TiCDC ResolvedTs stuck for half an hour

[hongyunyan]

What did you do?

Add sysbench workload in a cluster with lots of jxx tables, which leading to some create table ddls.

What did you expect to see?

ticdc works smoothly, with reasonable resolve ts and checkpoint ts

What did you see instead?

resolve ts and checkpoint ts stuck for half an hour.

Based on the log and metrics, we find it's due to the incremental scan cost too many times. And for the incremental scanning, the cpu is reach 100%.

[2024/03/25 07:50:18.403 +00:00] [INFO] [initializer.rs:545] ["cdc incremental scan uses ts filter: true"] [tables=13] [filtered_versions=41367935] [mvcc_versions=41367935] [hint_min_ts=448620924264776010] [region_id=63851] [thread_id=215]

Versions of the cluster

Upstream TiDB cluster version (execute SELECT tidb_version(); in a MySQL client):

(paste TiDB cluster version here)

Upstream TiKV version (execute tikv-server --version):

(paste TiKV version here)

TiCDC version (execute cdc version):

(paste TiCDC version here)

[hicqu]

Seems duplicate with tikv/tikv#16601.

[hicqu]

Seems the slow scan region is the last one in whole key space:

[2024/03/25 09:59:58.475 +00:00] [INFO] [region.go:740] ["region Version changed"] [region-id=63851] [detail="StartKey Changed:{7480000000000033FF6E5F72B1E04F919AFFD0F6830000000000FA} -> {7480000000000033FF725F698000000000FF00000103C1179678FFE93A796903C11796FF78E93A7969000000FC}, EndKey:{748000FFFFFFFFFFFFF900000000000000F8}"] [old-version=4570] [new-version=4572]

[fubinzh]

/severity major