使用变更数据捕获 - SQL Server 和 Azure SQL 时，事务日志会由于长时间运行的事务而增长
本文可帮助你监视和识别 SQL Server、Azure SQL 数据库 和Azure SQL 托管实例上启用了变更数据捕获（CDC）的数据库中长时间运行的事务导致持续事务日志增长的问题。
现象
假设出现了下面这种情景：
- 对数据库启用 变更数据捕获 。
- CDC 的更改数据源是事务日志。 在将插入、更新和删除应用于跟踪的源表时，将会在日志中添加说明这些更改的项。
- 数据库上的事务日志由于长时间运行的事务而增长。
- 查询 给定数据库的 sys.databases 时，列
log_reuse_wait_desc将显示REPLICATION。
在此方案中，数据库事务日志文件逐渐增长，导致事务日志空间消耗过多。 事务日志大小达到最大定义的限制后，写入数据库会失败。
原因
在启用了 CDC 的数据库上，捕获作业延迟会保留日志截断，以确保可以将更改从事务日志捕获到 CDC 更改表，从而防止更改数据丢失。
解决方法
可以使用 Transact-SQL（T-SQL）来指定事务日志阈值和监视事务日志的时间间隔。 如有必要，可以通过设置 @kill_oldest_tran = 1终止事务。
若要监视事务日志，请使用以下 T-SQL 查询：
DECLARE
-- Log Transactions that generated Txlog over this size
@transaction_log_bytes_used INT = 5242880, -- 5MB (UPDATE)
-- Log full threshold
@log_full_threshold INT = 30, -- Percent (UPDATE)
-- Kill Oldest Tran (0 = FALSE or 1 = TRUE)
@kill_oldest_tran BIT = 0, --(UPDATE)
-- Log Transactions over this duration
@active_tran_time_minutes INT = 15, --(UPDATE)
-- This variable specifies the loop delay, format is Hours:minutes:seconds
@delay VARCHAR(9) = '00:10:00',
@runtime DATETIME,
@starttime DATETIME,
@msg NVARCHAR(100),
@oldest_tran_id BIGINT,
@oldest_tran_session_id INT,
@oldest_tran_begin_time DATETIME,
@killstr NVARCHAR(100)
IF OBJECT_ID('tblDiagLongTransactions') IS NULL
BEGIN
CREATE TABLE tblDiagLongTransactions
(
[datacollectiontime] [datetime] NOT NULL,
[transaction_id] [bigint] NOT NULL,
[name] [nvarchar](32) NOT NULL,
[transaction_begin_time] [datetime] NOT NULL,
[transaction_type] [int] NOT NULL,
[transaction_state] [int] NOT NULL,
[session_id] [int] NOT NULL,
[is_user_transaction] [bit] NOT NULL,
[database_transaction_log_bytes_used] [bigint] NOT NULL,
[login_time] [datetime] NOT NULL,
[last_request_start_time] [datetime] NOT NULL,
[last_request_end_time] [datetime] NULL,
[transaction_isolation_level] [smallint] NOT NULL,
[host_name] [nvarchar](128) NULL,
[nt_user_name] [nvarchar](128) NULL,
[command] [nvarchar](32) NULL,
[status] [nvarchar](30) NULL,
[cpu_time] [int] NULL,
[total_elapsed_time] [int] NULL,
[Transaction_time_in_mins] [int] NULL,
[logical_reads] [bigint] NULL,
[wait_time] [int] NULL,
[wait_type] [nvarchar](60) NULL,
[wait_resource] [nvarchar](256) NULL,
[blocking_session_id] [smallint] NULL,
[program_name] [nvarchar](128) NULL,
[granted_query_memory] [int] NULL,
[writes] [bigint] NULL,
[Request Reads] [bigint] NULL,
[Session Reads] [bigint] NOT NULL,
[Session Logical Reads] [bigint] NOT NULL,
[statement_text] [nvarchar](max) NULL,
[batch_text] [nvarchar](max) NULL,
[objectid] [int] NULL,
[query_hash] BINARY(8),
[query_plan_hash] BINARY(8),
[mostrecentsqltext] [nvarchar](max) NULL
) ON [PRIMARY]
END
WHILE (1=1)
BEGIN
-- Check if the database log used space is over the threshold
SET @runtime = GETDATE()
INSERT INTO tblDiagLongTransactions
SELECT DISTINCT
@runtime AS datacollectiontime,
atr.transaction_id,
atr.name,
transaction_begin_time,
transaction_type,
transaction_state,
dsr.session_id,
dsr.is_user_transaction,
dtr.database_transaction_log_bytes_used,
s.login_time,
s.last_request_start_time,
s.last_request_end_time,
s.transaction_isolation_level,
s.host_name,
s.nt_user_name,
r.command,
r.status,
r.cpu_time,
r.total_elapsed_time,
DATEDIFF(mi, transaction_begin_time, getdate()) AS 'Transaction_time_in_mins',
r.logical_reads,
r.wait_time,
r.wait_type,
r.wait_resource,
r.blocking_session_id,
s.program_name,
r.granted_query_memory,
r.writes,
r.reads AS [Request Reads],
s.reads AS [Session Reads],
s.logical_reads AS [Session Logical Reads],
(REPLACE(REPLACE(REPLACE(REPLACE(SUBSTRING(qt.text, r.statement_start_offset / 2 + 1,
(CASE WHEN r.statement_end_offset = -1 THEN LEN(CONVERT(NVARCHAR(MAX), qt.text)) * 2
ELSE r.statement_end_offset
END - r.statement_start_offset) / 2), ' ', ''), CHAR(13), ''), CHAR(10), ''), CHAR(9), '')) AS statement_text,
SUBSTRING(REPLACE(REPLACE(REPLACE(REPLACE(qt.text, ' ', ''), CHAR(13), ''), CHAR(10), ''), CHAR(9), ''), 1, 256) AS batch_text,
qt.objectid,
query_hash,
query_plan_hash,
mqt.text
FROM sys.dm_tran_active_transactions atr
INNER JOIN sys.dm_tran_database_transactions dtr ON atr.transaction_id = dtr.transaction_id
INNER JOIN sys.dm_tran_session_transactions dsr ON atr.transaction_id = dsr.transaction_id
LEFT OUTER JOIN sys.dm_exec_sessions s ON dsr.session_id = s.session_id
LEFT OUTER JOIN sys.dm_exec_connections conn ON s.session_id = dsr.session_id
LEFT OUTER JOIN sys.dm_exec_requests r ON r.session_id = s.session_id
OUTER APPLY sys.dm_exec_sql_text(r.sql_handle) AS qt
OUTER APPLY sys.dm_exec_sql_text(conn.most_recent_sql_handle) AS mqt
WHERE s.session_id != @@spid
AND atr.transaction_type != 2
AND (database_transaction_log_bytes_used > @transaction_log_bytes_used
OR datediff(minute, transaction_begin_time, getdate()) > @active_tran_time_minutes)
-- Check Log full threshold
IF @kill_oldest_tran = 1
BEGIN
IF EXISTS (
SELECT 1 FROM sys.dm_db_log_space_usage
WHERE used_log_space_in_percent >= @log_full_threshold
)
BEGIN
SELECT TOP 1
@oldest_tran_id = atr.transaction_id,
@oldest_tran_begin_time = transaction_begin_time,
@oldest_tran_session_id = dsr.session_id
FROM sys.dm_tran_active_transactions atr
LEFT OUTER JOIN sys.dm_tran_database_transactions dtr ON atr.transaction_id = dtr.transaction_id
LEFT OUTER JOIN sys.dm_tran_session_transactions dsr ON atr.transaction_id = dsr.transaction_id
LEFT OUTER JOIN sys.dm_exec_sessions s ON dsr.session_id = s.session_id
WHERE dsr.session_id != @@spid
AND is_user_transaction = 1
ORDER BY transaction_begin_time DESC
SELECT @oldest_tran_id AS TranID, @oldest_tran_begin_time AS TranbeginTime, @oldest_tran_session_id AS SessionID
SET @killstr = 'KILL ' + CAST(@oldest_tran_session_id AS VARCHAR(100))
PRINT @killstr
-- Kill oldest tran
EXEC (@killstr)
-- Checkpoint
CHECKPOINT
END
END
-- Change the polling interval as required
WAITFOR DELAY @delay
END