本文分享自華為雲社群《GaussDB（DWS）監控工具指南（三）資源池級監控【綻放吧！GaussDB(DWS)雲原生數倉】》，作者：幕後小黑爪。
一、資源池
在資料庫最初階段，是沒有資源概念的，給資料庫輸入SQL語句，資料庫輸出結果，在簡單業務場景下，使用者獨佔資料庫是不存在資源爭搶問題的。隨著資料庫業務增長，使用者也越來越多，此時不同使用者間的SQL會搶佔作業系統的資源（CPU、記憶體、IO、網路等），如果不加限制的話就會影響整個叢集的使用者，造成叢集不可用的情況。為了防止在這種場景發生，需要對使用者業務SQL進行區分，對不同的使用者需要資源分配和管控。為此，資源池應運而生，資源池作為一種邏輯媒介，連線使用者和系統資源，管控每個使用者的資源使用，保證叢集的可用狀態。
二、GaussDB（DWS）中的資源池
當管理員建立使用者後，會自動繫結在預設資源池default_pool上，從網頁上，可以再建立資源池，然後繫結相應的使用者在對應的資源池上。此時使用者下發SQL語句執行，下發的語句就會收到資源池配置引數的管控。透過資源池可以劃分不同使用者的資源使用情況，簡化了叢集的管理，它可以統一管理所有的系統計算資源。這意味著管理員只需要管理資源池，而不是每個節點上的資源。
GaussDB（DWS）資源池（租戶）功能支援透過管控CPU、併發、記憶體、網路等手段對使用者下發的業務語句進行管控，在不同的維度對使用者語句進行管控。具體來說支援設定資源池引數的方式控制併發、記憶體、CPU利用率等能力。當使用者的語句是未知新語句時，也可以透過一些異常規則來控制語句情況，比如查殺超過限制的爛SQL等，在此基礎上，還支援設定黑名單的方式嚴格管控使用者下發的語句，這些功能後續會進行說明，本文聚焦資源池先不展開。
透過下列語句可查詢資源池資訊
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)
對於資源池的相關引數，使用者可透過管控面進行配置，同時也支援管理員透過語句來修改資源池配置，如下所示，可修改預設資源池的快車道併發限制。其他引數同理，不過該操作有風險，建議使用者在GaussDB（DWS）運維人員確認後執行。
alter resource pool default_pool with (max_dop=1);
三、資源池監控
GaussDB（DWS）為使用者提供了多維度的資源監控檢視，可支援從不同維度查詢叢集狀態。
GaussDB（DWS）提供資源池級別的監控能力，監控SQL語句的執行情況，主要包含實時監控和歷史監控，跟使用者監控類似，資源池監控展示了每個資源池的執行作業數、排隊作業數、記憶體使用、記憶體使用上限、 CPU使用情況、讀寫IO情況等，透過下列語句可進行查詢獲取
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 代表資源池所屬的邏輯叢集資訊。fast_run、slow_run代表資源池快車道執行數、慢車道執行數，fast_wait、slow_wait為快車道和慢車道的排隊作業數。其中，slow_wait也包含了CCN排隊的作業。fast_limit和slow_limit代表了快慢車道的併發上限，cpu_limit代表了資源池配置的cpu設定的限額是多少，透過used_cpu可以看到不同資源池的CPU使用率情況。estimate_mem為資源池內使用者下發作業的估算記憶體總和，used_mem為資源池實際使用的記憶體，mem_limit為資源池設定的可用記憶體上限。
同樣，歷史資源監控也提供了歷史問題定位的能力，該檢視會30s採集一次，timestamp為採集的時刻。
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、透過資源池監控分析定位問題（待補充）
1、當發現業務反饋語句阻塞嚴重，不執行作業，可查詢實時資源池監控或者歷史資源池監控進行分析，看是否作業堆積排隊，如果fast_limit上的限制為10，，fast_run欄位也為10，然後fast_wait較多，此時可嘗試修改資源池引數max_dop，適當調大併發上限。
2、業務反饋跑批業務在正常執行的情況下批次變慢，此時可觀察歷史資源池監控對比觀察，統計劣化前後同一批作業一段時間的記憶體資源使用情況，也可透過當時資源池作業排隊現象來定位問題。
五、更好用的監控檢視
為了提升系統可用性，GaussDB（DWS）也提供了更便捷，更易用的檢視用以幫使用者進行觀察系統狀態和定位問題。
在核心821版本中，使用者可使用gs_query_monitor、gs_user_monitor、gs_respool_monitor檢視進行語句級、使用者級、資源池的資源監控，這些檢視以GaussDB（DWS）監控工具指南系列中所講的檢視為基礎，選取常用的定位欄位，為現網使用者提供更易用的一套實時監控指令碼。
具體效果如下：
1. 作業監控
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. 使用者監控
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. 資源池監控
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)