading.
The logic of the replication was tried to change, but met
the new issues, so the suggested fix at commit:
b5b4809cf2e6d48230eb9e4301eac188b080e0f4
was reverted at commit:
766cd3e1015f6f76460a748c37212fb4c8791500

[029] --- replication/gc.result Mon Apr 15 14:58:09 2019
[029] +++ replication/gc.reject Tue Apr 16 09:17:47 2019
[029] @@ -290,7 +290,12 @@
[029] ...
[029] wait_xlog(1) or fio.listdir('./master')
[029] ---
[048] replication/gc.test.lua vinyl [ fail ]
[048]
[048] Test failed! Result content mismatch:
[029] -- true
[029] +- - 00000000000000000305.vylog
[029] + - 00000000000000000305.xlog
[029] + - '512'
[029] + - 00000000000000000310.xlog
[029] + - 00000000000000000310.vylog
[029] + - 00000000000000000310.snap
[029] ...
[029] -- Stop the replica.
[029] test_run:cmd("stop server replica")
[029] @@ -326,7 +331,13 @@
[029] ...
[029] wait_xlog(2) or fio.listdir('./master')
[029] ---
[029] -- true
[029] +- - 00000000000000000305.xlog
[029] + - 00000000000000000316.xlog
[029] + - 00000000000000000316.vylog
[029] + - '512'
[029] + - 00000000000000000310.xlog
[029] + - 00000000000000000317.vylog
[029] + - 00000000000000000317.snap
[029] ...
[029] -- The xlog should only be deleted after the replica
[029] -- is unregistered.
[029]

Close #4162
---

Github: https://github.com/tarantool/tarantool/tree/avtikhon/gh-4162-stop-kill
Issue: https://github.com/tarantool/tarantool/issues/4162

 test/replication/gc.result   | 16 +++++++---------
 test/replication/gc.test.lua | 12 ++++++------
 2 files changed, 13 insertions(+), 15 deletions(-)

diff --git a/test/replication/gc.result b/test/replication/gc.result
index 65785f47b..396e7ff10 100644
--- a/test/replication/gc.result
+++ b/test/replication/gc.result
@@ -252,20 +252,18 @@ wait_xlog(2) or fio.listdir('./master')
 ---
 - true
 ...
-test_run:cmd("switch replica")
+-- Imitate the replica crash and, then, wake up.
+-- Just 'stop server replica' (SIGTERM) is not sufficient to stop
+-- a tarantool instance when ERRINJ_WAL_DELAY is set, because
+-- "tarantool" thread wait for paused "wal" thread infinitely.
+test_run:cmd("stop server replica with signal=9")
 ---
 - true
 ...
--- Unblock the replica and break replication.
-box.error.injection.set("ERRINJ_WAL_DELAY", false)
----
-- ok
-...
-box.cfg{replication = {}}
+test_run:cmd("start server replica")
 ---
+- true
 ...
--- Restart the replica to reestablish replication.
-test_run:cmd("restart server replica")
 -- Wait for the replica to catch up.
 test_run:cmd("switch replica")
 ---
diff --git a/test/replication/gc.test.lua b/test/replication/gc.test.lua
index 890fe29ae..1ebf32cc8 100644
--- a/test/replication/gc.test.lua
+++ b/test/replication/gc.test.lua
@@ -122,12 +122,12 @@ fiber.sleep(0.1) -- wait for master to relay data
 -- the old snapshot.
 wait_gc(1) or box.info.gc()
 wait_xlog(2) or fio.listdir('./master')
-test_run:cmd("switch replica")
--- Unblock the replica and break replication.
-box.error.injection.set("ERRINJ_WAL_DELAY", false)
-box.cfg{replication = {}}
--- Restart the replica to reestablish replication.
-test_run:cmd("restart server replica")
+-- Imitate the replica crash and, then, wake up.
+-- Just 'stop server replica' (SIGTERM) is not sufficient to stop
+-- a tarantool instance when ERRINJ_WAL_DELAY is set, because
+-- "tarantool" thread wait for paused "wal" thread infinitely.
+test_run:cmd("stop server replica with signal=9")
+test_run:cmd("start server replica")
 -- Wait for the replica to catch up.
 test_run:cmd("switch replica")
 test_run:wait_cond(function() return box.space.test:count() == 310 end, 10)
-- 
2.17.1