[potential bug] TiFlash replica not available for about 10 minutes with error `tikv:9006`.

During running schrodinger/sqllogic test, I met the following problem:

The test process

1. created a table,
2. add a tiflash replica,
3. wait for the replica available, but it not available for about 10 minutes.

Note that before this operation, there was a heavy load in this tidb cluster produced by other threads of the sqllogic test.
And when it was waiting, all of the load was hanging.

tiflash_cluster_manager.log

Exception: can not get tiflash replica info from tidb: [('127.0.0.1:8539', Exception('unexpected status code 400 from http://127.0.0.1:8539/tiflash/replica'))]
2021-02-22 13:39:09,528 <ERROR> root: can not get tiflash replica info from tidb: [('127.0.0.1:8539', Exception('unexpected status code 400 from http://127.0.0.1:8539/tiflash/replica'))]
Traceback (most recent call last):
  File "flash_cluster_manager.py", line 286, in main
  File "flash_cluster_manager.py", line 129, in __init__
  File "flash_cluster_manager.py", line 29, in wrap_func
  File "flash_cluster_manager.py", line 238, in table_update
  File "tidb_tools.py", line 42, in db_flash_replica

http://h81:8539/tiflash/replica
(h81 is the host of the cluster, 8539 is tidb status port.)

[tikv:9006]GC life time is shorter than transaction duration, transaction starts at 2021-02-22 13:21:08.223 +0800 CST, GC safe point is 2021-02-22 13:21:08.923 +0800 CST

Start by reproducing the schrodinger/sqllogic sequence under heavy cluster load and inspect the tiflash_cluster_manager.py and tidb_tools.py call paths named in the traceback. Check the /tiflash/replica response alongside the tikv:9006 GC lifetime error. Done means identifying and resolving the cause of the delayed replica availability and hanging test load.