[next gen on clould] gc interface can blow up the PD during simulating one of pd follower network delay 100ms #9724
Description
Bug Report
What did you do?
1、run simulated workload
nohup sysbench --db-driver=mysql --mysql-user=root --mysql-host=xxx --mysql-port=4000 --mysql-password=xxx --mysql-db=bingo_dc --order_tables=38 --update_ratio=0.1 --cop_read=0.001 --batch_insert=0.0002 --threads=500 --report-interval=10 --time=0 c66_dc run | while IFS= read -r line; do printf '[%s] %s\n' "$(date '+%Y-%m-%d %H:%M:%S')" "$line"; done > sysbench_00907_1.log 2>&1 &
nohup sysbench --db-driver=mysql --mysql-user=root --mysql-host=xxx --mysql-port=4000 --mysql-password=xxx --mysql-db=bingo_dc --order_tables=38 --threads=1 --time=0 tiflash_read run | while IFS= read -r line; do printf '[%s] %s\n' "$(date '+%Y-%m-%d %H:%M:%S')" "$line"; done > sysbench_tiflash_00907_1.log 2>&1 &
2、one pd follower network is simulated with 100ms delay with all tikv, tidb and other pd
Chaos: 2025/09/06 22:59:40.402 +08:00 ~ 2025/09/06 23:09:40.403 +08:00
What did you expect to see?
The frequency of requests on the gc side needs to be controlled
What did you see instead?
workload report error
[2025-09-06 15:01:18] FATAL: mysql_drv_query() returned error 9001 (PD server timeout: start timestamp may fall behind safe point) for query 'SELECT ID, BILLNO, LOGINNAME, AGCODE, PRODUCT_ID, PLATFORM_ID,
tiflash query
[2025-09-06 15:00:27] start gd_api_select_4, table_id: 1
[2025-09-06 15:01:59] FATAL: mysql_drv_query() returned error 1317 (Query execution was interrupted) for query ' select
What version of PD are you using (pd-server -V)?
sh-5.1# ./pd-server -V
Release Version: v9.0.0-beta.2.pre-103-g2d131a5
Edition: Enterprise
Kernel Type: Next Generation
Git Commit Hash: 2d131a5
Git Branch: HEAD
UTC Build Time: 2025-09-04 11:12:01