【TiDB 使用环境】测试环境
【TiDB 版本】
7.5.3/8.5.6
【部署方式】
容器化部署，单机部署tiddb pd kv，tidb 官方镜像组件
【操作系统/CPU 架构/芯片详情】
pve 虚拟机 16c 32g almalinux
8.5.6 两种配置
store-io-pool-size = 1
batch-compression-threshold = “4KiB” （8x默认配置）
store-io-pool-size = 0
batch-compression-threshold = “8KiB” （手动配置7x相同配置）
【问题复现路径】
使用sysbench 进行压测 测试数据规模为 16 张表，每张表 100 万行；每项测试执行 3 轮，并记录 QPS/TPS、P95 延迟
结果如下
| 测试项 | TiDB 7.5.3 | 调整后 TiDB 8.5.6 | 变化 |
|---|---|---|---|
| 只读 QPS | 11,971.46 | 10,761.82 | -10.10% |
| 等值查询 QPS | 17,736.36 | 14,618.86 | -17.57% |
| 带索引更新 TPS | 3,275.79 | 2,958.41 | -9.69% |
| 不带索引更新 TPS | 5,204.07 | 5,070.71 | -2.56% |
| 插入 TPS | 3,854.85 | 3,629.99 | -5.83% |
| 删除 TPS | 3,408.10 | 3,238.51 | -4.98% |
| 测试项 | 7.5.3 | 调整后 8.5.6 | 变化 |
|---|---|---|---|
| 只读 | 117.82ms | 124.86ms | +5.98% |
| 等值查询 | 7.17ms | 8.08ms | +12.69% |
| 带索引更新 | 29.23ms | 32.15ms | +10.00% |
| 不带索引更新 | 18.39ms | 18.39ms | 基本持平 |
| 插入 | 23.81ms | 26.20ms | +10.04% |
| 删除 | 29.02ms | 30.08ms | +3.65% |
为什么看changelog 里都是性能优化，但是实际7.5.3 升级到8.5.6，在配置不变的情况下
性能反而下降了？
附上tikv 配置
# TiKV config template
# Human-readable big numbers:
# File size(based on byte): KB, MB, GB, TB, PB
# e.g.: 1_048_576 = "1MB"
# Time(based on ms): ms, s, m, h
# e.g.: 78_000 = "1.3m"
log-level = "info"
log-rotation-size = "30MB"
log-rotation-timespan = "1h"
memory_usage_limit = "4207MB"
[readpool]
[readpool.coprocessor]
high-concurrency = 64
low-concurrency = 64
normal-concurrency = 32
use-unified-pool = true
[readpool.storage]
normal-concurrency = 10
use-unified-pool = true
[readpool.unified]
auto-adjust-pool-size = true
[pessimistic-txn]
enabled = true
pipelined = true
[server]
grpc-concurrency = 16
grpc-raft-conn-num = 16
[storage]
[storage.block-cache]
capacity = "3183MB"
shared = true
[pd]
# This section will be overwritten by command line parameters
[metric]
[raftstore]
apply-max-batch-size = 1024
raftdb-path = ""
store-max-batch-size = 1024
store-io-pool-size = 0
[raft-engine]
batch-compression-threshold = "8KiB"
[coprocessor]
[rocksdb]
max-background-flushes = 4
max-background-jobs = 16
use-direct-io-for-flush-and-compaction = true
wal-bytes-per-sync = 128
[rocksdb.defaultcf]
compression-per-level = ["lz4", "lz4", "lz4", "lz4", "zstd", "zstd", "zstd"]
level0-slowdown-writes-trigger = 64
level0-stop-writes-trigger = 64
[rocksdb.lockcf]
level0-slowdown-writes-trigger = 64
level0-stop-writes-trigger = 64
[rocksdb.writecf]
compression-per-level = ["lz4", "lz4", "lz4", "lz4", "zstd", "zstd", "zstd"]
level0-slowdown-writes-trigger = 64
level0-stop-writes-trigger = 64
[raftdb]
use-direct-io-for-flush-and-compaction = true
[raftdb.defaultcf]
compression-per-level = ["lz4", "lz4", "lz4", "lz4", "zstd", "zstd", "zstd"]
[security]