table sink stucks after cdc owner switch

[fubinzh]

What did you do?

1. create changefeed
2. run sysbench workload
3. Inject cdc owner failure twice at 08:44 and 09:14

What did you expect to see?

CDC task should be OK

What did you see instead?

Table sink stucks sometimes, and finally changefeed "changefeed is in unretryable state, please check the error message, and you should manually handle it: table sink stuck"

[2023/08/14 08:46:53.019 +00:00] [WARN] [manager.go:984] ["Table checkpoint is stuck too long, will restart the sink backend"] [namespace=default] [changefeed=owner-switch-sync] [span={table_id:319,start_key:7480000000000001ff3f5f720000000000fa,end_key:7480000000000001ff3f5f730000000000fa}] [checkpointTs="{\"Mode\":0,\"Ts\":443548182204121088,\"BatchID\":18446744073709551615}"] [stuckCheck=150] [factoryVersion=1]
[2023/08/14 08:46:53.019 +00:00] [INFO] [manager.go:384] ["Sink manager tries to put an sink error"] [namespace=default] [changefeed=owner-switch-sync] [skipped=false] [error="table sink stuck"]
[2023/08/14 08:46:53.019 +00:00] [WARN] [manager.go:263] ["Sink manager backend sink fails"] [namespace=default] [changefeed=owner-switch-sync] [factoryVersion=1] [error="table sink stuck"] [errorVerbose="table sink stuck\ngithub.com/pingcap/tiflow/cdc/processor/sinkmanager.(*SinkManager).GetTableStats\n\tgithub.com/pingcap/tiflow/cdc/processor/sinkmanager/manager.go:992\ngithub.com/pingcap/tiflow/cdc/processor.(*processor).GetTableSpanStatus\n\tgithub.com/pingcap/tiflow/cdc/processor/processor.go:333\ngithub.com/pingcap/tiflow/cdc/scheduler/internal/v3/agent.(*tableSpan).getAndUpdateTableSpanState\n\tgithub.com/pingcap/tiflow/cdc/scheduler/internal/v3/agent/table.go:57\ngithub.com/pingcap/tiflow/cdc/scheduler/internal/v3/agent.(*tableSpanManager).poll.func1\n\tgithub.com/pingcap/tiflow/cdc/scheduler/internal/v3/agent/table.go:322\ngithub.com/pingcap/tiflow/cdc/scheduler/internal/v3/agent.(*tableSpanManager).poll.(*BtreeMap[...]).Ascend.func2\n\tgithub.com/pingcap/tiflow/pkg/spanz/btree_map.go:110\ngithub.com/google/btree.(*node[...]).iterate\n\tgithub.com/google/btree@v1.1.2/btree_generic.go:522\ngithub.com/google/btree.(*node[...]).iterate\n\tgithub.com/google/btree@v1.1.2/btree_generic.go:510\ngithub.com/google/btree.(*BTreeG[...]).Ascend\n\tgithub.com/google/btree@v1.1.2/btree_generic.go:779\ngithub.com/pingcap/tiflow/pkg/spanz.(*BtreeMap[...]).Ascend\n\tgithub.com/pingcap/tiflow/pkg/spanz/btree_map.go:109\ngithub.com/pingcap/tiflow/cdc/scheduler/internal/v3/agent.(*tableSpanManager).poll\n\tgithub.com/pingcap/tiflow/cdc/scheduler/internal/v3/agent/table.go:315\ngithub.com/pingcap/tiflow/cdc/scheduler/internal/v3/agent.(*agent).Tick\n\tgithub.com/pingcap/tiflow/cdc/scheduler/internal/v3/agent/agent.go:214\ngithub.com/pingcap/tiflow/cdc/processor.(*processor).tick\n\tgithub.com/pingcap/tiflow/cdc/processor/processor.go:586\ngithub.com/pingcap/tiflow/cdc/processor.(*processor).Tick\n\tgithub.com/pingcap/tiflow/cdc/processor/processor.go:474\ngithub.com/pingcap/tiflow/cdc/processor.(*managerImpl).Tick\n\tgithub.com/pingcap/tiflow/cdc/processor/manager.go:141\ngithub.com/pingcap/tiflow/pkg/orchestrator.(*EtcdWorker).Run\n\tgithub.com/pingcap/tiflow/pkg/orchestrator/etcd_worker.go:290\ngithub.com/pingcap/tiflow/cdc/capture.(*captureImpl).runEtcdWorker\n\tgithub.com/pingcap/tiflow/cdc/capture/capture.go:577\ngithub.com/pingcap/tiflow/cdc/capture.(*captureImpl).run.func4\n\tgithub.com/pingcap/tiflow/cdc/capture/capture.go:388\ngolang.org/x/sync/errgroup.(*Group).Go.func1\n\tgolang.org/x/sync@v0.2.0/errgroup/errgroup.go:75\nruntime.goexit\n\truntime/asm_amd64.s:1650"]

...

[2023/08/14 09:26:06.244 +00:00] [ERROR] [feed_state_manager.go:434] ["processor reports an error"] [namespace=default] [changefeed=owner-switch-sync] [captureID=f07dfb1f-f029-4a38-811a-23b898717b10] [error="{\"time\":\"2023-08-14T09:26:06.225683566Z\",\"addr\":\"upstream-ticdc-2.upstream-ticdc-peer.cdc-testbed-tps-1891603-1-66.svc:8301\",\"code\":\"CDC:ErrChangefeedUnretryable\",\"message\":\"[CDC:ErrChangefeedUnretryable]changefeed is in unretryable state, please check the error message, and you should manually handle it: table sink stuck\"}"]

Versions of the cluster

CDC version

["Welcome to Change Data Capture (CDC)"] [release-version=v7.4.0-alpha] [git-hash=dd8da8e725052e3abb8be12513f74ebb5e479a22] [git-branch=heads/refs/tags/v7.4.0-alpha] [utc-build-time="2023-08-12 11:36:23"] [go-version="go version go1.21.0 linux/amd64"] [failpoint-build=false]

[fubinzh]

cdc log:
ticdc-1_1891603.log

[hicqu]

[2023/08/14 08:46:53.019 +00:00] [WARN] [manager.go:984] ["Table checkpoint is stuck too long, will restart the sink backend"] [namespace=default] [changefeed=owner-switch-sync] [span={table_id:319,start_key:7480000000000001ff3f5f720000000000fa,end_key:7480000000000001ff3f5f730000000000fa}] [checkpointTs="{\"Mode\":0,\"Ts\":443548182204121088,\"BatchID\":18446744073709551615}"] [stuckCheck=150] [factoryVersion=1]

In the log the tso can be parsed to 2023-08-14 16:33:57.827. So I guess the root cause is sink is really slow. It can cause DML sink destroy and re-create, which makes sink more slow and the case becomes worse.

[fubinzh]

This issue seen twice during daily ci, and not reproduced after that.
Is there any workaround if it happens? @hicqu

[fubinzh]

/found automation
/severity moderate

[nongfushanquan]

fixed by #9747

[nongfushanquan]

/close

[ti-chi-bot]

@nongfushanquan: Closing this issue.

In response to this:

/close

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.