Giters

tikv / pd

Placement driver for TiKV

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

TestPrepareCheckerWithTransferLeader is unstable

lhy1024 opened this issue · comments

commented

Flaky Test

Which jobs are failing

--- FAIL: TestPrepareCheckerWithTransferLeader (38.21s)
    region_syncer_test.go:294: 
        	Error Trace:	/home/runner/work/pd/pd/tests/server/region_syncer/region_syncer_test.go:294
        	Error:      	Not equal: 
        	            	expected: "pd1"
        	            	actual  : "pd2"
        	            	
        	            	Diff:
        	            	--- Expected
        	            	+++ Actual
        	            	@@ -1 +1 @@
        	            	-pd1
        	            	+pd2
        	Test:       	TestPrepareCheckerWithTransferLeader
FAIL

CI link

https://github.com/tikv/pd/actions/runs/9737924850/job/26924893445

Reason for failure (if possible)

Anything else

commented

meet again https://github.com/tikv/pd/actions/runs/9762010374/job/26944878762

2024-07-02T13:54:15.0492409Z --- FAIL: TestPrepareCheckerWithTransferLeader (76.09s)
2024-07-02T13:54:15.0492734Z     region_syncer_test.go:289: 
2024-07-02T13:54:15.0493797Z         	Error Trace:	/home/runner/work/pd/pd/tests/server/region_syncer/region_syncer_test.go:289
2024-07-02T13:54:15.0494204Z         	Error:      	Not equal: 
2024-07-02T13:54:15.0494642Z         	            	expected: "pd2"
2024-07-02T13:54:15.0495194Z         	            	actual  : ""
2024-07-02T13:54:15.0495495Z         	            	
2024-07-02T13:54:15.0496104Z         	            	Diff:
2024-07-02T13:54:15.0496743Z         	            	--- Expected
2024-07-02T13:54:15.0497390Z         	            	+++ Actual
2024-07-02T13:54:15.0498020Z         	            	@@ -1 +1 @@
2024-07-02T13:54:15.0498413Z         	            	-pd2
2024-07-02T13:54:15.0498676Z         	            	+
2024-07-02T13:54:15.0499001Z         	Test:       	TestPrepareCheckerWithTransferLeader
2024-07-02T13:54:15.0499178Z FAIL
commented

meet again tikv/pd/actions/runs/9762010374/job/26944878762

2024-07-02T13:54:15.0492409Z --- FAIL: TestPrepareCheckerWithTransferLeader (76.09s)
2024-07-02T13:54:15.0492734Z     region_syncer_test.go:289: 
2024-07-02T13:54:15.0493797Z         	Error Trace:	/home/runner/work/pd/pd/tests/server/region_syncer/region_syncer_test.go:289
2024-07-02T13:54:15.0494204Z         	Error:      	Not equal: 
2024-07-02T13:54:15.0494642Z         	            	expected: "pd2"
2024-07-02T13:54:15.0495194Z         	            	actual  : ""
2024-07-02T13:54:15.0495495Z         	            	
2024-07-02T13:54:15.0496104Z         	            	Diff:
2024-07-02T13:54:15.0496743Z         	            	--- Expected
2024-07-02T13:54:15.0497390Z         	            	+++ Actual
2024-07-02T13:54:15.0498020Z         	            	@@ -1 +1 @@
2024-07-02T13:54:15.0498413Z         	            	-pd2
2024-07-02T13:54:15.0498676Z         	            	+
2024-07-02T13:54:15.0499001Z         	Test:       	TestPrepareCheckerWithTransferLeader
2024-07-02T13:54:15.0499178Z FAIL

I found this error is in an extreme situation. And I created an issue to focus on the Waitleader failure situation :) #8359

pd/tests/server/region_syncer/region_syncer_test.go

Lines 280 to 289 in 7e18a69

// join new PD
pd2, err := cluster.Join(ctx)
re.NoError(err)
err = pd2.Run()
re.NoError(err)
// waiting for synchronization to complete
time.Sleep(3 * time.Second)
err = cluster.ResignLeader()
re.NoError(err)
re.Equal("pd2", cluster.WaitLeader())

After resign leader to pd-2, which is happened in 2024/07/02 13:53:11.565 +00:00

!!!resign triggered by API!!!
2024-07-02T13:54:14.8419029Z [2024/07/02 13:53:11.565 +00:00] [INFO] [member.go:358] ["try to resign etcd leader to next pd-server"] [from=pd1] [to=]

Due to the slow disk, the leader changed 8 times in the 40s! and no leader check is in 40s which result error

2024-07-02T13:54:14.8776867Z [2024/07/02 13:53:21.728 +00:00] [WARN] [lease.go:175] ["lease keep alive failed"] [purpose="leader election"] [start=2024/07/02 13:53:21.726 +00:00] [error="context canceled"]
2024-07-02T13:54:14.8778148Z [2024/07/02 13:53:21.734 +00:00] [INFO] [lease.go:192] ["stop lease keep alive worker"] [purpose="leader election"]

For this extreme situation, I think we can failed directly.

The detailed content is

2024-07-02T13:54:14.7951174Z [2024/07/02 13:53:01.139 +00:00] [INFO] [server.go:1738] ["campaign PD leader ok"] [campaign-leader-name=pd1]
2024-07-02T13:54:14.8302504Z [2024/07/02 13:53:07.537 +00:00] [INFO] [server.go:1738] ["campaign PD leader ok"] [campaign-leader-name=pd1]

!!!resign triggered by API!!!
2024-07-02T13:54:14.8419029Z [2024/07/02 13:53:11.565 +00:00] [INFO] [member.go:358] ["try to resign etcd leader to next pd-server"] [from=pd1] [to=]

2024-07-02T13:54:14.8523786Z [2024/07/02 13:53:12.651 +00:00] [INFO] [server.go:1738] ["campaign PD leader ok"] [campaign-leader-name=pd2]
2024-07-02T13:54:14.8715386Z [2024/07/02 13:53:18.726 +00:00] [INFO] [server.go:1738] ["campaign PD leader ok"] [campaign-leader-name=pd2]
2024-07-02T13:54:14.8900582Z [2024/07/02 13:53:24.777 +00:00] [INFO] [server.go:1738] ["campaign PD leader ok"] [campaign-leader-name=pd2]

!!!resign pd-2 ->0!!!
2024-07-02T13:54:14.9121126Z [2024/07/02 13:53:30.827 +00:00] [INFO] [member.go:358] ["try to resign etcd leader to next pd-server"] [from=pd2] [to=]
2024-07-02T13:54:14.9167935Z [2024/07/02 13:53:31.329 +00:00] [ERROR] [server.go:1718] ["campaign PD leader meets error due to etcd error"] [campaign-leader-name=pd2] [error="[PD:server:ErrLeaderFrequentlyChange]leader pd2 frequently changed, leader-key is [/pd/7387036149936305935/leader]"]

2024-07-02T13:54:14.9189703Z [2024/07/02 13:53:31.833 +00:00] [INFO] [server.go:1738] ["campaign PD leader ok"] [campaign-leader-name=pd1]

!!!resign pd-1 -> 0!!!
2024-07-02T13:54:14.9374977Z [2024/07/02 13:53:37.894 +00:00] [INFO] [member.go:358] ["try to resign etcd leader to next pd-server"] [from=pd1] [to=]
2024-07-02T13:54:14.9440845Z [2024/07/02 13:53:38.395 +00:00] [ERROR] [server.go:1718] ["campaign PD leader meets error due to etcd error"] [campaign-leader-name=pd1] [error="[PD:server:ErrLeaderFrequentlyChange]leader pd1 frequently changed, leader-key is [/pd/7387036149936305935/leader]"]


2024-07-02T13:54:14.9468215Z [2024/07/02 13:53:38.897 +00:00] [INFO] [server.go:1738] ["campaign PD leader ok"] [campaign-leader-name=pd2]
2024-07-02T13:54:14.9755943Z [2024/07/02 13:53:44.934 +00:00] [INFO] [server.go:1738] ["campaign PD leader ok"] [campaign-leader-name=pd2]
2024-07-02T13:54:15.0006334Z [2024/07/02 13:53:50.978 +00:00] [INFO] [server.go:1738] ["campaign PD leader ok"] [campaign-leader-name=pd2]

!!!resign pd-2 -> 0!!!
2024-07-02T13:54:15.0132961Z [2024/07/02 13:53:57.018 +00:00] [INFO] [member.go:358] ["try to resign etcd leader to next pd-server"] [from=pd2] [to=]
2024-07-02T13:54:15.0152409Z [2024/07/02 13:53:57.520 +00:00] [ERROR] [server.go:1718] ["campaign PD leader meets error due to etcd error"] [campaign-leader-name=pd2] [error="[PD:server:ErrLeaderFrequentlyChange]leader pd2 frequently changed, leader-key is [/pd/7387036149936305935/leader]"]

2024-07-02T13:54:15.0162553Z [2024/07/02 13:53:58.021 +00:00] [INFO] [server.go:1738] ["campaign PD leader ok"] [campaign-leader-name=pd1]

And The WaitLeader times out because all pd's do not all recognize the leader.

When Checkleader's etcd get is too slow(> leader lease)

  • follower was blocked, failing to recognize the current leader.
  • The old leader exits leadership as lease time is reached within the follower blocking time.
  • follower watch with revision can still fetch the leader exit event and then exit watch status.

The above event loop occurred for 50 seconds, causing the WaitLeader to time out.

You can check

!!! pd2 was elected leader at 13:53:12.651 !!!
[2024/07/02 13:53:12.651 +00:00] [INFO] [server.go:1738] ["campaign PD leader ok"] [campaign-leader-name=pd2]

!!! pd1 skip campaigning stops at 13:53:12.479, at which point it goes to getPersistentLeader in CheckLeader, and then it takes 6s to get the leader and start watching !!!
[2024/07/02 13:53:12.278 +00:00] [INFO] [server.go:1700] ["skip campaigning of pd leader and check later"] [server-name=pd1] [etcd-leader-id=13430743067140231612] [member-id=15634440769998151696]
[2024/07/02 13:53:12.479 +00:00] [INFO] [server.go:1700] ["skip campaigning of pd leader and check later"] [server-name=pd1] [etcd-leader-id=13430743067140231612] [member-id=15634440769998151696]

!!! CheckLeader !!!
[2024/07/02 13:53:18.684 +00:00] [WARN] [etcdutil.go:153] ["kv gets too slow"] [request-key=/pd/7387036149936305935/leader] [cost=6.004986363s] []
[2024/07/02 13:53:18.725 +00:00] [INFO] [server.go:1669] ["start to watch pd leader"] [pd-leader="name:\"pd2\" member_id:13430743067140231612 peer_urls:\"http://127.0.0.1:38109\" client_urls:\"http://127.0.0.1:40999\" "]

!!! pd1 watch found the leader changed after only 150ms, resulting in the WaitLeader not being successful !!!
[2024/07/02 13:53:18.862 +00:00] [INFO] [leadership.go:388] ["current leadership is deleted"] [revision=57] [leader-key=/pd/7387036149936305935/leader] [purpose="leader election"]
[2024/07/02 13:53:19.728 +00:00] [INFO] [server.go:1673] ["pd leader has changed, try to re-campaign a pd leader"]
commented 
--- FAIL: TestPrepareCheckerWithTransferLeader (15.97s)
    region_syncer_test.go:301: 
        	Error Trace:	/home/runner/work/pd/pd/tests/server/region_syncer/region_syncer_test.go:301
        	Error:      	Should not be: "pd1"
        	Test:       	TestPrepareCheckerWithTransferLeader

https://github.com/tikv/pd/actions/runs/9969178086/job/27545694334

commented

Does this issue need reopen? @HuSharp