Safe Way to Reduce innodb_log_file_size on MariaDB 11.4 + Performance Tuning Advice

Hello,

A few months ago, I attempted to reduce the innodb_log_file_size value from 32G to 2G. Unfortunately, this resulted in a crash of the MariaDB server. The process corrupted ib_logfile0 and ibdata1, rendering all databases unusable. I had to restore everything from backups. At the time, I followed instructions provided by ChatGPT, but likely missed a critical step.

I would now like to try again—this time more carefully—with proper guidance.
My goal is to set innodb_log_file_size=2G, if that's appropriate and safe.

Server Specs:

• MariaDB version: 11.4

• Websites hosted: ~130 WordPress sites

• RAM: 196GB

• CPU: 50 cores

• Storage:

  • 2 × 3.84 TB NVMe in software RAID 1 (main OS & high-performance sites – AlmaLinux 8)

  • 2 × 16TB HDD in software RAID 1 (lower-traffic sites & mail storage)

• Web server: LiteSpeed (2 workers) + QUIC.cloud

Observations:

• Website frontend performance is excellent.

• Size of databases varies from 50MB to 1GB.

• WordPress admin panels (saving posts, plugin updates, etc.) feel slower than expected—especially compared to other hosting environments.

• CPU load is usually around 5% - 7%.

Questions:

1. What is the safest procedure to change innodb_log_file_size from 32G to 2G?

2. Can you review my my.cnf (below) and suggest any additional performance tuning, especially to improve WordPress admin responsiveness?

my.cnf content:

[client-server]

!includedir /etc/my.cnf.d

[mysqld]
# Server Optimization
log-error=/var/lib/mysql/xxxx.xxxx.xx.err
performance-schema=1
# innodb_force_recovery = 6
innodb_buffer_pool_size=96G
innodb_log_file_size=32G
innodb_log_buffer_size=512M
innodb_flush_log_at_trx_commit=2
innodb_flush_method=O_DIRECT
innodb_file_per_table=1
innodb_read_io_threads=16
innodb_write_io_threads=16
innodb_io_capacity=3000
innodb_io_capacity_max=6000
innodb_buffer_pool_instances=64
innodb_thread_concurrency=48
innodb_flush_neighbors=0
# innodb_numa_interleave=ON
innodb_page_cleaners=16
tmpdir = /dev/shm

# Connection Optimizations
max_connections=500
max_user_connections=450
wait_timeout=600
interactive_timeout=600
max_allowed_packet=512M
lock_wait_timeout=120

# Query Cache and Caching
query_cache_type=0
query_cache_size=0
join_buffer_size=1M
sort_buffer_size=1M
tmp_table_size=2G
max_heap_table_size=2G
table_open_cache=100000
table_definition_cache=100000
open_files_limit=250000
table_open_cache_instances=64
metadata_locks_cache_size=2048

# Buffering and Threading
thread_cache_size=256
thread_pool_size=36
thread_handling=pool-of-threads
read_buffer_size=1M
read_rnd_buffer_size=1M

# Authentication and Networking
plugin-load-add=auth_socket.so
unix_socket=OFF
bind-address=127.0.0.1

# Logging and Slow Query Log
log_slow_query=1
log_slow_query_file="/var/lib/mysql/slow.log"
log_slow_query_time=2
log_queries_not_using_indexes=0
log_slow_admin_statements=1
log_slow_slave_statements=1

# Replication and Binlog (if used)
binlog_format=ROW
expire_logs_days=7