cancel the workload script cause the resolved-ts spike

[3AceShowHand]

• run the following scripts with the command ./xxx 20000 0
• ctrl + c to stop the script

#!/bin/bash

batch_size=$1
sleep_duration=$2

TABLE_NAME="test.bank0"

i=0
while true
do
    sleep "$sleep_duration"

    i=$((i + 1))
    start_ts=$(date +%s%3N)

    mysql -h$HOST -P$PORT -e "DELETE FROM $TABLE_NAME LIMIT $batch_size;"

    end_ts=$(date +%s%3N)
    cost_ms=$((end_ts - start_ts))

    echo "Round $i: delete from $TABLE_NAME finished, cost=${cost_ms} ms"
done

The resolved-ts spikes, and TiKV-CDC endpoint CPU become full

This may caused by the stale lock, the resolved-ts drop to 0 after GC

At the same time, the TiKV-CDC endpoint CPU usage is full, this is not expected.

tikv：4855bdccc64e7a8551d30ebbbd5be75a42929265

[3AceShowHand]

The high CPU usage may caused by scan old value

After cancel the workload, seek operation count dropped from 16k ops / s to around 1.9k ops / s. But the seek duration increased from 700us to around 4.5ms.

At the same time, there is compaction related logs printed, is there any relationship ?

[3AceShowHand]

The resolved-ts drop to 0 may due to the ticdc reslove locks, this is good.