openGauss数据库自2020年6月30日开源至今已有10个月了，在这短短的10个月内，openGauss社区用户下载量已达13W+、issue合并2000+、发行商业版本6个。仅3月份就有11家企业完成CLA签署，包括虚谷伟业、云和恩墨、优炫软件、海量数据、柏睿数据、快立方，电信天翼云、东方通、宝兰德、新数科技、深信服，正式加入openGauss社区，越来越多的力量参与到社区建设中。4月24日，openGauss社区理事会筹备会议在深圳大学城召开，邀请到国内著名的数据库技术方向上的多个公司、组织和机构，包括华为、招商银行、中国电信、云和恩墨、海量数据、人大金仓、神舟通用、虚谷伟业、快立方、亚信、超图软件、深信服、哈工大等机构参与，为openGauss社区开放治理迈出了新的一步。
openGauss数据库在经历1.0.0/1.0.1/1.1.0三个版本的迭代发布后，于2021年3月31日发布openGauss 2.0.0版本，这是openGauss社区发布的第一个Release版本。深度融合华为在数据库领域多年的经验，结合企业级场景需求，持续构建竞争力特性。
作为一名从Oracle DBA转openGauss相关工作的“IT攻城狮”，在遇到性能诊断时念念不忘的还是以前经常使用的AWR报告，通过这份报告，DBA可以较为全面的分析出数据库的性能问题所在范围、为下一步的数据库性能优化和故障诊断提供有力支撑。很高兴在openGauss数据库中也看到了类似的功能，那就是openGauss的WDR报告。
本文针对openGauss 2.0.0的WDR报告进行详细解读，帮助大家梳理WDR报告的数据来源以及相关含义，以便在openGauss数据库的性能诊断工作中游刃有余。关于性能调优操作方法，由于涉及内容较多，这里就不再复述，再写下去就该被老板炒鱿鱼了。openGauss数据库归根结底，它的本质还是数据库软件，主流的数据库调优方法在openGauss数据库中也基本适用，大家仁者见仁、智者见智，根据WDR报告结合自己已有的数据库调优方法，完全可以满足openGauss绝大多数的性能调优工作。
干货内容如下：
1. 启用WDR报告的snapshot收集
$ gs_guc reload -N all -I all -c "enable_wdr_snapshot=on"
postgres=# select name,setting from pg_settings where name like '%wdr%';
name | setting
-----------------------------+---------
enable_wdr_snapshot | on -- 开启数据库监控快照功能
wdr_snapshot_interval | 60 -- 后台Snapshot线程执行监控快照的时间间隔
wdr_snapshot_query_timeout | 100 -- 快照操作相关的sql语句的执行超时时间
wdr_snapshot_retention_days | 8 -- 系统中数据库监控快照数据的保留天数
2. WDR信息表
1> snapshot.snapshot 【记录当前系统中存储的WDR快照信息】
postgres=# \d snapshot.snapshot
Table "snapshot.snapshot"
Column | Type | Modifiers
-------------+--------------------------+-----------
snapshot_id | bigint | not null -- WDR快照序列号
start_ts | timestamp with time zone | -- WDR快照的开始时间
end_ts | timestamp with time zone | -- WDR快照的结束时间
2> snapshot.tables_snap_timestamp【记录所有表的WDR快照信息】
postgres=# \d snapshot.tables_snap_timestamp
Table "snapshot.tables_snap_timestamp"
Column | Type | Modifiers
-------------+--------------------------+-----------
snapshot_id | bigint | not null -- WDR快照序列号
db_name | text | -- WDR snapshot对应的database
tablename | text | -- WDR snasphot对应的table
start_ts | timestamp with time zone | -- WDR快照的开始时间
end_ts | timestamp with time zone | -- WDR快照的结束时间
3. WDR数据表
说明：WDR的数据表保存在snapshot这个schema下以snap_开头的表，其数据来源于dbe_perf这个schema内的视图
postgres=# select relname from pg_class where relname like '%snap_%';
----------------------------------------------------------------------------------------------------------------
snapshot.tables_snap_timestamp -- 记录所有存储的WDR快照中数据库、表对象、数据采集的开始、结束时间
snapshot.snapshot -- 记录当前系统中存储的WDR快照数据的索引信息、开始、结束时间
snapshot.snapshot_pkey -- snapshot.snapshot表的primary key
snapshot.snap_seq -- 序列
snapshot.snap_global_os_runtime -- 操作系统运行状态信息
snapshot.snap_global_os_threads -- 线程状态信息
snapshot.snap_global_instance_time -- 各种时间消耗信息(时间类型见instance_time视图)
snapshot.snap_summary_workload_sql_count -- 各数据库主节点的workload上的SQL数量分布
snapshot.snap_summary_workload_sql_elapse_time -- 数据库主节点上workload（业务）负载的SQL耗时信息
snapshot.snap_global_workload_transaction -- 各节点上的workload的负载信息
snapshot.snap_summary_workload_transaction -- 汇总的负载事务信息
snapshot.snap_global_thread_wait_status -- 工作线程以及辅助线程的阻塞等待情况
snapshot.snap_global_memory_node_detail -- 节点的内存使用情况
snapshot.snap_global_shared_memory_detail -- 共享内存上下文的使用情况
snapshot.snap_global_stat_db_cu -- 数据库的CU命中情况，可以通过gs_stat_reset()进行清零
snapshot.snap_global_stat_database -- 数据库的统计信息
snapshot.snap_summary_stat_database -- 汇总的数据库统计信息
snapshot.snap_global_stat_database_conflicts -- 数据库冲突状态的统计信息
snapshot.snap_summary_stat_database_conflicts -- 汇总的数据库冲突状态的统计信息
snapshot.snap_global_stat_bad_block -- 表、索引等文件的读取失败信息
snapshot.snap_summary_stat_bad_block -- 汇总的表、索引等文件的读取失败信息
snapshot.snap_global_file_redo_iostat -- Redo(WAL)相关统计信息
snapshot.snap_summary_file_redo_iostat -- 汇总的Redo(WAL)相关统计信息
snapshot.snap_global_rel_iostat -- 数据对象IO统计信息
snapshot.snap_summary_rel_iostat -- 汇总的数据对象IO统计信息
snapshot.snap_global_file_iostat -- 数据文件IO统计信息
snapshot.snap_summary_file_iostat -- 汇总的数据文件IO统计信息
snapshot.snap_global_replication_slots -- 复制节点的信息
snapshot.snap_global_bgwriter_stat -- 后端写进程活动的统计信息
snapshot.snap_global_replication_stat -- 日志同步状态信息
snapshot.snap_global_transactions_running_xacts -- 各节点运行事务的信息
snapshot.snap_summary_transactions_running_xacts -- 汇总各节点运行事务的信息
snapshot.snap_global_transactions_prepared_xacts -- 当前准备好进行两阶段提交的事务的信息
snapshot.snap_summary_transactions_prepared_xacts -- 汇总的当前准备好进行两阶段提交的事务的信息
snapshot.snap_summary_statement -- SQL语句的全量信息
snapshot.snap_global_statement_count -- 当前时刻执行的DML/DDL/DQL/DCL语句统计信息
snapshot.snap_summary_statement_count -- 汇总的当前时刻执行的DML/DDL/DQL/DCL语句统计信息
snapshot.snap_global_config_settings -- 数据库运行时参数信息
snapshot.snap_global_wait_events -- event等待相关统计信息
snapshot.snap_summary_user_login -- 用户登录和退出次数的相关信息
snapshot.snap_global_ckpt_status -- 实例的检查点信息和各类日志刷页情况
snapshot.snap_global_double_write_status -- 实例的双写文件的情况
snapshot.snap_global_pagewriter_status -- 实例的刷页信息和检查点信息
snapshot.snap_global_redo_status -- 实例的日志回放情况
snapshot.snap_global_rto_status -- 极致RTO状态信息
snapshot.snap_global_recovery_status -- 主机和备机的日志流控信息
snapshot.snap_global_threadpool_status -- 节点上的线程池中工作线程及会话的状态信息
snapshot.snap_statement_responsetime_percentile -- SQL响应时间P80、P95分布信息
snapshot.snap_global_statio_all_indexes -- 数据库中的每个索引行、显示特定索引的I/O的统计
snapshot.snap_summary_statio_all_indexes -- 汇总的数据库中的每个索引行、显示特定索引的I/O的统计
snapshot.snap_global_statio_all_sequences -- 数据库中每个序列的每一行、显示特定序列关于I/O的统计
snapshot.snap_summary_statio_all_sequences -- 汇总的数据库中每个序列的每一行、显示特定序列关于I/O的统计
snapshot.snap_global_statio_all_tables -- 数据库中每个表（包括TOAST表）的I/O的统计
snapshot.snap_summary_statio_all_tables -- 汇总的数据库中每个表（包括TOAST表）的I/O的统计
snapshot.snap_global_stat_all_indexes -- 数据库中的每个索引行，显示访问特定索引的统计
snapshot.snap_summary_stat_all_indexes -- 汇总的数据库中的每个索引行，显示访问特定索引的统计
snapshot.snap_summary_stat_user_functions -- 汇总的数据库节点用户自定义函数的相关统计信息
snapshot.snap_global_stat_user_functions -- 用户所创建的函数的状态的统计信息
snapshot.snap_global_stat_all_tables -- 每个表的一行（包括TOAST表）的统计信息
snapshot.snap_summary_stat_all_tables -- 汇总的每个表的一行（包括TOAST表）的统计信息
snapshot.snap_class_vital_info -- 校验相同的表或者索引的Oid是否一致
snapshot.snap_global_record_reset_time -- 重置（重启，主备倒换，数据库删除）openGauss统计信息时间
snapshot.snap_summary_statio_indexes_name -- 表snap_summary_statio_all_indexes的索引
snapshot.snap_summary_statio_tables_name -- 表snap_summary_statio_all_tables的索引
snapshot.snap_summary_stat_indexes_name -- 表snap_summary_stat_all_indexes的索引
snapshot.snap_class_info_name -- 表snap_class_vital_info的索引
(66 rows)
----------------------------------------------------------------------------------------------------------------
4. WDR报告创建
4.1 创建snapshot
-- 当开启enable_wdr_snapshot参数时，数据库默认每小时自动执行一次snapshot操作。
-- 当然特定情况下，也可以手动使用函数创建snapshot，如：select create_wdr_snapshot();
postgres=# select * from snapshot.snapshot offset 20;
snapshot_id | start_ts | end_ts
-------------+-------------------------------+-------------------------------
21 | 2021-04-21 05:59:09.337877+08 | 2021-04-21 05:59:10.249162+08
22 | 2021-04-21 06:59:10.3209+08 | 2021-04-21 06:59:11.229808+08
23 | 2021-04-21 07:59:10.426882+08 | 2021-04-21 07:59:11.340277+08
24 | 2021-04-21 08:59:10.534251+08 | 2021-04-21 08:59:11.447762+08
25 | 2021-04-21 09:59:11.448225+08 | 2021-04-21 09:59:26.121124+08
4.2 查询数据库节点信息
postgres=# select * from pg_node_env;
node_name | host | process | port | installpath | datapath | log_directory
-----------+--------------+---------+-------+--------------+-------------------+---------------------------------
dn_6001 | 192.168.0.99 | 9442 | 26000 | /gaussdb/app | /gaussdb/data/db1 | /gaussdb/log/omm/pg_log/dn_6001
4.3 创建WDR Report[使用gsql客户端生成]
postgres=# \a \t \o WDR_20210421.html -- 打开格式化输出，输出WDR报告：WDR_20210421.html
postgres=# select generate_wdr_report(24,25,'all','node','dn_6001'); -- 生成WDR报告
postgres=# \o \a \t -- 关闭格式化输出
函数说明：generate_wdr_report()
-- 语法
select generate_wdr_report(begin_snap_id bigint, end_snap_id bigint, report_type cstring, report_scope cstring, node_name cstring);
-- 选项：
begin_snap_id：查询时间段开始的snapshot的id（表snapshot.snaoshot中的snapshot_id）
end_snap_id： 查询时间段结束snapshot的id。默认end_snap_id大于begin_snap_id（表snapshot.snaoshot中的snapshot_id）
report_type： 指定生成report的类型。例如，summary/detail/all，其中：summary[汇总数据]/detail[明细数据]/all[包含summary和detail]
report_scope： 指定生成report的范围，可以为cluster或者node，其中：cluster是数据库级别的信息，node是节点级别的信息。
node_name： 当report_scope指定为node时，需要把该参数指定为对应节点的名称。当report_scope为cluster时，该值可以省略或者指定为空或NULL。node[节点名称]、cluster[省略/空/NULL]
5. WDR报告解读
说明：为了使得WDR报告内容不空洞，本次在测试环境使用BenchmarkSQL对openGauss数据库进行压力测试。 本次解读的WDR报告样例来自于此时采集的snapshot数据。
解读：
这一部分是WDR报告的概况信息，从这一部分我们能得到如下信息：
| 信息分类 | 信息描述 |
| 报告采集类型 | Summary + Detail，即汇总数据+明细数据 |
| Snapshot信息 | 使用snapshot_id为24和25的快照采集2021-04-21(08:59 ~ 09:59)的运行信息 |
| 硬件配置 | 1*1c/4g |
| 节点名 | dn_6001 |
| openGauss版本 | openGauss 2.0.0 |
相关代码:
第一部分，Report Type/Report Scope/Report Node内容来源于执行generate_wdr_report函数时输入的参数，详见源码“GenReport::ShowReportType(report_params* params)”
第二部分查询SQL：(变量ld-->snapshot_id)
select snapshot_id as "Snapshot Id",
to_char(start_ts, 'YYYY-MM-DD HH24:MI:SS') as "Start Time",
to_char(end_ts, 'YYYY-MM-DD HH24:MI:SS') as "End Time"
from snapshot.snapshot
where snapshot_id = %ld or snapshot_id = %ld;
第三部分查询SQL：(变量ld-->snapshot_id)
select 'CPUS', x.snap_value
from (select * from pg_node_env) t,
(select * from snapshot.snap_global_os_runtime) x
where x.snap_node_name = t.node_name
and x.snapshot_id = %ld
and (x.snap_name = 'NUM_CPUS');
select 'CPU Cores', x.snap_value
from (select * from pg_node_env) t,
(select * from snapshot.snap_global_os_runtime) x
where x.snap_node_name = t.node_name
and x.snapshot_id = %ld
and x.snap_name = 'NUM_CPU_CORES';
select 'CPU Sockets', x.snap_value
from (select * from pg_node_env) t,
(select * from snapshot.snap_global_os_runtime) x
where x.snap_node_name = t.node_name
and x.snapshot_id = %ld
and x.snap_name = 'NUM_CPU_SOCKETS';
select 'Physical Memory', pg_size_pretty(x.snap_value)
from (select * from pg_node_env) t,
(select * from snapshot.snap_global_os_runtime) x
where x.snap_node_name = t.node_name
and x.snapshot_id = %ld
and x.snap_name = 'PHYSICAL_MEMORY_BYTES';
select node_name as "Host Node Name" from pg_node_env;
select version() as "openGauss Version";
解读：
这一部分是实例的效率百分比，目标值是100%，即越接近100%，数据库运行越健康。
Buffer Hit： 即数据库请求的数据在buffer中命中的比例，该指标越高代表openGauss在buffer中查询到目标数据的概率越高，数据读取性能越好。
Effective CPU： 即有效的CPU使用比例，该指标偏小则说明CPU的有效使用偏低，处于等待状态的比例可能较高。
WalWrite NoWait： 即WAL日志写入时不等待的比例，该指标接近100%，说明buffer容量充足，可以满足WAL写操作的需求，若指标值偏小则可能需要调大buffer容量。
Soft Parse： 即SQL软解析的比例，该指标接近100%，说明当前执行的SQL基本都可以在Buffer中找到，若指标值偏小则说明存在大量硬解析，需要分析原因，对DML语句进行适度优化。
Non-Parse CPU： 即非解析占用的CPU比例，该指标接近100%，说明SQL解析并没有占用较多的CPU时间。
相关代码：
-- 变量：ld指的是snapshot_id，手动执行以下SQL语句时请自行替换对应的snapshot_id
select
unnest(array['Buffer Hit %%', 'Effective CPU %%', 'WalWrite NoWait %%', 'Soft Parse %%', 'Non-Parse CPU %%']) as "Metric Name",
unnest(array[case when s1.all_reads = 0 then 1 else round(s1.blks_hit * 100 / s1.all_reads) end, s2.cpu_to_elapsd, s3.walwrite_nowait, s4.soft_parse, s5.non_parse]) as "Metric Value"
from
(select (snap_2.all_reads - coalesce(snap_1.all_reads, 0)) as all_reads,
(snap_2.blks_hit - coalesce(snap_1.blks_hit, 0)) as blks_hit
from
(select sum(coalesce(snap_blks_read, 0) + coalesce(snap_blks_hit, 0)) as all_reads,
coalesce(sum(snap_blks_hit), 0) as blks_hit
from snapshot.snap_summary_stat_database
where snapshot_id = %ld) snap_1,
(select sum(coalesce(snap_blks_read, 0) + coalesce(snap_blks_hit, 0)) as all_reads,
coalesce(sum(snap_blks_hit), 0) as blks_hit
from snapshot.snap_summary_stat_database
where snapshot_id = %ld) snap_2
) s1,
(select round(cpu_time.snap_value * 100 / greatest(db_time.snap_value, 1)) as cpu_to_elapsd
from
(select coalesce(snap_2.snap_value, 0) - coalesce(snap_1.snap_value, 0) as snap_value
from
(select snap_stat_name, snap_value from snapshot.snap_global_instance_time
where snapshot_id = %ld and snap_stat_name = 'CPU_TIME') snap_1,
(select snap_stat_name, snap_value from snapshot.snap_global_instance_time
where snapshot_id = %ld and snap_stat_name = 'CPU_TIME') snap_2) cpu_time,
(select coalesce(snap_2.snap_value, 0) - coalesce(snap_1.snap_value, 0) as snap_value
from
(select snap_stat_name, snap_value from snapshot.snap_global_instance_time
where snapshot_id = %ld and snap_stat_name = 'DB_TIME') snap_1,
(select snap_stat_name, snap_value from snapshot.snap_global_instance_time
where snapshot_id = %ld and snap_stat_name = 'DB_TIME') snap_2) db_time
) s2,
(select (bufferAccess.snap_wait - bufferFull.snap_wait) * 100 / greatest(bufferAccess.snap_wait, 1) as walwrite_nowait
from
(select coalesce(snap_2.snap_wait) - coalesce(snap_1.snap_wait, 0) as snap_wait
from
(select snap_wait from snapshot.snap_global_wait_events
where snapshot_id = %ld and snap_event = 'WALBufferFull') snap_1,
(select snap_wait from snapshot.snap_global_wait_events
where snapshot_id = %ld and snap_event = 'WALBufferFull') snap_2) bufferFull,
(select coalesce(snap_2.snap_wait) - coalesce(snap_1.snap_wait, 0) as snap_wait
from
(select snap_wait from snapshot.snap_global_wait_events
where snapshot_id = %ld and snap_event = 'WALBufferAccess') snap_1,
(select snap_wait from snapshot.snap_global_wait_events
where snapshot_id = %ld and snap_event = 'WALBufferAccess') snap_2) bufferAccess
) s3,
(select round((snap_2.soft_parse - snap_1.soft_parse) * 100 / greatest((snap_2.hard_parse + snap_2.soft_parse)-(snap_1.hard_parse + snap_1.soft_parse), 1)) as soft_parse
from
(select sum(snap_n_soft_parse) as soft_parse, sum(snap_n_hard_parse) as hard_parse from snapshot.snap_summary_statement
where snapshot_id = %ld ) snap_1,
(select sum(snap_n_soft_parse) as soft_parse, sum(snap_n_hard_parse) as hard_parse from snapshot.snap_summary_statement
where snapshot_id = %ld ) snap_2
) s4,
(select round((snap_2.elapse_time - snap_1.elapse_time) * 100 /greatest((snap_2.elapse_time + snap_2.parse_time)-(snap_1.elapse_time + snap_1.parse_time), 1)) as non_parse
from
(select sum(snap_total_elapse_time) as elapse_time, sum(snap_parse_time) as parse_time from snapshot.snap_summary_statement
where snapshot_id = %ld ) snap_1,
(select sum(snap_total_elapse_time) as elapse_time, sum(snap_parse_time) as parse_time from snapshot.snap_summary_statement
where snapshot_id = %ld ) snap_2
) s5;
解读：
这一部分列出了数据库Top 10的等待事件、等待次数、总等待时间、平均等待时间、等待事件类型。
等待事件主要分为等待状态、等待轻量级锁、等待IO、等待事务锁这4类，详见下表所示：
- 等待状态列表
当wait_status为acquire lwlock、acquire lock或者wait io时，表示有等待事件。
正在等待获取wait_event列对应类型的轻量级锁、事务锁，或者正在进行IO。
- 轻量级锁等待事件列表
当wait_status值为acquire lwlock（轻量级锁）时对应的wait_event等待事件类型即为轻量级锁等待。
wait_event为extension时，表示此时的轻量级锁是动态分配的锁，未被监控。
|
wait_event类型 | |
|---|---|
- IO等待事件列表
当wait_status值为wait io时对应的wait_event等待事件类型即为IO等待事件。
- 事务锁等待事件列表
当wait_status值为acquire lock（事务锁）时对应的wait_event等待事件类型为事务锁等待事件。
| wait_event类型 | 类型描述 |
|---|---|
| relation | 对表加锁 |
| extend | 对表扩展空间时加锁 |
| partition | 对分区表加锁 |
| partition_seq | 对分区表的分区加锁 |
| page | 对表页面加锁 |
| tuple | 对页面上的tuple加锁 |
| transactionid | 对事务ID加锁 |
| virtualxid | 对虚拟事务ID加锁 |
| object | 加对象锁 |
| cstore_freespace | 对列存空闲空间加锁 |
| userlock | 加用户锁 |
| advisory | 加advisory锁 |
相关代码：
-- 说明：变量ld-->snapshot_id, 变量s-->node_name
select snap_event as "Event",
snap_wait as "Waits",
snap_total_wait_time as "Total Wait Times(us)",
round(snap_total_wait_time/snap_wait) as "Wait Avg(us)",
snap_type as "Wait Class"
from (
select snap_2.snap_event as snap_event,
snap_2.snap_type as snap_type,
snap_2.snap_wait - snap_1.snap_wait as snap_wait,
snap_2.total_time - snap_1.total_time as snap_total_wait_time
from
(select snap_event,
snap_wait,
snap_total_wait_time as total_time,
snap_type
from snapshot.snap_global_wait_events
where snapshot_id = %ld
and snap_event != 'none'
and snap_event != 'wait cmd'
and snap_event != 'unknown_lwlock_event'
and snap_nodename = '%s') snap_1,
(select snap_event,
snap_wait,
snap_total_wait_time as total_time,
snap_type
from snapshot.snap_global_wait_events
where snapshot_id = %ld
and snap_event != 'none'
and snap_event != 'wait cmd'
and snap_event != 'unknown_lwlock_event'
and snap_nodename = '%s') snap_2
where snap_2.snap_event = snap_1.snap_event
order by snap_total_wait_time desc limit 10)
where snap_wait != 0;
解读：
这一部分按照等待类型（STATUS、IO_EVENT、LWLOCK_EVENT、LOCK_EVENT），分类统计等待次数、总等待时间、平均等待时间。
相关代码：
-- 说明： 变量ld-->snapshot_id, 变量s-->node_name
select
snap_2.type as "Wait Class",
(snap_2.wait - snap_1.wait) as "Waits",
(snap_2.total_wait_time - snap_1.total_wait_time) as "Total Wait Time(us)",
round((snap_2.total_wait_time - snap_1.total_wait_time) / greatest((snap_2.wait - snap_1.wait), 1)) as "Wait Avg(us)"
from
(select
snap_type as type,
sum(snap_total_wait_time) as total_wait_time,
sum(snap_wait) as wait from snapshot.snap_global_wait_events
where snapshot_id = %ld
and snap_nodename = '%s'
and snap_event != 'unknown_lwlock_event'
and snap_event != 'none'
group by snap_type) snap_2
left join
(select
snap_type as type,
sum(snap_total_wait_time) as total_wait_time,
sum(snap_wait) as wait
from snapshot.snap_global_wait_events
where snapshot_id = %ld
and snap_nodename = '%s'
and snap_event != 'unknown_lwlock_event' and snap_event != 'none'
group by snap_type) snap_1
on snap_2.type = snap_1.type
order by "Total Wait Time(us)" desc;
解读：
这一部分主机CPU的负载情况：CPU的平均负载、用户使用占比、系统使用占比、IO等待占比、空闲占比。
相关SQL：
select
snap_2.cpus as "Cpus",
snap_2.cores as "Cores",
snap_2.sockets as "Sockets",
snap_1.load as "Load Average Begin",
snap_2.load as "Load Average End",
round(coalesce((snap_2.user_time - snap_1.user_time), 0) / greatest(coalesce((snap_2.total_time - snap_1.total_time), 0), 1) * 100, 2) as "%User",
round(coalesce((snap_2.sys_time - snap_1.sys_time), 0) / greatest(coalesce((snap_2.total_time - snap_1.total_time), 0), 1) * 100, 2) as "%System",
round(coalesce((snap_2.iowait_time - snap_1.iowait_time), 0) / greatest(coalesce((snap_2.total_time - snap_1.total_time), 0), 1) * 100, 2) as "%WIO",
round(coalesce((snap_2.idle_time - snap_1.idle_time), 0) / greatest(coalesce((snap_2.total_time - snap_1.total_time), 0), 1) * 100, 2) as "%Idle"
from
(select H.cpus, H.cores, H.sockets, H.idle_time, H.user_time, H.sys_time, H.iowait_time,
(H.idle_time + H.user_time + H.sys_time + H.iowait_time) AS total_time, H.load from
(select C.cpus, E.cores, T.sockets, I.idle_time, U.user_time, S.sys_time, W.iowait_time, L.load from
(select snap_value as cpus from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'NUM_CPUS')) AS C,
(select snap_value as cores from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'NUM_CPU_CORES')) AS E,
(select snap_value as sockets from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'NUM_CPU_SOCKETS')) AS T,
(select snap_value as idle_time from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'IDLE_TIME')) AS I,
(select snap_value as user_time from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'USER_TIME')) AS U,
(select snap_value as sys_time from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'SYS_TIME')) AS S,
(select snap_value as iowait_time from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'IOWAIT_TIME')) AS W,
(select snap_value as load from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'LOAD')) AS L ) as H ) as snap_2,
(select H.cpus, H.cores, H.sockets, H.idle_time, H.user_time, H.sys_time, H.iowait_time,
(H.idle_time + H.user_time + H.sys_time + H.iowait_time) AS total_time, H.load from
(select C.cpus, E.cores, T.sockets, I.idle_time, U.user_time, S.sys_time, W.iowait_time, L.load from
(select snap_value as cpus from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'NUM_CPUS')) AS C,
(select snap_value as cores from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'NUM_CPU_CORES')) AS E,
(select snap_value as sockets from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'NUM_CPU_SOCKETS')) AS T,
(select snap_value as idle_time from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'IDLE_TIME')) AS I,
(select snap_value as user_time from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'USER_TIME')) AS U,
(select snap_value as sys_time from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'SYS_TIME')) AS S,
(select snap_value as iowait_time from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'IOWAIT_TIME')) AS W,
(select snap_value as load from snapshot.snap_global_os_runtime
where (snapshot_id = %ld and snap_node_name = '%s' and snap_name = 'LOAD')) AS L ) as H ) as snap_1 ;
解读：
这一部分描述了openGauss在快照期间的IO负载情况。
Database requests： 即每秒IO请求次数，包括请求次数总和、读请求次数、写请求次数.
Database(blocks)： 即每秒block请求数量，包含请求的block总和数量、读block的数量和写block的数量.
Database(MB): 即将block换算成容量(MB)[如：blocks * 8/1024]，增加数据的可读性。
Redo requests和Redo(MB) 分别表示每秒redo的写请求次数和redo写的数据量。
相关代码：
-- 由于源码中相关SQL融合了C++程序语法，像我这种不做开发的DBA读起来有些难以理解【如：(phyblkwrt * %d) >> 20 这个段没有很好理解】。
-- 但是依旧不影响我们对这些数据采集方法的理解，相关SQL如下：
-- 两个snapshot_id(24和25)期间，数据块的IO统计信息(数值除以3600即换算成以秒为单位的WDR数据)
postgres=# select
(snap_2.phytotal - snap_1.phytotal) as phytotal,
(snap_2.phyblktotal - snap_1.phyblktotal) as phyblktotal,
(snap_2.phyrds - snap_1.phyrds) as phyrds,
(snap_2.phyblkrd - snap_1.phyblkrd) as phyblkrd,
(snap_2.phywrts - snap_1.phywrts) as phywrts,
(snap_2.phyblkwrt - snap_1.phyblkwrt) as phyblkwrt
from
(select (snap_phyrds + snap_phywrts) as phytotal,
(snap_phyblkwrt + snap_phyblkrd) as phyblktotal,
snap_phyrds as phyrds, snap_phyblkrd as phyblkrd,
snap_phywrts as phywrts, snap_phyblkwrt as phyblkwrt
from snapshot.snap_global_rel_iostat
where snapshot_id = 24 and snap_node_name = 'dn_6001') snap_1,
(select (snap_phyrds + snap_phywrts) as phytotal,
(snap_phyblkwrt + snap_phyblkrd) as phyblktotal,
snap_phyrds as phyrds, snap_phyblkrd as phyblkrd,
snap_phywrts as phywrts, snap_phyblkwrt as phyblkwrt
from snapshot.snap_global_rel_iostat
where snapshot_id = 25 and snap_node_name = 'dn_6001') snap_2;
phytotal | phyblktotal | phyrds | phyblkrd | phywrts | phyblkwrt
----------+-------------+---------+----------+---------+-----------
4626892 | 4626892 | 2955639 | 2955639 | 1671253 | 1671253
-- 两个snapshot_id(24和25)期间，redo的统计信息(数值除以3600即换算成以秒为单位的WDR数据)
postgres=# select
(snap_2.phywrts - snap_1.phywrts) as phywrts,
(snap_2.phyblkwrt - snap_1.phyblkwrt) as phyblkwrt
from
(select sum(snap_phywrts) as phywrts, sum(snap_phyblkwrt) as phyblkwrt
from snapshot.snap_global_file_redo_iostat
where snapshot_id = 24 and snap_node_name = 'dn_6001') snap_1,
(select sum(snap_phywrts) as phywrts, sum(snap_phyblkwrt) as phyblkwrt
from snapshot.snap_global_file_redo_iostat
where snapshot_id = 25 and snap_node_name = 'dn_6001') snap_2;
phywrts | phyblkwrt
---------+-----------
132721 | 509414
解读：
这一部分描述了节点内存的变化信息,通过这些变化信息，我们可以了解到在两次快照期间，数据库的内存变化情况，作为数据库性能分析或异常分析的参考。数据来源于snapshot.snap_global_memory_node_detail。
这部分分别描述了： 内存的类型 以及 对应的起始大小和终止大小。
这里没有采集到数据的原因： 测试环境内存太小，导致启动时将memory protect关闭了，从而导致无法查询dbe_perf.global_memory_node_detail视图。 而WDR的内存统计数据(snapshot.snap_global_memory_node_detail)则来源于该视图。
另外，请确保disable_memory_protect=off。
关于这部分Memory Type常见类型如下：
| Memory 类型 | 说明 |
|---|---|
| max_process_memory | openGauss实例所占用的内存大小 |
| process_used_memory | 进程所使用的内存大小 |
| max_dynamic_memory | 最大动态内存 |
| dynamic_used_memory | 已使用的动态内存 |
| dynamic_peak_memory | 内存的动态峰值 |
| dynamic_used_shrctx | 最大动态共享内存上下文 |
| dynamic_peak_shrctx | 共享内存上下文的动态峰值 |
| max_shared_memory | 最大共享内存 |
| shared_used_memory | 已使用的共享内存 |
| max_cstore_memory | 列存所允许使用的最大内存 |
| cstore_used_memory | 列存已使用的内存大小 |
| max_sctpcomm_memory | sctp通信所允许使用的最大内存 |
| sctpcomm_used_memory | sctp通信已使用的内存大小 |
| sctpcomm_peak_memory | sctp通信的内存峰值 |
| other_used_memory | 其他已使用的内存大小 |
| gpu_max_dynamic_memory | GPU最大动态内存 |
| gpu_dynamic_used_memory | GPU已使用的动态内存 |
| gpu_dynamic_peak_memory | GPU内存的动态峰值 |
| pooler_conn_memory | 链接池申请内存计数 |
| pooler_freeconn_memory | 链接池空闲连接的内存计数 |
| storage_compress_memory | 存储模块压缩使用的内存大小 |
| udf_reserved_memory | UDF预留的内存大小 |
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id
select
snap_2.snap_memorytype as "Memory Type",
snap_1.snap_memorymbytes as "Begin(MB)",
snap_2.snap_memorymbytes as "End(MB)"
from
(select snap_memorytype, snap_memorymbytes
from
snapshot.snap_global_memory_node_detail
where (snapshot_id = %ld and snap_nodename = '%s')
and
(snap_memorytype = 'max_process_memory'
or snap_memorytype = 'process_used_memory'
or snap_memorytype = 'max_shared_memory'
or snap_memorytype = 'shared_used_memory'))
as snap_2
left join
(select snap_memorytype, snap_memorymbytes
from
snapshot.snap_global_memory_node_detail
where (snapshot_id = %ld and snap_nodename = '%s')
and (snap_memorytype = 'max_process_memory'
or snap_memorytype = 'process_used_memory'
or snap_memorytype = 'max_shared_memory'
or snap_memorytype = 'shared_used_memory'))
as snap_1
on snap_2.snap_memorytype = snap_1.snap_memorytype;
解读：
这一部分描述了数据库各种状态所消耗的时间，关于Stat Name的解释如下：
| Stat Name | 说明 |
|---|---|
| DB_TIME | 作业在多核下的有效时间花销 |
| CPU_TIME | CPU的时间花销 |
| EXECUTION_TIME | 执行器内的时间花销 |
| PARSE_TIME | SQL解析的时间花销 |
| PLAN_TIME | 生成Plan的时间花销 |
| REWRITE_TIME | SQL重写的时间花销 |
| PL_EXECUTION_TIME | plpgsql（存储过程）执行的时间花销 |
| PL_COMPILATION_TIME | plpgsql（存储过程）编译的时间花销 |
| NET_SEND_TIME | 网络上的时间花销 |
| DATA_IO_TIME | IO上的时间花销 |
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id
select t2.snap_node_name as "Node Name",
t2.snap_stat_name as "Stat Name",
(t2.snap_value - coalesce(t1.snap_value, 0)) as "Value(us)"
from
(select * from snapshot.snap_global_instance_time
where snapshot_id = %ld
and snap_node_name = '%s') t1
right join
(select * from snapshot.snap_global_instance_time
where snapshot_id = %ld
and snap_node_name = '%s') t2
on t1.snap_stat_name = t2.snap_stat_name
order by "Value(us)"
desc limit 200;
解读：
这一部分分别从SQL执行时间、SQL消耗CPU的时间、SQL返回的行数、SQL扫描的行数、SQL执行的次数、SQL物理读的次数、SQL逻辑读的次数等多维度对两次快照期间的SQL执行情况进行统计。
关于表中列的含义，如下所示：
| 列名称 | 备注 |
|---|---|
| Unique SQL Id | 归一化的SQL ID, 即SQL唯一标识 |
| User Name | 执行SQL的用户 |
| Logical Read | 逻辑读，即Buffer的块访问次数 |
| Calls | SQL的调用次数 |
| Min Elapse Time(us) | SQL在内核中的最小运行时间（单位：微秒） |
| Max Elapse Time(us) | SQL在内核中的最大运行时间（单位：微秒） |
| Total Elapse Time(us) | SQL在内核中的总运行时间 （单位：微秒） |
| Avg Elapse Time(us) | SQL在内核中的平均运行时间（单位：微秒） |
| Returned Rows | SELECT返回的结果集行数 |
| Tuples Read | SQL扫描的总行数(包括顺序扫描和随机扫描) |
| Tuples Affected | SQL删除的行数 |
| Physical Read | 物理读，即下盘读取block进入buffer的次数 |
| CPU Time(us) | SQL消耗的CPU时间（单位：微秒） |
| Data IO Time(us) | IO上的时间花费（单位：微秒） |
| Sort Count | SQL排序执行的次数 |
| Sort Time(us) | SQL排序执行的时间（单位：微秒） |
| Sort Mem Used(KB) | 排序过程中使用的work memory大小（单位：KB） |
| Sort Spill Count | 排序过程中发生落盘写文件的次数 |
| Sort Spill Size(KB) | 排序过程中发生落盘写文件的大小（单位：KB） |
| Hash Count | hash执行的次数 |
| Hash Time(us) | hash执行的时间（单位：微秒） |
| Hash Mem Used(KB) | hash过程中使用的work memory大小（单位：KB） |
| Hash Spill Count | hash过程中发生落盘写文件的次数 |
| Hash Spill Size(KB) | hash过程中发生落盘写文件的大小（单位：KB） |
| SQL Text | SQL语句内容 |
Tips：Top200显得有些冗余，多余的SQL信息并没有太大用处，反而降低了可读性，希望将来能优化到Top20,。
相关代码：
-- 由于多个SQL统计信息的SQL语句类似，这里仅列举SQL执行时间的统计SQL，其他的类似。
-- 说明：%s代表node_name，%ld代表snapshot_id
select t2.snap_unique_sql_id as "Unique SQL Id",
t2.snap_user_name as "User Name",
(t2.snap_total_elapse_time - coalesce(t1.snap_total_elapse_time, 0)) as "Total Elapse Time(us)",
(t2.snap_n_calls - coalesce(t1.snap_n_calls, 0)) as "Calls",
round("Total Elapse Time(us)"/greatest("Calls", 1), 0) as "Avg Elapse Time(us)",
t2.snap_min_elapse_time as "Min Elapse Time(us)",
t2.snap_max_elapse_time as "Max Elapse Time(us)",
(t2.snap_n_returned_rows - coalesce(t1.snap_n_returned_rows, 0)) as "Returned Rows",
((t2.snap_n_tuples_fetched - coalesce(t1.snap_n_tuples_fetched, 0)) +
(t2.snap_n_tuples_returned - coalesce(t1.snap_n_tuples_returned, 0))) as "Tuples Read",
((t2.snap_n_tuples_inserted - coalesce(t1.snap_n_tuples_inserted, 0)) +
(t2.snap_n_tuples_updated - coalesce(t1.snap_n_tuples_updated, 0)) +
(t2.snap_n_tuples_deleted - coalesce(t1.snap_n_tuples_deleted, 0))) as "Tuples Affected",
(t2.snap_n_blocks_fetched - coalesce(t1.snap_n_blocks_fetched, 0)) as "Logical Read",
((t2.snap_n_blocks_fetched - coalesce(t1.snap_n_blocks_fetched, 0)) -
(t2.snap_n_blocks_hit - coalesce(t1.snap_n_blocks_hit, 0))) as "Physical Read",
(t2.snap_cpu_time - coalesce(t1.snap_cpu_time, 0)) as "CPU Time(us)",
(t2.snap_data_io_time - coalesce(t1.snap_data_io_time, 0)) as "Data IO Time(us)",
(t2.snap_sort_count - coalesce(t1.snap_sort_count, 0)) as "Sort Count",
(t2.snap_sort_time - coalesce(t1.snap_sort_time, 0)) as "Sort Time(us)",
(t2.snap_sort_mem_used - coalesce(t1.snap_sort_mem_used, 0)) as "Sort Mem Used(KB)",
(t2.snap_sort_spill_count - coalesce(t1.snap_sort_spill_count, 0)) as "Sort Spill Count",
(t2.snap_sort_spill_size - coalesce(t1.snap_sort_spill_size, 0)) as "Sort Spill Size(KB)",
(t2.snap_hash_count - coalesce(t1.snap_hash_count, 0)) as "Hash Count",
(t2.snap_hash_time - coalesce(t1.snap_hash_time, 0)) as "Hash Time(us)",
(t2.snap_hash_mem_used - coalesce(t1.snap_hash_mem_used, 0)) as "Hash Mem Used(KB)",
(t2.snap_hash_spill_count - coalesce(t1.snap_hash_spill_count, 0)) as "Hash Spill Count",
(t2.snap_hash_spill_size - coalesce(t1.snap_hash_spill_size, 0)) as "Hash Spill Size(KB)",
LEFT(t2.snap_query, 25) as "SQL Text"
from
(select * from snapshot.snap_summary_statement where snapshot_id = %ld and snap_node_name = '%s') t1
right join
(select * from snapshot.snap_summary_statement where snapshot_id = %ld and snap_node_name = '%s') t2
on t1.snap_unique_sql_id = t2.snap_unique_sql_id
and t1.snap_user_id = t2.snap_user_id
order by "Total Elapse Time(us)"
desc limit 200;
解读：
这一部分分别从等待时长、等待次数这两个维度对等待事件进行统计。
表格中列的含义即就是列的英文翻译，这里就不再复述了。
具体的等待事件的介绍详见前文：“Top 10 Events by Total Wait Time”部分的内容，这里也不再复述。
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id，无论从哪个维度统计，基本的SQL语句差异不大，这里仅列举"by wait time"的SQL示例
select t2.snap_type as "Type",
t2.snap_event as "Event",
(t2.snap_total_wait_time - coalesce(t1.snap_total_wait_time, 0)) as "Total Wait Time (us)",
(t2.snap_wait - coalesce(t1.snap_wait, 0)) as "Waits",
(t2.snap_failed_wait - coalesce(t1.snap_failed_wait, 0)) as "Failed Waits",
(case "Waits"
when 0 then 0
else round("Total Wait Time (us)" / "Waits", 2)
end) as "Avg Wait Time (us)",
t2.snap_max_wait_time as "Max Wait Time (us)"
from
(select * from snapshot.snap_global_wait_events
where snapshot_id = %ld
and snap_nodename = '%s'
and snap_event != 'unknown_lwlock_event'
and snap_event != 'none') t1
right join
(select * from snapshot.snap_global_wait_events
where snapshot_id = %ld
and snap_nodename = '%s'
and snap_event != 'unknown_lwlock_event'
and snap_event != 'none') t2
on t1.snap_event = t2.snap_event
order by "Total Wait Time (us)"
desc limit 200;
解读：
这一部分根据Heap block的命中率排序统计用户表的IO活动状态。
数据来源于snapshot.snap_global_statio_all_indexes表和snapshot.snap_global_statio_all_tables表。
该表相关列的介绍如下：
| 列名 | 描述 |
|---|---|
| DB Name | 数据库名 |
| Schema Name | 模式名 |
| Table Name | 表名 |
| %Heap Blks Hit Ratio | 数据块读取缓存命中率=heap_blks_hit/(heap_blks_read+heap_blks_hit)*100 |
| Heap Blks Read | 从该表中读取的磁盘块数 |
| Heap Blks Hit | 此表缓存命中数 |
| Idx Blks Read | 从表中所有索引读取的磁盘块数 |
| Idx Blks Hit | 表中所有索引命中缓存数 |
| Toast Blks Read | 此表的TOAST表读取的磁盘块数（如果存在） |
| Toast Blks Hit | 此表的TOAST表命中缓冲区数（如果存在） |
| Tidx Blks Read | 此表的TOAST表索引读取的磁盘块数（如果存在） |
| Tidx Blks Hit | 此表的TOAST表索引命中缓冲区数（如果存在） |
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id
SELECT table_io.db_name as "DB Name",
table_io.snap_schemaname as "Schema Name",
table_io.snap_relname as "Table Name",
table_io.heap_blks_hit_ratio as "%Heap Blks Hit Ratio",
table_io.heap_blks_read as "Heap Blks Read",
table_io.heap_blks_hit as "Heap Blks Hit",
idx_io.idx_blks_read as "Idx Blks Read",
idx_io.idx_blks_hit as "Idx Blks Hit",
table_io.toast_blks_read as "Toast Blks Read",
table_io.toast_blks_hit as "Toast Blks Hit",
table_io.tidx_blks_read as "Tidx Blks Read",
table_io.tidx_blks_hit as "Tidx Blks Hit"
FROM
(select t2.db_name, t2.snap_schemaname , t2.snap_relname ,
(case
when ((t2.snap_heap_blks_read - coalesce(t1.snap_heap_blks_read, 0)) + (t2.snap_heap_blks_hit - coalesce(t1.snap_heap_blks_hit, 0))) = 0
then 0
else round((t2.snap_heap_blks_hit - coalesce(t1.snap_heap_blks_hit, 0))/
((t2.snap_heap_blks_read - coalesce(t1.snap_heap_blks_read, 0)) + (t2.snap_heap_blks_hit - coalesce(t1.snap_heap_blks_hit, 0))) * 100, 2)
end ) as heap_blks_hit_ratio,
(t2.snap_heap_blks_read - coalesce(t1.snap_heap_blks_read, 0)) as heap_blks_read,
(t2.snap_heap_blks_hit - coalesce(t1.snap_heap_blks_hit, 0)) as heap_blks_hit,
(t2.snap_idx_blks_read - coalesce(t1.snap_idx_blks_read, 0)) as idx_blks_read,
(t2.snap_idx_blks_hit - coalesce(t1.snap_idx_blks_hit, 0)) as idx_blks_hit,
(t2.snap_toast_blks_read - coalesce(t1.snap_toast_blks_read, 0)) as toast_blks_read,
(t2.snap_toast_blks_hit - coalesce(t1.snap_toast_blks_hit, 0)) as toast_blks_hit,
(t2.snap_tidx_blks_read - coalesce(t1.snap_tidx_blks_read, 0)) as tidx_blks_read,
(t2.snap_tidx_blks_hit - coalesce(t1.snap_tidx_blks_hit, 0)) as tidx_blks_hit from
(select * from snapshot.snap_global_statio_all_tables
where snapshot_id = %ld and snap_node_name = '%s'
and snap_schemaname NOT IN ('pg_catalog', 'information_schema', 'snapshot')
and snap_schemaname !~ '^pg_toast') t1
right join
(select * from snapshot.snap_global_statio_all_tables
where snapshot_id = %ld
and snap_node_name = '%s'
and snap_schemaname NOT IN ('pg_catalog', 'information_schema', 'snapshot')
and snap_schemaname !~ '^pg_toast') t2
on t1.snap_relid = t2.snap_relid
and t2.db_name = t1.db_name
and t2.snap_schemaname = t1.snap_schemaname ) as table_io
LEFT JOIN
(select t2.db_name , t2.snap_schemaname , t2.snap_relname,
(t2.snap_idx_blks_read - coalesce(t1.snap_idx_blks_read, 0)) as idx_blks_read,
(t2.snap_idx_blks_hit - coalesce(t1.snap_idx_blks_hit, 0)) as idx_blks_hit
from
(select * from snapshot.snap_global_statio_all_indexes
where snapshot_id = %ld
and snap_node_name = '%s'
and snap_schemaname NOT IN ('pg_catalog', 'information_schema', 'snapshot')
and snap_schemaname !~ '^pg_toast') t1
right join
(select * from snapshot.snap_global_statio_all_indexes
where snapshot_id = %ld
and snap_node_name = '%s'
and snap_schemaname NOT IN ('pg_catalog', 'information_schema', 'snapshot')
AND snap_schemaname !~ '^pg_toast') t2
on t1.snap_relid = t2.snap_relid
and t2.snap_indexrelid = t1.snap_indexrelid
and t2.db_name = t1.db_name and t2.snap_schemaname = t1.snap_schemaname) as idx_io
on table_io.db_name = idx_io.db_name
and table_io.snap_schemaname = idx_io.snap_schemaname
and table_io.snap_relname = idx_io.snap_relname
order by "%%Heap Blks Hit Ratio"
asc limit 200;
解读：
这一部分根据索引缓存命中率，统计用户索引IO活动信息。
数据来源于snapshot.snap_global_statio_all_indexes表。
相关列信息如下：
| 列名 | 介绍 |
|---|---|
| DB Name | 数据库名 |
| Schema Name | 模式名 |
| Table Name | 表名 |
| Index Name | 索引名 |
| %Idx Blks Hit Ratio | 索引缓冲命中率=“Idx Blks Hit”/(“Idx Blks Hit”+“Idx Blks Read”)*100 |
| Idx Blks Read | 从索引中读取的磁盘块数 |
| Idx Blks Hit | 索引命中缓存数 |
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id
select t2.db_name as "DB Name",
t2.snap_schemaname as "Schema Name",
t2.snap_relname as "Table Name",
t2.snap_indexrelname as "Index Name",
(case
when ((t2.snap_idx_blks_read - coalesce(t1.snap_idx_blks_read, 0)) + (t2.snap_idx_blks_hit - coalesce(t1.snap_idx_blks_hit, 0))) = 0 then 0
else
round((t2.snap_idx_blks_hit - coalesce(t1.snap_idx_blks_hit, 0))/((t2.snap_idx_blks_hit - coalesce(t1.snap_idx_blks_hit, 0)) +
(t2.snap_idx_blks_read - coalesce(t1.snap_idx_blks_read, 0))) * 100, 2)
end) as "%Idx Blks Hit Ratio",
(t2.snap_idx_blks_read - coalesce(t1.snap_idx_blks_read, 0)) as "Idx Blks Read",
(t2.snap_idx_blks_hit - coalesce(t1.snap_idx_blks_hit, 0)) as "Idx Blks Hit"
from
(select * from snapshot.snap_global_statio_all_indexes
where snapshot_id = %ld
and snap_node_name = '%s'
and snap_schemaname NOT IN ('pg_catalog', 'information_schema', 'snapshot')
and snap_schemaname !~ '^pg_toast') t1
right join
(select * from snapshot.snap_global_statio_all_indexes
where snapshot_id = %ld
and snap_node_name = '%s'
and snap_schemaname NOT IN ('pg_catalog', 'information_schema', 'snapshot')
and snap_schemaname !~ '^pg_toast') t2
on t1.snap_relid = t2.snap_relid
and t2.snap_indexrelid = t1.snap_indexrelid
and t2.db_name = t1.db_name
and t2.snap_schemaname = t1.snap_schemaname
order by "%Idx Blks Hit Ratio"
asc limit 200;
解读：
此处存在缺陷，统计的数据有问题，个人认为SQL语句需要修改，详见“相关代码”。
这一部分描述的是后台写操作的统计信息，数据来源于snapshot.snap_global_bgwriter_stat表。
具体内容如下：
| 列名 | 数据获取相关函数 | 说明 |
|---|---|---|
| Checkpoints Timed | pg_stat_get_bgwriter_timed_checkpoints() | 执行的定期检查点数 |
| Checkpoints Require | pg_stat_get_bgwriter_requested_checkpoints() | 执行的需求检查点数 |
| Checkpoint Write Time(ms) | pg_stat_get_checkpoint_write_time() | 检查点操作中，文件写入到磁盘消耗的时间（单位：毫秒） |
| Checkpoint Sync Time(ms) | pg_stat_get_checkpoint_sync_time() | 检查点操作中，文件同步到磁盘消耗的时间（单位：毫秒） |
| Buffers Checkpoint | pg_stat_get_bgwriter_buf_written_checkpoints() | 检查点写缓冲区数量 |
| Buffers Clean | pg_stat_get_bgwriter_buf_written_clean() | 后端写线程写缓冲区数量 |
| Maxwritten Clean | pg_stat_get_bgwriter_maxwritten_clean() | 后端写线程停止清理Buffer的次数 |
| Buffers Backend | pg_stat_get_buf_written_backend() | 通过后端直接写缓冲区数 |
| Buffers Backend | Fsync pg_stat_get_buf_fsync_backend() | 后端不得不执行自己的fsync调用的时间数（通常后端写进程处理这些即使后端确实自己写） |
| Buffers Alloc | pg_stat_get_buf_alloc() 分配的缓冲区数量 | |
| Stats Reset | pg_stat_get_bgwriter_stat_reset_time() | 这些统计被重置的时间 |
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id
select (snap_2.snap_checkpoints_timed - coalesce(snap_1.snap_checkpoints_timed, 0)) AS "Checkpoints Timed",
(snap_2.snap_checkpoints_req - coalesce(snap_1.snap_checkpoints_req, 0)) AS "Checkpoints Require",
(snap_2.snap_checkpoint_write_time - coalesce(snap_1.snap_checkpoint_write_time, 0)) AS "Checkpoint Write Time(ms)",
(snap_2.snap_checkpoint_sync_time - coalesce(snap_1.snap_checkpoint_sync_time, 0)) AS "Checkpoint Sync Time(ms)",
(snap_2.snap_buffers_checkpoint - coalesce(snap_1.snap_buffers_checkpoint, 0)) AS "Buffers Checkpoint",
(snap_2.snap_buffers_clean - coalesce(snap_1.snap_buffers_clean, 0)) AS "Buffers Clean",
(snap_2.snap_maxwritten_clean - coalesce(snap_1.snap_maxwritten_clean, 0)) AS "Maxwritten Clean",
(snap_2.snap_buffers_backend - coalesce(snap_1.snap_buffers_backend, 0)) AS "Buffers Backend",
(snap_2.snap_buffers_backend_fsync - coalesce(snap_1.snap_buffers_backend_fsync, 0)) AS "Buffers Backend Fsync",
(snap_2.snap_buffers_alloc - coalesce(snap_1.snap_buffers_alloc, 0)) AS "Buffers Alloc",
to_char(snap_2.snap_stats_reset, 'YYYY-MM-DD HH24:MI:SS') AS "Stats Reset"
from
(select * from snapshot.snap_global_bgwriter_stat
where snapshot_id = %ld
and snap_node_name = '%s') snap_2
LEFT JOIN
(select * from snapshot.snap_global_bgwriter_stat
where snapshot_id = %ld
and snap_node_name = '%s') snap_1
on snap_2.snapshot_id = snap_1.snapshot_id --错误点：snap_2.snapshot_id = snap_1.snapshot_id ？ 这其实还是同一个snapshot
and snap_2.snap_node_name = snap_1.snap_node_name
and snap_2.snap_stats_reset = snap_1.snap_stats_reset
limit 200;
-- 统计信息应该是2次snapshot之间的数据，而以上SQL并不能正确输出相关数据。个人觉得可以删除LEFT JOIN连接。
-- 建议修改如下：
select (snap_2.snap_checkpoints_timed - coalesce(snap_1.snap_checkpoints_timed, 0)) AS "Checkpoints Timed",
(snap_2.snap_checkpoints_req - coalesce(snap_1.snap_checkpoints_req, 0)) AS "Checkpoints Require",
(snap_2.snap_checkpoint_write_time - coalesce(snap_1.snap_checkpoint_write_time, 0)) AS "Checkpoint Write Time(ms)",
(snap_2.snap_checkpoint_sync_time - coalesce(snap_1.snap_checkpoint_sync_time, 0)) AS "Checkpoint Sync Time(ms)",
(snap_2.snap_buffers_checkpoint - coalesce(snap_1.snap_buffers_checkpoint, 0)) AS "Buffers Checkpoint",
(snap_2.snap_buffers_clean - coalesce(snap_1.snap_buffers_clean, 0)) AS "Buffers Clean",
(snap_2.snap_maxwritten_clean - coalesce(snap_1.snap_maxwritten_clean, 0)) AS "Maxwritten Clean",
(snap_2.snap_buffers_backend - coalesce(snap_1.snap_buffers_backend, 0)) AS "Buffers Backend",
(snap_2.snap_buffers_backend_fsync - coalesce(snap_1.snap_buffers_backend_fsync, 0)) AS "Buffers Backend Fsync",
(snap_2.snap_buffers_alloc - coalesce(snap_1.snap_buffers_alloc, 0)) AS "Buffers Alloc",
to_char(snap_2.snap_stats_reset, 'YYYY-MM-DD HH24:MI:SS') AS "Stats Reset"
from
(select * from snapshot.snap_global_bgwriter_stat
where snapshot_id = %ld
and snap_node_name = '%s') snap_2,
(select * from snapshot.snap_global_bgwriter_stat
where snapshot_id = %ld
and snap_node_name = '%s') snap_1
limit 200;
解读：[本次实验环境是单机，没有复制槽数据]
这一部分描述的是复制槽的相关信息。数据来源于：snapshot.snap_global_replication_slots表。
信息内容如下所示：
| 列名 | 描述 |
|---|---|
| Slot Name | 复制槽名称 |
| Slot Type | 复制槽类型 |
| DB Name | 数据库名称 |
| Active | 是否为激活状态 |
| Xmin | 事务标识，最早的事务ID(txid) |
| Restart Lsn | 事务槽开始复制时的LSN信息，用来判断哪些事务需要被复制 |
| Dummy Standby | 是否为假备 |
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id
SELECT snap_slot_name as "Slot Name",
snap_slot_type as "Slot Type",
snap_database as "DB Name",
snap_active as "Active",
snap_x_min as "Xmin",
snap_restart_lsn as "Restart Lsn",
snap_dummy_standby as "Dummy Standby"
FROM snapshot.snap_global_replication_slots
WHERE snapshot_id = %ld
and snap_node_name = '%s'
limit 200;
解读：[本次实验环境是单机，没有复制槽数据]
这一部分描述事务槽详细的状态信息，数据源于snapshot.snap_global_replication_stat表。
信息内容如下所示：
| 列名 | 描述 |
|---|---|
| Thread Id | 线程ID |
| Usesys Id | 用户OID |
| Usename | 用户名 |
| Application Name | 应用程序名称 |
| Client Addr | 客户端地址 |
| Client Hostname | 客户端主机名 |
| Client Port | 客户端端口 |
| Backend Start | 程序启动时间 |
| State | 日志复制状态【追赶状态/一直的流状态】 |
| Sender Sent Location | 日志发送的位置 |
| Receiver Write Location | 日志接收端write的位置 |
| Receiver Flush Location | 日志接收端flush的位置 |
| Receiver Replay Location | 日志接收端replay的位置 |
| Sync Priority | 同步复制的优先级(0表示异步复制) |
| Sync State | 同步状态【异步复制/同步复制/潜在同步复制】 |
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id
SELECT snap_pid as "Thread Id",
snap_usesysid as "Usesys Id",
snap_usename as "Usename",
snap_application_name as "Application Name",
snap_client_addr as "Client Addr",
snap_client_hostname as "Client Hostname",
snap_client_port as "Client Port",
snap_backend_start as "Backend Start",
snap_state as "State",
snap_sender_sent_location as "Sender Sent Location",
snap_receiver_write_location as "Receiver Write Location",
snap_receiver_flush_location as "Receiver Flush Location",
snap_receiver_replay_location as "Receiver Replay Location",
snap_sync_priority as "Sync Priority",
snap_sync_state as "Sync State"
FROM snapshot.snap_global_replication_stat
WHERE snapshot_id = %ld
and snap_node_name = '%s' limit 200;
解读：
这一部分描述用户表状态的统计信息，数据源于snapshot.snap_global_stat_all_tables表。
信息内容如下所示：
| 列名 | 描述 |
|---|---|
| DB Name | 数据库名称 |
| Schema | 模式名称 |
| Relname | 表名称 |
| Seq Scan | 顺序扫描的次数 |
| Seq Tup Read | 顺序扫描获取的活跃行数 |
| Index Scan | 索引扫描次数 |
| Index Tup Fetch | 索引扫描获取的活跃行数 |
| Tuple Insert | 插入的行数 |
| Tuple Update | 更新的行数 |
| Tuple Delete | 删除的行数 |
| Tuple Hot Update | HOT(Heap Only Tuple)更新行数备注：HOT更新指，如果更新后的新行和旧行位于同一个数据块内，则旧行会有一个指针指向新行，这样就不用更新索引了，通过索引访问到旧行数据，进而访问到新行数据。 |
| Live Tuple | 活跃行数(估值) |
| Dead Tuple | 死行数(估值) |
| Last Vacuum | 最后一次手动Vacuum的时间(不包含VACUUM FULL) |
| Last Autovacuum | 最后一次autovacuum的时间 |
| Last Analyze | 最后一次手动Analyze表的时间 |
| Last Autoanalyze | 最后一次autovacuum线程Analyze表的时间 |
| Vacuum Count | 手动vacuum的次数(不包含VACUUM FULL) |
| Autovacuum Count | autovacuum的次数 |
| Analyze Count | 手动Analyze的次数 |
| Autoanalyze Count | autovacuum线程Analyze表的次数 |
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id
SELECT snap_2.db_name as "DB Name",
snap_2.snap_schemaname as "Schema",
snap_2.snap_relname as "Relname",
(snap_2.snap_seq_scan - coalesce(snap_1.snap_seq_scan, 0)) as "Seq Scan",
(snap_2.snap_seq_tup_read - coalesce(snap_1.snap_seq_tup_read, 0)) as "Seq Tup Read",
(snap_2.snap_idx_scan - coalesce(snap_1.snap_idx_scan, 0)) as "Index Scan",
(snap_2.snap_idx_tup_fetch - coalesce(snap_1.snap_idx_tup_fetch, 0)) as "Index Tup Fetch",
(snap_2.snap_n_tup_ins - coalesce(snap_1.snap_n_tup_ins, 0)) as "Tuple Insert",
(snap_2.snap_n_tup_upd - coalesce(snap_1.snap_n_tup_upd, 0)) as "Tuple Update",
(snap_2.snap_n_tup_del - coalesce(snap_1.snap_n_tup_del, 0)) as "Tuple Delete",
(snap_2.snap_n_tup_hot_upd - coalesce(snap_1.snap_n_tup_hot_upd, 0)) as "Tuple Hot Update",
snap_2.snap_n_live_tup as "Live Tuple",
snap_2.snap_n_dead_tup as "Dead Tuple",
to_char(snap_2.snap_last_vacuum, 'YYYY-MM-DD HH24:MI:SS') as "Last Vacuum",
to_char(snap_2.snap_last_autovacuum, 'YYYY-MM-DD HH24:MI:SS') as "Last Autovacuum",
to_char(snap_2.snap_last_analyze, 'YYYY-MM-DD HH24:MI:SS') as "Last Analyze",
to_char(snap_2.snap_last_autoanalyze, 'YYYY-MM-DD HH24:MI:SS') as "Last Autoanalyze",
(snap_2.snap_vacuum_count - coalesce(snap_1.snap_vacuum_count, 0)) as "Vacuum Count",
(snap_2.snap_autovacuum_count - coalesce(snap_1.snap_autovacuum_count, 0)) as "Autovacuum Count",
(snap_2.snap_analyze_count - coalesce(snap_1.snap_analyze_count, 0)) as "Analyze Count",
(snap_2.snap_autoanalyze_count - coalesce(snap_1.snap_autoanalyze_count, 0)) as "Autoanalyze Count"
FROM
(SELECT * FROM snapshot.snap_global_stat_all_tables
WHERE snapshot_id = %ld
and snap_node_name = '%s'
and snap_schemaname NOT IN ('pg_catalog', 'information_schema', 'snapshot')
AND snap_schemaname !~ '^pg_toast') snap_2
LEFT JOIN
(SELECT * FROM snapshot.snap_global_stat_all_tables
WHERE snapshot_id = %ld
and snap_node_name = '%s'
and snap_schemaname NOT IN ('pg_catalog', 'information_schema', 'snapshot')
AND snap_schemaname !~ '^pg_toast') snap_1
ON snap_2.snap_relid = snap_1.snap_relid
AND snap_2.snap_schemaname = snap_1.snap_schemaname
AND snap_2.snap_relname = snap_1.snap_relname
AND snap_2.db_name = snap_1.db_name
order by snap_2.db_name, snap_2.snap_schemaname
limit 200;
解读：
这一部分描述用户索引状态的统计信息，数据源于snapshot.snap_global_stat_all_indexes表。
信息内容如下所示：
| 列名 | 描述 |
|---|---|
| DB Name | 数据库名称 |
| Schema | 模式名称 |
| Relname | 表名称 |
| Index Relname | 索引名称 |
| Index Scan | 索引扫描次数 |
| Index Tuple Read | 索引扫描返回的索引条目数 |
| Index Tuple Fetch | 索引扫描获取的活跃行数 |
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id
SELECT snap_2.db_name as "DB Name",
snap_2.snap_schemaname as "Schema",
snap_2.snap_relname as "Relname",
snap_2.snap_indexrelname as "Index Relname",
(snap_2.snap_idx_scan - coalesce(snap_1.snap_idx_scan, 0)) as "Index Scan",
(snap_2.snap_idx_tup_read - coalesce(snap_1.snap_idx_tup_read, 0)) as "Index Tuple Read",
(snap_2.snap_idx_tup_fetch - coalesce(snap_1.snap_idx_tup_fetch, 0)) as "Index Tuple Fetch"
FROM
(SELECT * FROM snapshot.snap_global_stat_all_indexes
WHERE snapshot_id = %ld
and snap_node_name = '%s'
and snap_schemaname NOT IN ('pg_catalog', 'information_schema', 'snapshot')
and snap_schemaname !~ '^pg_toast') snap_2
LEFT JOIN
(SELECT * FROM snapshot.snap_global_stat_all_indexes
WHERE snapshot_id = %ld
and snap_node_name = '%s'
and snap_schemaname NOT IN ('pg_catalog', 'information_schema', 'snapshot')
and snap_schemaname !~ '^pg_toast') snap_1
ON snap_2.snap_relid = snap_1.snap_relid
and snap_2.snap_indexrelid = snap_1.snap_indexrelid
and snap_2.snap_schemaname = snap_1.snap_schemaname
and snap_2.snap_relname = snap_1.snap_relname
and snap_2.snap_indexrelname = snap_1.snap_indexrelname
and snap_2.db_name = snap_1.db_name
order by snap_2.db_name, snap_2.snap_schemaname
limit 200;
解读：
这一部分描述坏块的统计信息，数据源于snapshot.snap_global_stat_bad_block表。
信息内容如下所示：
| 列名 | 描述 |
|---|---|
| DB Id | 数据库OID |
| Tablespace Id | 表空间OID |
| Relfilenode | relation的filenode号 |
| Fork Number | fork编号 |
| Error Count | error数量 |
| First Time | 坏块第一次出现的时间 |
| Last Time | 坏块最近一次出现的时间 |
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id
SELECT snap_2.snap_databaseid AS "DB Id",
snap_2.snap_tablespaceid AS "Tablespace Id",
snap_2.snap_relfilenode AS "Relfilenode",
snap_2.snap_forknum AS "Fork Number",
(snap_2.snap_error_count - coalesce(snap_1.snap_error_count, 0)) AS "Error Count",
snap_2.snap_first_time AS "First Time",
snap_2.snap_last_time AS "Last Time"
FROM
(SELECT * FROM snapshot.snap_global_stat_bad_block
WHERE snapshot_id = %ld
and snap_node_name = '%s') snap_2
LEFT JOIN
(SELECT * FROM snapshot.snap_global_stat_bad_block
WHERE snapshot_id = %ld
and snap_node_name = '%s') snap_1
ON snap_2.snap_databaseid = snap_1.snap_databaseid
and snap_2.snap_tablespaceid = snap_1.snap_tablespaceid
and snap_2.snap_relfilenode = snap_1.snap_relfilenode
limit 200;
解读：
这一部分描述的是数据库参数配置信息，数据源于snapshot.snap_global_config_settings表。
信息内容如下所示：
| 列名 | 描述 |
|---|---|
| Name | 参数名称 |
| Abstract | 参数的简单描述 |
| Type | 参数类型(bool/enum/integer/real/string) |
| Curent Value | 参数当前值 |
| Min Value | 最小参数值 |
| Max Value | 最大参数值 |
| Category | 参数逻辑组 |
| Enum Values | 参数枚举值 |
| Default Value | 启动时的默认参数值 |
| Reset Value | 重置时的默认参数值 |
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id
select snap_name as "Name",
snap_short_desc as "Abstract",
snap_vartype as "Type",
snap_setting as "Curent Value",
snap_min_val as "Min Value",
snap_max_val as "Max Value",
snap_category as "Category",
snap_enumvals as "Enum Values",
snap_boot_val as "Default Value",
snap_reset_val as "Reset Value"
FROM
snapshot.snap_global_config_settings
WHERE snapshot_id = %ld
and snap_node_name = '%s';
解读：
这一部分描述的是SQL语句的详细信息，数据来源于snapshot.snap_summary_statement表。
Unique SQL Id： 即SQL的唯一识别ID
SQL Text： 即SQL的具体内容。
相关代码：
-- 说明：%s代表node_name，%ld代表snapshot_id
select (t2.snap_unique_sql_id) as "Unique SQL Id",
(t2.snap_query) as "SQL Text"
from
snapshot.snap_summary_statement t2
where snapshot_id = %ld
and snap_node_name = '%s';
小结：
关于openGauss WDR报告的梳理和解读就到这里，这些内容对于刚刚接触openGauss的老DBA已经足够了，只要读懂WDR报告的数据计算方法和含义即意味着看懂了这份WDR报告内容，剩下的数据库优化工作完全可以参照已有的数据库优化经验操作。当然了，“小白”DBA也完全可以参照网络上大佬们分享的数据库优化案例以及类似Oracle等主流数据库的AWR报告解读，学习openGauss的WDR报告，从而执行openGauss数据库相关的优化工作