If the queries you are interested in are dynamic queries that fail intermittently, you could log the SQL and the datetime and user in a table at the time the dynamic statement is created. It would be done on a case-by case basis though as it requires specific programming to happen and it takes a littel extra processing time, so do it only for those few queries you are most concerned about. But having a log of the specific statements executed can really help when you are trying to find out why it fails once a month only. Dynamic queries are hard to thoroughly test and sometimes you get one specific input value that just won't work and doing this logging at the time the SQL is created is often the best way to see what specifically wasn in the sql that was built.

[因为这个问题可能会作为一个重复的问题结束。]

如果SQL Server还没有重新启动(并且计划还没有被清除等等)，您可能能够在计划缓存中找到该查询。

SELECT t.[text]
FROM sys.dm_exec_cached_plans AS p
CROSS APPLY sys.dm_exec_sql_text(p.plan_handle) AS t
WHERE t.[text] LIKE N'%something unique about your query%';

如果你因为Management Studio崩溃而丢失了文件，你可以在这里找到恢复文件:

C:\Users\<you>\Documents\SQL Server Management Studio\Backup Files\

否则，你将需要使用其他工具来帮助你保存查询历史，比如Ed Harper的回答中提到的SSMS Tools Pack——尽管它在SQL Server 2012+中不是免费的。或者您可以设置一些轻量级跟踪过滤您的登录名或主机名(但请使用服务器端跟踪，而不是Profiler)。

正如@Nenad-Zivkovic评论的那样，加入sys可能会有帮助。Dm_exec_query_stats和last_execution_time排序:

SELECT t.[text], s.last_execution_time
FROM sys.dm_exec_cached_plans AS p
INNER JOIN sys.dm_exec_query_stats AS s
   ON p.plan_handle = s.plan_handle
CROSS APPLY sys.dm_exec_sql_text(p.plan_handle) AS t
WHERE t.[text] LIKE N'%something unique about your query%'
ORDER BY s.last_execution_time DESC;

一个稍微开箱即用的方法是在AutoHotKey中编写一个解决方案。我用了这个，它不是完美的，但是有效而且免费。从本质上讲，这个脚本为CTRL+SHIFT+R分配了一个热键，它将复制SSMS (CTRL+C)中所选的SQL，保存一个日期戳SQL文件，然后执行突出显示的查询(F5)。如果您不习惯AHK脚本，则前导分号是注释。

;CTRL+SHIFT+R to run a query that is first saved off
^+r::
;Copy
Send, ^c
; Set variables
EnvGet, HomeDir, USERPROFILE
FormatTime, DateString,,yyyyMMdd
FormatTime, TimeString,,hhmmss
; Make a spot to save the clipboard
FileCreateDir %HomeDir%\Documents\sqlhist\%DateString%
FileAppend, %Clipboard%, %HomeDir%\Documents\sqlhist\%DateString%\%TimeString%.sql
; execute the query
Send, {f5}
Return

最大的限制是，如果你点击“执行”而不是使用键盘快捷键，这个脚本将无法工作，并且这个脚本不会保存整个文件-只保存选中的文本。但是，您总是可以修改脚本以执行查询，然后在复制/保存之前选择所有(CTRL+A)。

使用具有“在文件中查找”功能的现代编辑器可以让您搜索SQL历史记录。您甚至可以把您的文件刮到SQLite3数据库中以查询您的查询。

正如其他人所指出的，您可以使用SQL Profiler，但也可以通过sp_trace_*系统存储过程来利用它的功能。例如，这个SQL片段将(至少在2000年;我认为这对SQL 2008是一样的，但你必须仔细检查)捕获RPC:Completed和SQL:BatchCompleted事件的所有查询需要超过10秒运行，并将输出保存到一个跟踪文件，您可以在稍后的日期在SQL分析器中打开:

DECLARE @TraceID INT
DECLARE @ON BIT
DECLARE @RetVal INT
SET @ON = 1

exec @RetVal = sp_trace_create @TraceID OUTPUT, 2, N'Y:\TraceFile.trc'
print 'This trace is Trace ID = ' + CAST(@TraceID AS NVARCHAR)
print 'Return value = ' + CAST(@RetVal AS NVARCHAR)
-- 10 = RPC:Completed
exec sp_trace_setevent @TraceID, 10, 1, @ON     -- Textdata
exec sp_trace_setevent @TraceID, 10, 3, @ON     -- DatabaseID
exec sp_trace_setevent @TraceID, 10, 12, @ON        -- SPID
exec sp_trace_setevent @TraceID, 10, 13, @ON        -- Duration
exec sp_trace_setevent @TraceID, 10, 14, @ON        -- StartTime
exec sp_trace_setevent @TraceID, 10, 15, @ON        -- EndTime

-- 12 = SQL:BatchCompleted
exec sp_trace_setevent @TraceID, 12, 1, @ON     -- Textdata
exec sp_trace_setevent @TraceID, 12, 3, @ON     -- DatabaseID
exec sp_trace_setevent @TraceID, 12, 12, @ON        -- SPID
exec sp_trace_setevent @TraceID, 12, 13, @ON        -- Duration
exec sp_trace_setevent @TraceID, 12, 14, @ON        -- StartTime
exec sp_trace_setevent @TraceID, 12, 15, @ON        -- EndTime

-- Filter for duration [column 13] greater than [operation 2] 10 seconds (= 10,000ms)
declare @duration bigint
set @duration = 10000
exec sp_trace_setfilter @TraceID, 13, 0, 2, @duration

您可以从Books Online中找到每个跟踪事件、列等的ID;只需搜索sp_trace_create、sp_trace_setevent和sp_trace_setfiler sppros即可。然后，您可以像下面这样控制跟踪:

exec sp_trace_setstatus 15, 0       -- Stop the trace
exec sp_trace_setstatus 15, 1       -- Start the trace
exec sp_trace_setstatus 15, 2       -- Close the trace file and delete the trace settings

.．.其中'15'是跟踪ID(如上面的第一个脚本所删除的sp_trace_create所报告的)。

您可以查看哪些跟踪正在运行:

select * from ::fn_trace_getinfo(default)

我唯一要谨慎说的是——我不知道这会给你的系统带来多少负载;它将添加一些，但“一些”的大小可能取决于您的服务器有多忙。

我使用下面的查询来跟踪未启用跟踪分析器的SQL服务器上的应用程序活动。 该方法使用查询存储(SQL Server 2016+)而不是DMV的。这提供了更好的查看历史数据的能力，以及更快的查找。 捕获sp_who/sp_whoisactive无法捕获的短时间运行的查询是非常有效的。

/* Adjust script to your needs.
    Run full script (F5) -> Interact with UI -> Run full script again (F5)
    Output will contain the queries completed in that timeframe.
*/

/* Requires Query Store to be enabled:
    ALTER DATABASE <db> SET QUERY_STORE = ON
    ALTER DATABASE <db> SET QUERY_STORE (OPERATION_MODE = READ_WRITE, MAX_STORAGE_SIZE_MB = 100000)
*/

USE <db> /* Select your DB */

IF OBJECT_ID('tempdb..#lastendtime') IS NULL
    SELECT GETUTCDATE() AS dt INTO #lastendtime
ELSE IF NOT EXISTS (SELECT * FROM #lastendtime)
    INSERT INTO #lastendtime VALUES (GETUTCDATE()) 

;WITH T AS (
SELECT 
    DB_NAME() AS DBName
    , s.name + '.' + o.name AS ObjectName
    , qt.query_sql_text
    , rs.runtime_stats_id
    , p.query_id
    , p.plan_id
    , CAST(p.last_execution_time AS DATETIME) AS last_execution_time
    , CASE WHEN p.last_execution_time > #lastendtime.dt THEN 'X' ELSE '' END AS New
    , CAST(rs.last_duration / 1.0e6 AS DECIMAL(9,3)) last_duration_s
    , rs.count_executions
    , rs.last_rowcount
    , rs.last_logical_io_reads
    , rs.last_physical_io_reads
    , q.query_parameterization_type_desc
FROM (
    SELECT *, ROW_NUMBER() OVER (PARTITION BY plan_id, runtime_stats_id ORDER BY runtime_stats_id DESC) AS recent_stats_in_current_priod
    FROM sys.query_store_runtime_stats 
    ) AS rs
INNER JOIN sys.query_store_runtime_stats_interval AS rsi ON rsi.runtime_stats_interval_id = rs.runtime_stats_interval_id
INNER JOIN sys.query_store_plan AS p ON p.plan_id = rs.plan_id
INNER JOIN sys.query_store_query AS q ON q.query_id = p.query_id
INNER JOIN sys.query_store_query_text AS qt ON qt.query_text_id = q.query_text_id
LEFT OUTER JOIN sys.objects AS o ON o.object_id = q.object_id
LEFT OUTER JOIN sys.schemas AS s ON s.schema_id = o.schema_id
CROSS APPLY #lastendtime
WHERE rsi.start_time <= GETUTCDATE() AND GETUTCDATE() < rsi.end_time
    AND recent_stats_in_current_priod = 1
    /* Adjust your filters: */
    -- AND (s.name IN ('<myschema>') OR s.name IS NULL)
UNION
SELECT NULL,NULL,NULL,NULL,NULL,NULL,dt,NULL,NULL,NULL,NULL,NULL,NULL, NULL
FROM #lastendtime
)
SELECT * FROM T
WHERE T.query_sql_text IS NULL OR T.query_sql_text NOT LIKE '%#lastendtime%' -- do not show myself
ORDER BY last_execution_time DESC

TRUNCATE TABLE #lastendtime
INSERT INTO #lastendtime VALUES (GETUTCDATE()) 

SELECT deqs.last_execution_time AS [Time], dest.text AS [Query], dest.*
FROM sys.dm_exec_query_stats AS deqs
CROSS APPLY sys.dm_exec_sql_text(deqs.sql_handle) AS dest
WHERE dest.dbid = DB_ID('msdb')
ORDER BY deqs.last_execution_time DESC

这将显示查询运行的时间和日期