Oracle資料庫出現WARNING: too many parse errors告警的分析思路

潇湘隐者發表於2024-04-23
Oracle資料庫Error

Oracle資料庫的告警日誌中出WARNING: too many parse errors這些告警資訊的話，如果遇到這個問題，我們應該如何分析呢？

下面簡單聊一下如何分析這個錯誤。該告警資訊其實是12.2版本中的一個特性增強。在以前的Oracle版本中，資料庫出現瞭解析錯誤時，資料庫的alert日誌中不會有任何相關的提示，我們一般只能透過AWR報告才能瞭解資料庫出現瞭解析錯誤，例如，從"failed parse elapsed time" 和"parse count(failures)"指標中檢視解析出錯資訊，如下截圖所示：

.....................

如果資料庫中解析錯誤的次數非常頻繁時，可能會造成大量的Library Cache Lock等待，整個資料庫可能會處於hang死的狀態。要找出解析錯誤的root cause，則需要在資料庫中設定10035 event，如果再次出現解析錯誤時，會向資料庫的alert日誌中寫入解析錯誤的詳細資訊。

ALTER SYSTEM SET EVENTS '10035 trace name context forever, level 1';
ALTER SESSION SET EVENTS '10035 trace name context forever, level 1';
EVENT="10035 trace name context forever, level 1"

Levels:
level 1+ Print out failed parses of SQL statements to

Note:
The event can be turned off as follows:

ALTER SYSTEM SET EVENTS '10035 trace name context off';
ALTER SESSION SET EVENTS '10035 trace name context off';

而從12.2版本開始，即使未設定10035 event，當資料庫出現解析錯誤的情況時，仍然會向資料庫的alert日誌中寫入一條解析錯誤的告警資訊。

如下所示，你可能會看到類似這樣的報錯資訊：

2024-04-18T00:26:00.288821+08:00
*******(3):WARNING: too many parse errors, count=592 SQL hash=0xd4b65b68
*******(3):PARSE ERROR: ospid=969851, error=903 for statement: 
*******(3):Additional information: hd=0x1c6a4c5b80 phd=0x1f598d0500 flg=0x28 cisid=290 sid=290 ciuid=290 uid=290 sqlid=9mj0cyvabcqv8
*******(3):...Current username=***
*******(3):...Application: IgniteMonitor Action:

這裡比較關鍵的資訊是第二行錯誤資訊的錯誤程式碼："PARSE ERROR: ospid=969851, error=903 for statement"，這個例子中，它提示SQL解析出錯是因為遇到了ORA-903這個錯誤

$ oerr ora 903
00903, 00000, "invalid table name"
// *Cause:     A table or cluster name was invalid or does not exist.
//             This message was also issued if an invalid cluster name or no
//             cluster name was specified in an ALTER CLUSTER or DROP CLUSTER
//             statement.
// *Action:    Check spelling. A valid table name or cluster name
//             must begin with a letter and may contain only alphanumeric
//             characters and the special characters $, _, and #. The name
//             must be less than or equal to 30 characters and cannot be a
//             reserved word.

我們可以嘗試透過SQL_ID找到對應的SQL，但是有時候，可能透過SQL_ID可能已無法找到SQL語句，只能等到下一次出現時及時定位。

SELECT c.username,
      ,a.program
      ,b.sql_text
      ,b.command_type
      ,a.sample_time
FROM dba_hist_active_sess_history a
JOIN dba_hist_sqltext b
ON a.sql_id = b.sql_id
JOIN dba_users c
ON a.user_id = c.user_id
WHERE a.sample_time BETWEEN SYSDATE - 1 AND SYSDATE
and a.sql_id='9mj0cyvabcqv8'
ORDER BY a.sample_time DESC;

這裡如果可以找出具體SQL語句，就可以找出SQL解析出錯的原因，跟開發人員一起修復這個問題，像官方文件Doc ID 2649163.1[1]中提及的案例中

2020-01-07T11:35:33.918516+10:30
WARNING: too many parse errors, count=1091700 SQL hash=0xbbcb647d
PARSE ERROR: ospid=33376, error=923 for statement:
2020-01-07T11:35:33.918632+10:30
select 1
Additional information: hd=0xb336ab08 phd=0xb336af30 flg=0x28 cisid=120 sid=120 ciuid=120 uid=120
2020-01-07T11:39:04.673714+10:30
WARNING: too many parse errors, count=1091800 SQL hash=0xbbcb647d
PARSE ERROR: ospid=38578, error=923 for statement:sdkjfdsfkjsadfkjsadfkj
2020-01-07T11:39:04.673839+10:30
select 1

出現這個錯誤，是因為應用程式中輸入的SQL不完整，沒有from關鍵字，從錯誤程式碼error=923也能看出出錯的可能性。如下所示：

$ oerr ora 923
00923, 00000, "FROM keyword not found where expected"
// *Cause:     In a SELECT or REVOKE statement, the keyword FROM was
//             either missing, misplaced, or misspelled. The keyword FROM
//             must follow the last selected item in a SELECT statement or
//             the privileges in a REVOKE statement.
// *Action:    Correct the syntax. Insert the keyword FROM where
//             appropriate. The SELECT list itself also may be in error. If
//             quotation marks were used in an alias, check that double
//             quotation marks enclose the alias. Also, check to see if a
//             reserved word was used as an alias.

還有一些bug會引起WARNING: too many parse errors，此時就必須在Oracle metalink上進行搜尋，仔細匹配了。例如Doc ID 2976229.1 [2]。它的現象是Oracle DG的備庫中一直出現"WARNING: too many parse errors",而且資料庫版本為Oracle Database - Enterprise Edition - Version 19.3.0.0.0 to 19.21.0.0.0

2023-09-04T13:19:41.797929+00:00

WARNING: too many parse errors, count=13900 SQL hash=0x0cd5bf3b

PARSE ERROR: ospid=15822, error=1219 for statement:

2023-09-04T13:19:41.798049+00:00

select count(*) from cdb_service$

Additional information: hd=0x6cbc40d0 phd=0x6cbc4828 flg=0x20 cisid=0 sid=0

ciuid=2147483620 uid=2147483620 sqlid=<SQL ID>

...Current username=SYSRAC

...Application: oraagent.bin@<HOSTNAME> (TNS V1-V3) Action:

WARNING: too many parse errors, count=13900 SQL hash=0xeb8d02bf

PARSE ERROR: ospid=15822, error=1219 for statement

引起這個的原因是Bug,官方描述如下

The issue is analyzed and discussed in internal / unpublished Bug 34046765 - ORAAGENT DAGENT::SETCONNECTIONPOOLMAX GENERATES TOO MANY PARSE ERRORS ON STANDBY DATABASE

另外，最重要的就是監控資料庫的alert日誌，一旦出現這類告警就必須發出告警郵件或告警提示。以前我們監控資料庫的alert日誌，一般是過濾ORA-這類關鍵字，而這樣過濾的話，是無法獲取WARNING這類的告警資訊的。所以監控指令碼過濾關鍵字時，必須增加WARNING這個關鍵字資訊。

參考資料

[1]

1: https://support.oracle.com/epmos/faces/DocumentDisplay?_afrLoop=510335864406137&id=2649163.1&_afrWindowMode=0&_adf.ctrl-state=im4fw4kqq_78

 [2]

2: https://support.oracle.com/epmos/faces/DocumentDisplay?_afrLoop=510414734632502&id=2976229.1&_afrWindowMode=0&_adf.ctrl-state=im4fw4kqq_127

相關文章