S1-B D1.1/D1.2: compose SSD 基线改造——MySQL buffer_pool 6G→10G+io_capacity 2000/10000+redo 1G(设计规范§五,31G 内存容器组预算留痕);Kafka 恢复 compose 管理(KRaft 单节点/QUORUM_VOTERS=服务名自解析/flush 放宽 100000msg·10s,老机事故史归档注记)

This commit is contained in:
2026-11-13 16:26:14 +08:00
parent daa123fa23
commit 59866809fb
+56 -48
View File
@@ -2,19 +2,16 @@
# go-sms(Go-sms)本地开发环境中间件(Day 1)
# 使用:make up(启动并等待 healthy)/ make down(完全清理)/ make logs
# 端口规划(避免本机与中间件默认端口冲突):
# MySQL 13306 | Redis 16379 | Kafka 19092(外部容器 go-sms-kafka 提供,见下) | ETCD 12379
# MySQL 13306 | Redis 16379 | Kafka 19092(compose 内 go-sms-kafka 提供,见下) | ETCD 12379
# ClickHouse HTTP 18123 / Native 19000 | MinIO API 19010 / Console 19011
#
# ⚠️ Kafka 说明(2026-09-11 Day47 运维处置,详见 docs/runbook/incident-response.md):
# 本 compose 曾含 kafka 服务(go-sms-kafka),历史原因容器脱离 go-sms-dev_default 网络,
# KRaft QUORUM_VOTERS(1@kafka:29093)解析失败→broker 永不就绪→重启循环 1271 次,
# 且与外部 Kafka 容器抢宿主端口 19092。已停用(docker update --restart=no + stop;2026-09-12 尸体已删除腾名)。
# 当前本地开发 Kafka = go-sms-kafka(2026-09-12 由 hwsms-kafka 更名;手动 docker run,非 compose 管理;
# 2026-09-13 网络迁移:历史遗留网 hwsms-dev_default→go-sms-dev_default,显式别名 kafka 随迁
# (KRaft QUORUM_VOTERS=1@kafka:29093 依赖它,解析 172.22.0.8),宿主 19092,
# 网关 configs/config.dev.yaml brokers=127.0.0.1:19092 即它;Day45/46 冒烟均其承载。
# 注意:advertised=127.0.0.1:19092 为宿主观,容器内 CLI 连不通属预期,验证须走 host 网络容器(scripts/init_kafka_topics.sh 同法)。
# 若未来恢复本服务:①取消下段注释 ②QUORUM_VOTERS 主机名与网络别名对齐 ③换宿主端口避免冲突。
# ℹ️ Kafka 说明(2026-11-13 S1-B 恢复 compose 管理):
# POC 老机该服务曾因容器脱离 compose 网络→QUORUM_VOTERS 解析失败而停用(2026-09-11
# Day47 运维处置,事故史见归档仓 docs/runbook/incident-response.md),期间以手动
# docker run 的 go-sms-kafka 承载。新机( debian13-smsDev )为干净网络,自 S1-B 起
# 恢复由本 compose 管理:服务名 kafka 即网络别名,QUORUM_VOTERS=1@kafka:29093 可解析。
# advertised=127.0.0.1:19092 为宿主观(configs/config.dev.yaml brokers 同址),
# 容器内 CLI 连不通属预期,验证须走 host 网络容器(scripts/init_kafka_topics.sh 同法)。
# ============================================================
name: go-sms-dev
@@ -43,7 +40,15 @@ services:
# Day66 单表化后再调 6G:百万级/天写入冲刷 4G 池导致二级索引页换出,
# 冷查询 HDD 随机回表 29.5s;6G 给 detail 索引+数据页更多常驻余量
# (表+索引已 4.5G+;宿主 31G,在线 SET GLOBAL 生效,此处持久化)。
- --innodb-buffer-pool-size=6G
# 2026-11-13 S1-B SSD 基线(重建设计规范 §五):新机 /data 为 SSD(fio 4K
# 随机读 76.3k IOPS,老机 HDD 500 倍级),6G→10G(31G 内存容器组预算:
# MySQL 10G+CH 2G+Kafka 1G+Redis 0.5G+监控四件 1G≈14.5G,宿主留 16G 给
# Go 服务/编译/vscode-server);io_capacity 2000/10000 与 redo 1G 同步到位。
# 老机 HDD 三轮调参史(4G→6G)自此翻页,压测后(S5)再校。
- --innodb-buffer-pool-size=10G
- --innodb-io-capacity=2000
- --innodb-io-capacity-max=10000
- --innodb-redo-log-capacity=1G
ports:
- "13306:3306"
volumes:
@@ -76,42 +81,45 @@ services:
retries: 30
start_period: 10s
# ---- kafka 服务已于 2026-09-11 停用(见文件头说明;恢复时取消注释并修 QUORUM_VOTERS/端口) ----
# kafka:
# image: apache/kafka:3.7.0
# container_name: go-sms-kafka
# restart: unless-stopped
# environment:
# # ---- KRaft 单节点模式(无 ZooKeeper)----
# CLUSTER_ID: "5L6g3nShT-eMCtK--X86sw"
# KAFKA_NODE_ID: 1
# KAFKA_PROCESS_ROLES: broker,controller
# KAFKA_CONTROLLER_QUORUM_VOTERS: 1@kafka:29093
# # ---- 监听器:单一 listener,advertise 宿主机地址 ----
# # 消费者组需经 metadata 返回的 endpoint 连接 Coordinator,
# # 故统一广播 127.0.0.1:19092(宿主机端口映射 19092 -> 29092)
# KAFKA_LISTENERS: PLAINTEXT://0.0.0.0:29092,CONTROLLER://0.0.0.0:29093
# KAFKA_ADVERTISED_LISTENERS: PLAINTEXT://127.0.0.1:19092
# KAFKA_LISTENER_SECURITY_PROTOCOL_MAP: PLAINTEXT:PLAINTEXT,CONTROLLER:PLAINTEXT
# KAFKA_CONTROLLER_LISTENER_NAMES: CONTROLLER
# KAFKA_INTER_BROKER_LISTENER_NAME: PLAINTEXT
# # ---- 单节点副本数必须为 1 ----
# KAFKA_OFFSETS_TOPIC_REPLICATION_FACTOR: 1
# KAFKA_TRANSACTION_STATE_LOG_REPLICATION_FACTOR: 1
# KAFKA_TRANSACTION_STATE_LOG_MIN_ISR: 1
# KAFKA_GROUP_INITIAL_REBALANCE_DELAY_MS: 0
# KAFKA_AUTO_CREATE_TOPICS_ENABLE: "true"
# KAFKA_LOG_DIRS: /var/lib/kafka/data
# ports:
# - "19092:29092"
# volumes:
# - kafka-data:/var/lib/kafka/data
# healthcheck:
# test: ["CMD-SHELL", "bash -c '</dev/tcp/127.0.0.1/29092' || exit 1"]
# interval: 10s
# timeout: 5s
# retries: 30
# start_period: 60s
kafka:
image: apache/kafka:3.7.0
container_name: go-sms-kafka
restart: unless-stopped
environment:
# ---- KRaft 单节点模式(无 ZooKeeper);S1-B 恢复 compose 管理,事故史见文件头 ----
CLUSTER_ID: "5L6g3nShT-eMCtK--X86sw"
KAFKA_NODE_ID: 1
KAFKA_PROCESS_ROLES: broker,controller
KAFKA_CONTROLLER_QUORUM_VOTERS: 1@kafka:29093
# ---- 监听器:单一 listener,advertise 宿主机地址 ----
# 消费者组需经 metadata 返回的 endpoint 连接 Coordinator,
# 故统一广播 127.0.0.1:19092(宿主机端口映射 19092 -> 29092)
KAFKA_LISTENERS: PLAINTEXT://0.0.0.0:29092,CONTROLLER://0.0.0.0:29093
KAFKA_ADVERTISED_LISTENERS: PLAINTEXT://127.0.0.1:19092
KAFKA_LISTENER_SECURITY_PROTOCOL_MAP: PLAINTEXT:PLAINTEXT,CONTROLLER:PLAINTEXT
KAFKA_CONTROLLER_LISTENER_NAMES: CONTROLLER
KAFKA_INTER_BROKER_LISTENER_NAME: PLAINTEXT
# ---- 单节点副本数必须为 1 ----
KAFKA_OFFSETS_TOPIC_REPLICATION_FACTOR: 1
KAFKA_TRANSACTION_STATE_LOG_REPLICATION_FACTOR: 1
KAFKA_TRANSACTION_STATE_LOG_MIN_ISR: 1
KAFKA_GROUP_INITIAL_REBALANCE_DELAY_MS: 0
KAFKA_AUTO_CREATE_TOPICS_ENABLE: "true"
KAFKA_LOG_DIRS: /var/lib/kafka/data
# ---- 2026-11-13 S1-B SSD 基线(设计规范 §五):flush 放宽,掉电容忍靠
# page cache + 副本兜底(dev 单节点与 MySQL 2/0 同口径) ----
KAFKA_LOG_FLUSH_INTERVAL_MESSAGES: "100000"
KAFKA_LOG_FLUSH_INTERVAL_MS: "10000"
ports:
- "19092:29092"
volumes:
- kafka-data:/var/lib/kafka/data
healthcheck:
test: ["CMD-SHELL", "bash -c '</dev/tcp/127.0.0.1/29092' || exit 1"]
interval: 10s
timeout: 5s
retries: 30
start_period: 60s
# ---- 监控告警栈(Day 53 · runbook §6.2 P0) ----
# Prometheus 抓宿主网关 /metrics(host.docker.internal:host-gateway 映射);