Server hangs when the bufferpool is shrunk to 8M and innodb_page_size=64K

Observation on

origin/preview-10.8-MDEV-14425-innodb f6b1e6fbae27842f78b28ad8ceb898e6008841d1 2022-01-07T18:59:05+02:00

One RQG test starts the server with innodb_page_size 64K and bufferpool_size 256M,

performs a bufferpool resizing down to 8M (message in server error log) and fails during or short after that with

[ERROR] [FATAL] InnoDB: innodb_fatal_semaphore_wait_threshold was exceeded for dict_sys.latch.

Per my experiences with server startup on official versions of MariaDB

InnoDB denies to work in case the page_size is 64K and the bufferpool_size

is < 24M because of whatever reasons.

Nevertheless InnoDB does not deny a resizing of the bufferpool to 8M at runtime.

The MTR based test reports on

- origin/10.7  2022-01-06

   SET GLOBAL innodb_buffer_pool_size = 8388608;

   Warnings:

   Warning 1210    innodb_buffer_pool_size must be at least innodb_buffer_pool_chunk_size=25165824

- origin/10.8 897d8c57b68858a58eea6cda4d971c3a94f6cc4a 2021-11-29

  a warning like above

- origin/10.8 commit baace572dc1d2ab35145556e9db231dca8c18147 (HEAD)

   MDEV-25342: autosize innodb_buffer_pool_chunk_size

   mysqltest: At line 11: query 'SET GLOBAL innodb_buffer_pool_size = @innodb_buffer_pool_size_save' failed: ER_WRONG_ARGUMENTS (1210): Another buffer pool resize is already in progress.

   which is caused by the ongoing asynchronous buffer pool resizing to 8M issued earlier.

   The server error log shows

   2022-01-10 17:59:45 4 [Note] InnoDB: Requested to resize buffer pool. (new size: 8388608 bytes)

   2022-01-10 17:59:45 0 [Note] InnoDB: Resizing buffer pool from 25165824 to 8388608 (unit=1048576).