理论+应用，带你了解数据库资源池

本文分享自华为云社区《GaussDB（DWS）监控工具指南（三）资源池级监控【绽放吧！GaussDB(DWS)云原生数仓】》，作者：幕后小黑爪。

一、资源池

在数据库最初阶段，是没有资源概念的，给数据库输入SQL语句，数据库输出结果，在简单业务场景下，用户独占数据库是不存在资源争抢问题的。随着数据库业务增长，用户也越来越多，此时不同用户间的SQL会抢占操作系统的资源（CPU、内存、IO、网络等），如果不加限制的话就会影响整个集群的用户，造成集群不可用的情况。为了防止在这种场景发生，需要对用户业务SQL进行区分，对不同的用户需要资源分配和管控。为此，资源池应运而生，资源池作为一种逻辑媒介，连接用户和系统资源，管控每个用户的资源使用，保证集群的可用状态。

二、GaussDB（DWS）中的资源池

当管理员创建用户后，会自动绑定在默认资源池default_pool上，从网页上，可以再创建资源池，然后绑定相应的用户在对应的资源池上。此时用户下发SQL语句执行，下发的语句就会收到资源池配置参数的管控。通过资源池可以划分不同用户的资源使用情况，简化了集群的管理，它可以统一管理所有的系统计算资源。这意味着管理员只需要管理资源池，而不是每个节点上的资源。

GaussDB（DWS）资源池（租户）功能支持通过管控CPU、并发、内存、网络等手段对用户下发的业务语句进行管控，在不同的维度对用户语句进行管控。具体来说支持设置资源池参数的方式控制并发、内存、CPU利用率等能力。当用户的语句是未知新语句时，也可以通过一些异常规则来控制语句情况，比如查杀超过限制的烂SQL等，在此基础上，还支持设置黑名单的方式严格管控用户下发的语句，这些功能后续会进行说明，本文聚焦资源池先不展开。

通过下列语句可查询资源池信息

postgres=# select * from pg_resource_pool;

respool_name | mem_percent | cpu_affinity | control_group | active_statements | max_dop | memory_limit | parentid | io_limits | io_priority | nodegroup | is_foreign | short_acc | except_rule | weight

----------------------+-------------+--------------+---------------------+-------------------+---------+--------------+------------+-----------+-------------+------------------+------------+-----------+-------------+--------

default_pool | 0 | -1 | DefaultClass:Medium | -1 | -1 | default | 0 | 0 | None | installation | f | t | None | -1

respool_1 | 0 | -1 | ClassN1:wn1 | 10 | -1 | default | 0 | 0 | None | logical_cluster1 | f | t | None | -1

respool_grp_1 | 20 | -1 | ClassG1 | 10 | -1 | default | 0 | 0 | None | logical_cluster1 | f | t | None | -1

respool_g1_job_1 | 20 | -1 | ClassG1:wg1_1 | 10 | -1 | default | 2147484586 | 0 | None | logical_cluster1 | f | t | None | -1

respool_g1_job_2 | 20 | -1 | ClassG1:wg1_2 | 10 | -1 | default | 2147484586 | 0 | None | logical_cluster1 | f | t | None | -1

respool_0_mempercent | 0 | -1 | DefaultClass:Medium | 10 | -1 | default | 0 | 0 | None | logical_cluster1 | f | t | None | -1

(6 rows)

对于资源池的相关参数，用户可通过管控面进行配置，同时也支持管理员通过语句来修改资源池配置，如下所示，可修改默认资源池的快车道并发限制。其他参数同理，不过该操作有风险，建议用户在GaussDB（DWS）运维人员确认后执行。

alter resource pool default_pool with (max_dop=1);

三、资源池监控

GaussDB（DWS）为用户提供了多维度的资源监控视图，可支持从不同维度查询集群状态。

GaussDB（DWS）提供资源池级别的监控能力，监控SQL语句的运行情况，主要包含实时监控和历史监控，跟用户监控类似，资源池监控展示了每个资源池的运行作业数、排队作业数、内存使用、内存使用上限、 CPU使用情况、读写IO情况等，通过下列语句可进行查询获取

postgres=# select * from gs_respool_resource_info;

nodegroup | rpname | cgroup | ref_count | fast_run | fast_wait | fast_limit | slow_run | slow_wait | slow_limit | used_cpu | cpu_limit | used_mem | estimate_mem | mem_limit | read_kbytes | write_kbytes | read_counts | write_counts | read_speed | write_speed

-----------+--------------+---------------------+-----------+----------+-----------+------------+----------+-----------+------------+----------+-----------+----------+--------------+-----------+-------------+--------------+-------------+--------------+------------+-------------

lc1 | pool_group | ClassN | 0 | 0 | 0 | -1 | 0 | 0 | 10 | 0 | 312 | 0 | 0 | 116844 | 0 | 0 | 0 | 0 | 0 | 0

lc1 | pool_work | ClassN:wg1 | 0 | 0 | 0 | 10 | 0 | 0 | 10 | 0 | 312 | 0 | 0 | 23364 | 0 | 0 | 0 | 0 | 0 | 0

lc2 | default_pool | DefaultClass:Medium | 0 | 0 | 0 | -1 | 0 | 0 | -1 | 0 | 208 | 0 | 0 | 584220 | 0 | 0 | 0 | 0 | 0 | 0

lc1 | resp_other | DefaultClass:Medium | 0 | 0 | 0 | -1 | 0 | 0 | 100 | 0 | 312 | 0 | 0 | 175260 | 0 | 0 | 0 | 0 | 0 | 0

lc1 | default_pool | DefaultClass:Medium | 0 | 0 | 0 | -1 | 0 | 0 | -1 | 0 | 312 | 0 | 0 | 584220 | 0 | 0 | 0 | 0 | 0 | 0

(5 rows)

其中，nodegroup 代表资源池所属的逻辑集群信息。fast_run、slow_run代表资源池快车道运行数、慢车道运行数，fast_wait、slow_wait为快车道和慢车道的排队作业数。其中，slow_wait也包含了CCN排队的作业。fast_limit和slow_limit代表了快慢车道的并发上限，cpu_limit代表了资源池配置的cpu设置的限额是多少，通过used_cpu可以看到不同资源池的CPU使用率情况。estimate_mem为资源池内用户下发作业的估算内存总和，used_mem为资源池实际使用的内存，mem_limit为资源池设置的可用内存上限。

同样，历史资源监控也提供了历史问题定位的能力，该视图会30s采集一次，timestamp为采集的时刻。

postgres=# select * from gs_respool_resource_history;

timestamp | nodegroup | rpname | cgroup | ref_count | fast_run | fast_wait | fast_limit | slow_run | slow_wait | slow_limit | used_cpu | cpu_limit | used_mem | estimate_mem | mem_limit | read_kbytes | write_kbytes | read_counts | write_counts | read_speed | write_speed

-------------------------------+-----------+--------------+---------------------+-----------+----------+-----------+------------+----------+-----------+------------+----------+-----------+----------+--------------+-----------+-------------+--------------+-------------+--------------+------------+-------------

2023-10-20 20:24:14.715107+08 | lc1 | pool_group | ClassN | 0 | 0 | 0 | -1 | 0 | 0 | 10 | 0 | 312 | 0 | 0 | 116844 | 0 | 0 | 0 | 0 | 0 | 0

2023-10-20 20:24:14.715107+08 | lc1 | pool_work | ClassN:wg1 | 0 | 0 | 0 | 10 | 0 | 0 | 10 | 0 | 312 | 0 | 0 | 23364 | 0 | 0 | 0 | 0 | 0 | 0

2023-10-20 20:24:14.715107+08 | lc2 | default_pool | DefaultClass:Medium | 0 | 0 | 0 | -1 | 0 | 0 | -1 | 0 | 208 | 0 | 0 | 584220 | 0 | 0 | 0 | 0 | 0 | 0

2023-10-20 20:24:14.715107+08 | lc1 | resp_other | DefaultClass:Medium | 0 | 0 | 0 | -1 | 0 | 0 | 100 | 0 | 312 | 0 | 0 | 175260 | 0 | 0 | 0 | 0 | 0 | 0

2023-10-20 20:24:14.715107+08 | lc1 | default_pool | DefaultClass:Medium | 0 | 0 | 0 | -1 | 0 | 0 | -1 | 0 | 312 | 0 | 0 | 584220 | 0 | 0 | 0 | 0 | 0 | 0

2023-10-20 20:24:44.791512+08 | lc1 | pool_group | ClassN | 0 | 0 | 0 | -1 | 0 | 0 | 10 | 0 | 312 | 0 | 0 | 116844 | 0 | 0 | 0 | 0 | 0 | 0

2023-10-20 20:24:44.791512+08 | lc1 | pool_work | ClassN:wg1 | 0 | 0 | 0 | 10 | 0 | 0 | 10 | 0 | 312 | 0 | 0 | 23364 | 0 | 0 | 0 | 0 | 0 | 0

2023-10-20 20:24:44.791512+08 | lc2 | default_pool | DefaultClass:Medium | 0 | 0 | 0 | -1 | 0 | 0 | -1 | 0 | 208 | 0 | 0 | 584220 | 0 | 0 | 0 | 0 | 0 | 0

2023-10-20 20:24:44.791512+08 | lc1 | resp_other | DefaultClass:Medium | 0 | 0 | 0 | -1 | 0 | 0 | 100 | 0 | 312 | 0 | 0 | 175260 | 0 | 0 | 0 | 0 | 0 | 0

2023-10-20 20:24:44.791512+08 | lc1 | default_pool | DefaultClass:Medium | 0 | 0 | 0 | -1 | 0 | 0 | -1 | 0 | 312 | 0 | 0 | 584220 | 0 | 0 | 0 | 0 | 0 | 0

...

四、通过资源池监控分析定位问题（待补充）

1、当发现业务反馈语句阻塞严重，不执行作业，可查询实时资源池监控或者历史资源池监控进行分析，看是否作业堆积排队，如果fast_limit上的限制为10，，fast_run字段也为10，然后fast_wait较多，此时可尝试修改资源池参数max_dop，适当调大并发上限。

2、业务反馈跑批业务在正常运行的情况下批量变慢，此时可观察历史资源池监控对比观察，统计劣化前后同一批作业一段时间的内存资源使用情况，也可通过当时资源池作业排队现象来定位问题。

五、更好用的监控视图

为了提升系统可用性，GaussDB（DWS）也提供了更便捷，更易用的视图用以帮用户进行观察系统状态和定位问题。

在内核821版本中，用户可使用gs_query_monitor、gs_user_monitor、gs_respool_monitor视图进行语句级、用户级、资源池的资源监控，这些视图以GaussDB（DWS）监控工具指南系列中所讲的视图为基础，选取常用的定位字段，为现网用户提供更易用的一套实时监控脚本。

具体效果如下：

1. 作业监控

postgres=# select * from gs_query_monitor;

usename | nodename | nodegroup | rpname | priority | xact_start | query_start | block_time | duration | query_band | attribute | lane | status | queue | used_mem | estimate_mem | used_cpu | read_speed | write_speed | send_speed | recv_speed | dn_count | stream_count | pid | lw

tid | query_id | unique_sql_id | query

--------------+----------+------------------+--------------+----------+-------------------------------+-------------------------------+------------+----------+------------+-------------+------+---------+-------+----------+--------------+----------+------------+-------------+------------+------------+----------+--------------+-----------------+---

-----+-------------------+---------------+--------------------------------------------------

user_default | cn_5001 | logical_cluster1 | default_pool | Medium | 2023-10-30 16:39:28.754207+08 | 2023-10-30 16:39:28.748855+08 | 59 | 0 | | Complicated | slow | pending | CCN | 0 | 4360 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 139906878865264 | 98

2280 | 72902018968076864 | 2372000271 | INSERT INTO t1 SELECT generate_series(1,100000);

user_default | cn_5001 | logical_cluster1 | default_pool | Medium | 2023-10-30 16:39:28.760305+08 | 2023-10-30 16:39:28.754861+08 | 59 | 0 | | Complicated | slow | pending | CCN | 0 | 4360 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 139906878866632 | 98

2283 | 72902018968076871 | 2372000271 | INSERT INTO t1 SELECT generate_series(1,100000);

user_default | cn_5001 | logical_cluster1 | default_pool | Medium | 2023-10-30 16:39:28.761491+08 | 2023-10-30 16:39:28.756124+08 | 59 | 0 | | Complicated | slow | pending | CCN | 0 | 4360 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 139906878865720 | 98

2281 | 72902018968076872 | 2372000271 | INSERT INTO t1 SELECT generate_series(1,100000);

user_default | cn_5001 | logical_cluster1 | default_pool | Medium | 2023-10-30 16:39:28.768333+08 | 2023-10-30 16:39:28.762653+08 | 59 | 0 | | Complicated | slow | pending | CCN | 0 | 4360 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 139906878867544 | 98

2285 | 72902018968076877 | 2372000271 | INSERT INTO t1 SELECT generate_series(1,100000);

user_default | cn_5001 | logical_cluster1 | default_pool | Medium | 2023-10-30 16:39:28.772288+08 | 2023-10-30 16:39:28.766933+08 | 59 | 0 | | Complicated | slow | pending | CCN | 0 | 4360 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 139906878868912 | 98

2288 | 72902018968076881 | 2372000271 | INSERT INTO t1 SELECT generate_series(1,100000);

user_default | cn_5001 | logical_cluster1 | default_pool | Medium | 2023-10-30 16:39:28.772304+08 | 2023-10-30 16:39:28.766966+08 | 59 | 0 | | Complicated | slow | pending | CCN | 0 | 4360 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 139906878867088 | 98

2284 | 72902018968076882 | 2372000271 | INSERT INTO t1 SELECT generate_series(1,100000);

user_default | cn_5001 | logical_cluster1 | default_pool | Medium | 2023-10-30 16:39:28.777958+08 | 2023-10-30 16:39:28.772572+08 | 59 | 0 | | Complicated | slow | pending | CCN | 0 | 4360 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 139906878868000 | 98

2286 | 72902018968076888 | 2372000271 | INSERT INTO t1 SELECT generate_series(1,100000);

user_default | cn_5001 | logical_cluster1 | default_pool | Medium | 2023-10-30 16:39:28.779373+08 | 2023-10-30 16:39:28.773997+08 | 59 | 0 | | Complicated | slow | pending | CCN | 0 | 4360 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 139906878868456 | 98

2287 | 72902018968076889 | 2372000271 | INSERT INTO t1 SELECT generate_series(1,100000);

user_default | cn_5001 | logical_cluster1 | default_pool | Medium | 2023-10-30 16:39:28.753845+08 | 2023-10-30 16:39:28.748498+08 | 0 | 59 | | Complicated | slow | running | None | 4 | 4360 | .289 | 0 | 0 | 0 | 0 | 0 | 0 | 139906878864808 | 98

2279 | 72902018968076862 | 2372000271 | INSERT INTO t1 SELECT generate_series(1,100000);

user_default | cn_5001 | logical_cluster1 | default_pool | Medium | 2023-10-30 16:39:28.753957+08 | 2023-10-30 16:39:28.748609+08 | 0 | 59 | | Complicated | slow | running | None | 4 | 4360 | .288 | 0 | 0 | 17 | 23 | 0 | 0 | 139906878866176 | 98

2282 | 72902018968076863 | 2372000271 | INSERT INTO t1 SELECT generate_series(1,100000);

(10 rows)

2. 用户监控

postgres=# select * from gs_user_monitor;

usename | rpname | nodegroup | session_count | active_count | global_wait | fast_run | fast_wait | slow_run | slow_wait | used_mem | estimate_mem | used_cpu | read_speed | write_speed | send_speed | recv_speed | used_space | space_limit | used_temp_space | temp_space_limit | used_spill_space | spill_space_limit

------------------+---------------+------------------+---------------+--------------+-------------+----------+-----------+----------+-----------+----------+--------------+----------+------------+-------------+------------+------------+------------+-------------+-----------------+------------------+------------------+-------------------

logical_cluster2 | default_pool | logical_cluster2 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | -1 | 0 | -1 | 0 | -1

user_grp_1 | respool_grp_1 | logical_cluster1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | -1 | 0 | -1 | 0 | -1

logical_cluster1 | default_pool | logical_cluster1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1834424 | -1 | 0 | -1 | 0 | -1

user_normal | respool_1 | logical_cluster1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | -1 | 0 | -1 | 0 | -1

user_default | default_pool | logical_cluster1 | 10 | 10 | 0 | 0 | 0 | 2 | 8 | 8 | 8720 | .563 | 0 | 15 | 0 | 0 | 640080 | -1 | 0 | -1 | 0 | -1

(5 rows)

3. 资源池监控

postgres=# select * from gs_respool_monitor;

rpname | nodegroup | cn_count | short_acc | session_count | active_count | global_wait | fast_run | fast_wait | fast_limit | slow_run | slow_wait | slow_limit | used_mem | estimate_mem | mem_limit | query_mem_limit | used_cpu | cpu_limit | read_speed | write_speed | send_speed | recv_speed

----------------------+------------------+----------+-----------+---------------+--------------+-------------+----------+-----------+------------+----------+-----------+------------+----------+--------------+-----------+-----------------+----------+-----------+------------+-------------+------------+------------

default_pool | logical_cluster2 | 3 | t | 0 | 0 | 0 | 0 | 0 | -1 | 0 | 0 | -1 | 0 bytes | 0 bytes | 11 GB | 4376 MB | 0 | 8 | 0 bytes/s | 0 bytes/s | 0 bytes/s | 0 bytes/s

respool_g1_job_1 | logical_cluster1 | 3 | t | 0 | 0 | 0 | 0 | 0 | -1 | 0 | 0 | 10 | 0 bytes | 0 bytes | 437 MB | 175 MB | 0 | 8 | 0 bytes/s | 0 bytes/s | 0 bytes/s | 0 bytes/s

respool_1 | logical_cluster1 | 3 | t | 0 | 0 | 0 | 0 | 0 | -1 | 0 | 0 | 10 | 0 bytes | 0 bytes | 11 GB | 4376 MB | 0 | 8 | 0 bytes/s | 0 bytes/s | 0 bytes/s | 0 bytes/s

respool_0_mempercent | logical_cluster1 | 3 | t | 0 | 0 | 0 | 0 | 0 | -1 | 0 | 0 | 10 | 0 bytes | 0 bytes | 11 GB | 4376 MB | 0 | 8 | 0 bytes/s | 0 bytes/s | 0 bytes/s | 0 bytes/s

respool_g1_job_2 | logical_cluster1 | 3 | t | 0 | 0 | 0 | 0 | 0 | -1 | 0 | 0 | 10 | 0 bytes | 0 bytes | 437 MB | 175 MB | 0 | 8 | 0 bytes/s | 0 bytes/s | 0 bytes/s | 0 bytes/s

default_pool | logical_cluster1 | 3 | t | 10 | 10 | 0 | 0 | 0 | -1 | 2 | 8 | -1 | 8192 KB | 8720 MB | 11 GB | 4376 MB | .577 | 8 | 0 bytes/s | 17 KB/s | 0 bytes/s | 0 bytes/s

(6 rows)

点击关注，第一时间了解华为云新鲜技术~