sqlserver数据库服务异常停止分析
背景
2019年11月13日早上9:37分钟,我们的运维监控系统告警提示某项目数据库服务1433端口无法访问,我们赶紧登陆服务器查看详情,发现数据库服务以及数据库agent服务都已停止,重新手工启动对应的服务恢复生产。接下来我们就需赶紧分析引起这个问题的原因以及如何去修复;
服务器环境情况如下:
Microsoft SQL Server 2017 (RTM-CU14) (KB4484710) - 14.0.3076.1 (X64) Mar 12 2019 19:29:19 Copyright (C) 2017 Microsoft Corporation Enterprise Edition: Core-based Licensing (64-bit) on Windows Server 2012 R2 Datacenter 6.3 <X64> (Build 9600: ) (Hypervisor)
硬件配置:8CPU , 32G memory,数据盘SSD云盘 550GiB (18300 IOPS)
分析过程
第一步 查询系统日志
a.有个eventid 1000的错误
- <Event xmlns="http://schemas.microsoft.com/win/2004/08/events/event">
- <System>
<Provider Name="Application Error" />
<EventID Qualifiers="0">1000</EventID>
<Level>2</Level>
<Task>100</Task>
<Keywords>0x80000000000000</Keywords>
<TimeCreated SystemTime="2019-11-13T01:37:40.000000000Z" />
<EventRecordID>8587</EventRecordID>
<Channel>Application</Channel>
<Computer>****************</Computer>
<Security />
</System>
- <EventData>
<Data>sqlservr.exe</Data>
<Data>2017.140.3076.1</Data>
<Data>5c88918c</Data>
<Data>ntdll.dll</Data>
<Data>6.3.9600.19304</Data>
<Data>5c7f684f</Data>
<Data>c0000374</Data>
<Data>00000000000f1cd0</Data>
<Data>1b8c</Data>
<Data>01d564b93362c212</Data>
<Data>E:Program FilesMicrosoft SQL ServerMSSQL14.MSSQLSERVERMSSQLBinnsqlservr.exe</Data>
<Data>C:WindowsSYSTEM32
tdll.dll</Data>
<Data>2db02342-05b6-11ea-80ba-a469ceaa5caa</Data>
<Data />
<Data />
</EventData>
</Event>
b.有个eventid 19019的错误;
- <Event xmlns="http://schemas.microsoft.com/win/2004/08/events/event">
- <System>
<Provider Name="MSSQLSERVER" />
<EventID Qualifiers="16384">19019</EventID>
<Level>2</Level>
<Task>2</Task>
<Keywords>0x80000000000000</Keywords>
<TimeCreated SystemTime="2019-11-13T01:37:51.000000000Z" />
<EventRecordID>8589</EventRecordID>
<Channel>Application</Channel>
<Computer>*******************</Computer>
<Security />
</System>
- <EventData>
<Data>MSSQLSERVER 服务意外终止.</Data>
<Binary>B9370000190000001000000069005A006100630075006900650062006400780031007300370069005A000000070000006D00610073007400650072000000</Binary>
</EventData>
</Event>
我是觉的这些信息不是非常明了,只是告知MSSQLSERVER服务确实在问题时间点意外终止;
c.夹在两个错误日志中间,注意到一个eventid 1001
- <Event xmlns="http://schemas.microsoft.com/win/2004/08/events/event">
- <System>
<Provider Name="Windows Error Reporting" />
<EventID Qualifiers="0">1001</EventID>
<Level>4</Level>
<Task>0</Task>
<Keywords>0x80000000000000</Keywords>
<TimeCreated SystemTime="2019-11-13T01:37:50.000000000Z" />
<EventRecordID>8588</EventRecordID>
<Channel>Application</Channel>
<Computer>**********************</Computer>
<Security />
</System>
- <EventData>
<Data />
<Data>0</Data>
<Data>APPCRASH</Data>
<Data>不可用</Data>
<Data>0</Data>
<Data>sqlservr.exe</Data>
<Data>2017.140.3076.1</Data>
<Data>5c88918c</Data>
<Data>StackHash_f18b</Data>
<Data>6.3.9600.19304</Data>
<Data>5c7f684f</Data>
<Data>c0000374</Data>
<Data>PCH_99_FROM_ntdll+0x0000000000090D0A</Data>
<Data />
<Data />
<Data>C:UsersMSSQLSERVERAppDataLocalTempWERA49D.tmp.appcompat.txt C:UsersMSSQLSERVERAppDataLocalTempWERA6EF.tmp.WERInternalMetadata.xml C:ProgramDataMicrosoftWindowsWERReportQueueAppCrash_sqlservr.exe_4b5e12dc9b4c86bd34896c996afe289e6149ec98_cac3d1cb_cab_4d0da6ee riagedump.dmp WERGenerationLog.txt</Data>
<Data>C:ProgramDataMicrosoftWindowsWERReportQueueAppCrash_sqlservr.exe_4b5e12dc9b4c86bd34896c996afe289e6149ec98_cac3d1cb_cab_4d0da6ee</Data>
<Data />
<Data>0</Data>
<Data>2db02342-05b6-11ea-80ba-a469ceaa5caa</Data>
<Data>4</Data>
<Data />
</EventData>
</Event>
发现系统已打出了dump文件了,直接把上述地址的文件,拷到本地来分析。
通过windbg查看如下:
# Child-SP RetAddr Call Site 00 000000e1`14c192b8 00007ff8`6012b61e ntdll!NtWaitForMultipleObjects+0xa 01 000000e1`14c192c0 00007ff8`6012b17c ntdll!RtlReportExceptionEx+0x452 02 000000e1`14c19890 00007ff8`60151d26 ntdll!RtlReportException+0xbc 03 000000e1`14c19920 00007ff8`600e1e26 ntdll!RtlReportCriticalFailure$filt$0+0x33 04 000000e1`14c19950 00007ff8`600f030e ntdll!_C_specific_handler+0x96 05 000000e1`14c199c0 00007ff8`600f349d ntdll!_GSHandlerCheck_SEH+0x76 06 000000e1`14c199f0 00007ff8`600b48d7 ntdll!RtlpExecuteHandlerForException+0xd 07 000000e1`14c19a20 00007ff8`600b3afd ntdll!RtlDispatchException+0x197 08 000000e1`14c1a0f0 00007ff8`60151cd0 ntdll!RtlRaiseException+0x18d 09 000000e1`14c1a8b0 00007ff8`60154f12 ntdll!RtlReportCriticalFailure+0x8c 0a 000000e1`14c1a9c0 00007ff8`60155b10 ntdll!RtlpHeapHandleError+0x12 0b 000000e1`14c1a9f0 00007ff8`6010a5ff ntdll!RtlpLogHeapFailure+0xa4 0c 000000e1`14c1aa20 00007ff8`40d269d8 ntdll!RtlFreeHeap+0x74f2f 0d 000000e1`14c1aac0 00007ff8`40d27692 msvcr120!free+0x1c 0e 000000e1`14c1aaf0 00007ff8`31aa748e msvcr120!realloc+0x2e 0f 000000e1`14c1ab20 00000000`00000000 psqlodbc35w+0x5748e从分析结果来看像是第三方功能引起,接下来就要去数据库去验证猜测
第二步 数据库验证猜测
select * from sys.dm_os_loaded_modules
ODBC中设置如下:
数据库中有个dblink如下:
跟对应项目负责人员了解到,最近他们有个临时需求使用job定时在9:30通过dblink与postgres做数据同步,基本时间点也能对上了,那问题基本已明确是第三方的dblink调用造成sql server 异常停止;
那么接下的问题就是分析9:30调用的job的代码,就是运行一个存储过程,存储过程的详细代码如下:
--同步并清洗**销量数据
--1、分别将通过**link获取到的**和**销量数据同步到sfa_t_pflh_sales和sfa_t_drf_sales中
--2、将同步过来的销量数据进行清洗,并按客户需求保存数据到sfa_t_KaSales
CREATE PROCEDURE [dbo].[sfa_p_KaSalesLYNC]
AS
--同步****销量数据
INSERT INTO sfa_t_pflh_sales
SELECT *
FROM OPENQUERY
(POSTGRES,
'select * from "tn_bflh_sales" where "tn_tenant_id" = ''****''and tn_product_code<>''合计''AND tn_dealstatus=0 '
);
--更新*****销量销量数据同步状态tn_dealstatus=1代表已同步
UPDATE OPENQUERY
(POSTGRES, 'select * from "tn_bflh_sales"')
SET tn_dealstatus = 1
WHERE tn_tenant_id = '****'
AND tn_dealstatus = 0;
--同步***销量数据
INSERT INTO sfa_t_drf_sales
SELECT *
FROM OPENQUERY
(POSTGRES, 'select * from "tn_drf_salses" where "tn_tenant_id" = ''****'' AND tn_dealstatus=0 ');
--更新****销量销量数据同步状态tn_dealstatus=1代表已同步
UPDATE OPENQUERY
(POSTGRES, 'select * from "tn_drf_salses"')
SET tn_dealstatus = 1
WHERE tn_tenant_id = '****'
AND tn_dealstatus = 0;
--同步***link作业计划日志
TRUNCATE TABLE sfa_ka_tasklog;
INSERT INTO sfa_ka_tasklog
SELECT *
FROM OPENQUERY
(POSTGRES,
'SELECT t.tn_task_id,t.tn_ka_id,tn_ka_code,tn_dealers_account,t.tn_task_date,i.tn_status,i.tn_task_steps,i.tn_reak FROM tn_ka_task t LEFT JOIN tn_ka_task_info i on t.tn_task_id=i.tn_task_id
where t.tn_tenant_id=''*****'''
)
ORDER BY tn_task_date,tn_task_steps
DELETE FROM sfa_t_KaSales WHERE DATEDIFF(DAY, tn_date, GETDATE()) = 1--清洗数据前,先删除之前清洗好的前一天销量
INSERT INTO sfa_t_KaSales
--清洗******前一天销量数据
SELECT tn_tenant_id,'****'kaname,
tn_dealer_number,
case WHEN tn_supplier_code = '******' THEN
'***'
WHEN tn_supplier_code = '*****' THEN
'***'
END quyu,
tn_supplier_code,
tn_supplier_name,
tn_end_date tn_date,
tn_product_code,
tn_product_desc producname,
tn_store_code,
tn_store_name,
MAX(tn_sale_quantity) AS tn_day_sales,
CAST(MAX(tn_sale_amount) AS NVARCHAR(20)) AS tn_sale_amount
FROM sfa_t_pflh_sales
WHERE DATEDIFF(DAY, tn_create_date, GETDATE()) = 0
--AND tn_dealstatus = 0
GROUP BY tn_tenant_id,
tn_dealer_number,
tn_supplier_code,
tn_supplier_name,
tn_end_date,
tn_product_code,
tn_product_desc,
tn_store_code,
tn_store_name
UNION
--清洗****前一天销量数据
SELECT tn_tenant_id,'****'kaname,
tn_dealer_number,
CASE
WHEN tn_dealer_number = '*****' THEN
'***'
WHEN tn_dealer_number = '****' THEN
'***'
WHEN tn_dealer_number = '****' THEN
'***'
END quyu,
CASE
WHEN tn_dealer_number = '****' THEN
'****'
WHEN tn_dealer_number = '*****' THEN
'****'
WHEN tn_dealer_number = '****' THEN
'***'
END tn_supplier_code,
'******' tn_supplier_name,
DATEADD(DAY, -1, tn_date) tn_date,
tn_product_code,
p.producname,
'' tn_store_code,
tn_store_name,
MAX(tn_day_sales) tn_day_sales,
'' tn_sale_amount
FROM dbo.sfa_t_drf_sales s
LEFT JOIN dbo.sfa_t_KaProduct p
ON s.tn_product_code = p.productcode
AND p.KaName = 'drf'
WHERE DATEDIFF(DAY, s.tn_create_date, GETDATE()) = 0
--AND s.tn_dealstatus = 0
GROUP BY tn_tenant_id,
tn_dealer_number,
tn_product_code,
tn_store_name,
p.producname,
DATEADD(DAY, -1, tn_date);
--为保持数据清洗速度,各***系统销量临时表只保留15天数据
DELETE FROM sfa_t_pflh_sales
WHERE DATEDIFF(DAY, tn_create_date, GETDATE()) = 15
DELETE FROM sfa_t_drf_sales
WHERE DATEDIFF(DAY, tn_create_date, GETDATE()) = 15
GO
/*--同步并清洗**销量数据
--1、分别将通过**link获取到的**和**销量数据同步到sfa_t_pflh_sales和sfa_t_drf_sales中
--2、将同步过来的销量数据进行清洗,并按客户需求保存数据到sfa_t_KaSales
*/
CREATE PROCEDURE [dbo].[sfa_p_KaSalesLYNC]
AS
/*--同步****销量数据*/
INSERT INTO sfa_t_pflh_sales
SELECT *
FROM OPENQUERY
(POSTGRES,
'select * from "tn_bflh_sales" where "tn_tenant_id" = ''****''and tn_product_code<>''合计''AND tn_dealstatus=0 '
);
/*--更新*****销量销量数据同步状态tn_dealstatus=1代表已同步*/
UPDATE OPENQUERY
(POSTGRES, 'select * from "tn_bflh_sales"')
SET tn_dealstatus = 1
WHERE tn_tenant_id = '****'
AND tn_dealstatus = 0;
/*--同步***销量数据*/
INSERT INTO sfa_t_drf_sales
SELECT *
FROM OPENQUERY
(POSTGRES, 'select * from "tn_drf_salses" where "tn_tenant_id" = ''****'' AND tn_dealstatus=0 ');
/*--更新****销量销量数据同步状态tn_dealstatus=1代表已同步*/
UPDATE OPENQUERY
(POSTGRES, 'select * from "tn_drf_salses"')
SET tn_dealstatus = 1
WHERE tn_tenant_id = '****'
AND tn_dealstatus = 0;
/*--同步***link作业计划日志*/
TRUNCATE TABLE sfa_ka_tasklog;
INSERT INTO sfa_ka_tasklog
SELECT *
FROM OPENQUERY
(POSTGRES,
'SELECT t.tn_task_id,t.tn_ka_id,tn_ka_code,tn_dealers_account,t.tn_task_date,i.tn_status,i.tn_task_steps,i.tn_reak FROM tn_ka_task t LEFT JOIN tn_ka_task_info i on t.tn_task_id=i.tn_task_id
where t.tn_tenant_id=''*****'''
)
ORDER BY tn_task_date,tn_task_steps
DELETE FROM sfa_t_KaSales WHERE DATEDIFF(DAY, tn_date, GETDATE()) = 1 /*--清洗数据前,先删除之前清洗好的前一天销量*/
INSERT INTO sfa_t_KaSales
/*--清洗******前一天销量数据*/
SELECT tn_tenant_id,'****'kaname,
tn_dealer_number,
case WHEN tn_supplier_code = '******' THEN
'***'
WHEN tn_supplier_code = '*****' THEN
'***'
END quyu,
tn_supplier_code,
tn_supplier_name,
tn_end_date tn_date,
tn_product_code,
tn_product_desc producname,
tn_store_code,
tn_store_name,
MAX(tn_sale_quantity) AS tn_day_sales,
CAST(MAX(tn_sale_amount) AS NVARCHAR(20)) AS tn_sale_amount
FROM sfa_t_pflh_sales
WHERE DATEDIFF(DAY, tn_create_date, GETDATE()) = 0
/* --AND tn_dealstatus = 0 */
GROUP BY tn_tenant_id,
tn_dealer_number,
tn_supplier_code,
tn_supplier_name,
tn_end_date,
tn_product_code,
tn_product_desc,
tn_store_code,
tn_store_name
UNION
/*--清洗****前一天销量数据*/
SELECT tn_tenant_id,'****'kaname,
tn_dealer_number,
CASE
WHEN tn_dealer_number = '*****' THEN
'***'
WHEN tn_dealer_number = '****' THEN
'***'
WHEN tn_dealer_number = '****' THEN
'***'
END quyu,
CASE
WHEN tn_dealer_number = '****' THEN
'****'
WHEN tn_dealer_number = '*****' THEN
'****'
WHEN tn_dealer_number = '****' THEN
'***'
END tn_supplier_code,
'******' tn_supplier_name,
DATEADD(DAY, -1, tn_date) tn_date,
tn_product_code,
p.producname,
'' tn_store_code,
tn_store_name,
MAX(tn_day_sales) tn_day_sales,
'' tn_sale_amount
FROM dbo.sfa_t_drf_sales s
LEFT JOIN dbo.sfa_t_KaProduct p
ON s.tn_product_code = p.productcode
AND p.KaName = 'drf'
WHERE DATEDIFF(DAY, s.tn_create_date, GETDATE()) = 0
/* --AND s.tn_dealstatus = 0 */
GROUP BY tn_tenant_id,
tn_dealer_number,
tn_product_code,
tn_store_name,
p.producname,
DATEADD(DAY, -1, tn_date);
/*--为保持数据清洗速度,各***系统销量临时表只保留15天数据 */
DELETE FROM sfa_t_pflh_sales
WHERE DATEDIFF(DAY, tn_create_date, GETDATE()) = 15
DELETE FROM sfa_t_drf_sales
WHERE DATEDIFF(DAY, tn_create_date, GETDATE()) = 15
GO
后记
后续通过观察,发现该job某些时间段是可以正常执行,某些时间段是会失败;
直接报错信息如下:
Msg 109, Level 20, State 0, Line 6
A transport-level error has occurred when receiving results from the server. (provider: Shared Memory Provider, error: 0 - The pipe has been ended.)
微软官方有个hotfix针对这种情况,可惜下载执行提示已不适应当前系统了;
该job使用dblink的内存与sql server设置的最大内存以及是否锁定内存页都没有直接关系,我们甚至把max memroy 值从28G调小至10G,该现象还是会出现,观察该job执行的时候,其中有一段脚本reads超15万,查询有类似的案例如下:
后续跟该项目负责人沟通之后,他们评估认为直接停用该job,该问题是没有再重现了。但我认为如他们要实现需求还可以尝试:
- 替换现有的ODBC的驱动;
- 在job中代码做些优化;
- 或换其他模式如SSIS等;
结论
从这个案例来看,直接使用DBlink做数据对接,且每次数据量超5万以上不是一个首选方案;