What is DeadLock in SQL Server? How to Minimize it?

What is DeadLock in SQL Server?

A deadlock occurs when two or more tasks permanently block each other by each task having a lock on a resource which the other tasks are trying to lock. For example:

• Transaction A acquires a share lock on row 1.
• Transaction B acquires a share lock on row 2.
• Transaction A now requests an exclusive lock on row 2, and is blocked until transaction B finishes and releases the share lock it has on row 2.
• Transaction B now requests an exclusive lock on row 1, and is blocked until transaction A finishes and releases the share lock it has on row 1.

Transaction A cannot complete until transaction B completes, but transaction B is blocked by transaction A. This condition is also called a cyclic dependency: Transaction A has a dependency on transaction B, and transaction B closes the circle by having a dependency on transaction A.

Both transactions in a deadlock will wait forever unless the deadlock is broken by an external process. The Microsoft SQL Server Database Engine deadlock monitor periodically checks for tasks that are in a deadlock. If the monitor detects a cyclic dependency, it chooses one of the tasks as a victim and terminates its transaction with an error. This allows the other task to complete its transaction. The application with the transaction that terminated with an error can retry the transaction, which usually completes after the other deadlocked transaction has finished.

Using certain coding conventions in applications reduces the chance that applications will cause deadlocks. For more information, see Minimizing Deadlocks.

Deadlocking is often confused with normal blocking. When a transaction requests a lock on a resource locked by another transaction, the requesting transaction waits until the lock is released. By default, SQL Server transactions do not time out, unless LOCK_TIMEOUT is set. The requesting transaction is blocked, not deadlocked, because the requesting transaction has not done anything to block the transaction owning the lock. Eventually, the owning transaction will complete and release the lock, and then the requesting transaction will be granted the lock and proceed.

Deadlocks are sometimes called a deadly embrace.

Deadlock is a condition that can occur on any system with multiple threads, not just on a relational database management system, and can occur for resources other than locks on database objects. For example, a thread in a multithreaded operating system might acquire one or more resources, such as blocks of memory. If the resource being acquired is currently owned by another thread, the first thread may have to wait for the owning thread to release the target resource. The waiting thread is said to have a dependency on the owning thread for that particular resource. In an instance of the Database Engine, sessions can deadlock when acquiring nondatabase resources, such as memory or threads.

In the illustration, transaction T1 has a dependency on transaction T2 for the Part table lock resource. Similarly, transaction T2 has a dependency on transaction T1 for the Suppliertable lock resource. Because these dependencies form a cycle, there is a deadlock between transactions T1 and T2.

Deadlocks can also occur when a table is partitioned and the LOCK_ESCALATION setting of ALTER TABLE is set to AUTO. When LOCK_ESCALATION is set to AUTO, concurrency increases by allowing the Database Engine to lock table partitions at the HoBT level instead of at the TABLE level. However, when separate transactions hold partition locks in a table and want a lock somewhere on the other transactions partition, this causes a deadlock. This type of deadlock can be avoided by setting LOCK_ESCALATION to TABLE; although this setting will reduce concurrency by forcing large updates to a partition to wait for a table lock.

Although deadlocks cannot be completely avoided, following certain coding conventions can minimize the chance of generating a deadlock. Minimizing deadlocks can increase transaction throughput and reduce system overhead because fewer transactions are:

• Rolled back, undoing all the work performed by the transaction.
• Resubmitted by applications because they were rolled back when deadlocked.

To help minimize deadlocks:

• Access objects in the same order.
• Avoid user interaction in transactions.
• Keep transactions short and in one batch.
• Use a lower isolation level.
• Use a row versioning-based isolation level.
  • Set READ_COMMITTED_SNAPSHOT database option ON to enable read-committed transactions to use row versioning.
  • Use snapshot isolation.
• Use bound connections.

Access Objects in the Same Order

If all concurrent transactions access objects in the same order, deadlocks are less likely to occur. For example, if two concurrent transactions obtain a lock on the Supplier table and then on the Part table, one transaction is blocked on the Supplier table until the other transaction is completed. After the first transaction commits or rolls back, the second continues, and a deadlock does not occur. Using stored procedures for all data modifications can standardize the order of accessing objects.

Avoid User Interaction in Transactions

Avoid writing transactions that include user interaction, because the speed of batches running without user intervention is much faster than the speed at which a user must manually respond to queries, such as replying to a prompt for a parameter requested by an application. For example, if a transaction is waiting for user input and the user goes to lunch or even home for the weekend, the user delays the transaction from completing. This degrades system throughput because any locks held by the transaction are released only when the transaction is committed or rolled back. Even if a deadlock situation does not arise, other transactions accessing the same resources are blocked while waiting for the transaction to complete.

Keep Transactions Short and in One Batch

A deadlock typically occurs when several long-running transactions execute concurrently in the same database. The longer the transaction, the longer the exclusive or update locks are held, blocking other activity and leading to possible deadlock situations.

Keeping transactions in one batch minimizes network roundtrips during a transaction, reducing possible delays in completing the transaction and releasing locks.

Use a Lower Isolation Level

Determine whether a transaction can run at a lower isolation level. Implementing read committed allows a transaction to read data previously read (not modified) by another transaction without waiting for the first transaction to complete. Using a lower isolation level, such as read committed, holds shared locks for a shorter duration than a higher isolation level, such as serializable. This reduces locking contention.

Use a Row Versioning-based Isolation Level

When the READ_COMMITTED_SNAPSHOT database option is set ON, a transaction running under read committed isolation level uses row versioning rather than shared locks during read operations.

Note
Some applications rely upon locking and blocking behavior of read committed isolation. For these applications, some change is required before this option can be enabled.

Snapshot isolation also uses row versioning, which does not use shared locks during read operations. Before a transaction can run under snapshot isolation, the ALLOW_SNAPSHOT_ISOLATION database option must be set ON.

Implement these isolation levels to minimize deadlocks that can occur between read and write operations.

Use Bound Connections

Using bound connections, two or more connections opened by the same application can cooperate with each other. Any locks acquired by the secondary connections are held as if they were acquired by the primary connection, and vice versa. Therefore they do not block each other.

Detecting and Ending Deadlocks

A deadlock occurs when two or more tasks permanently block each other by each task having a lock on a resource which the other tasks are trying to lock. The following graph presents a high level view of a deadlock state where:

• Task T1 has a lock on resource R1 (indicated by the arrow from R1 to T1) and has requested a lock on resource R2 (indicated by the arrow from T1 to R2).
• Task T2 has a lock on resource R2 (indicated by the arrow from R2 to T2) and has requested a lock on resource R1 (indicated by the arrow from T2 to R1).
• Because neither task can continue until a resource is available and neither resource can be released until a task continues, a deadlock state exists.

The SQL Server Database Engine automatically detects deadlock cycles within SQL Server. The Database Engine chooses one of the sessions as a deadlock victim and the current transaction is terminated with an error to break the deadlock.

Resources That Can Deadlock

Each user session might have one or more tasks running on its behalf where each task might acquire or wait to acquire a variety of resources. The following types of resources can cause blocking that could result in a deadlock.

• Locks. Waiting to acquire locks on resources, such as objects, pages, rows, metadata, and applications can cause deadlock. For example, transaction T1 has a shared (S) lock on row r1 and is waiting to get an exclusive (X) lock on r2. Transaction T2 has a shared (S) lock on r2 and is waiting to get an exclusive (X) lock on row r1. This results in a lock cycle in which T1 and T2 wait for each other to release the locked resources.
• Worker threads. A queued task waiting for an available worker thread can cause deadlock. If the queued task owns resources that are blocking all worker threads, a deadlock will result. For example, session S1 starts a transaction and acquires a shared (S) lock on row r1 and then goes to sleep. Active sessions running on all available worker threads are trying to acquire exclusive (X) locks on row r1. Because session S1 cannot acquire a worker thread, it cannot commit the transaction and release the lock on row r1. This results in a deadlock.
• Memory. When concurrent requests are waiting for memory grants that cannot be satisfied with the available memory, a deadlock can occur. For example, two concurrent queries, Q1 and Q2, execute as user-defined functions that acquire 10MB and 20MB of memory respectively. If each query needs 30MB and the total available memory is 20MB, then Q1 and Q2 must wait for each other to release memory, and this results in a deadlock.
• Parallel query execution-related resources Coordinator, producer, or consumer threads associated with an exchange port may block each other causing a deadlock usually when including at least one other process that is not a part of the parallel query. Also, when a parallel query starts execution, SQL Server determines the degree of parallelism, or the number of worker threads, based upon the current workload. If the system workload unexpectedly changes, for example, where new queries start running on the server or the system runs out of worker threads, then a deadlock could occur.
• Multiple Active Result Sets (MARS) resources. These resources are used to control interleaving of multiple active requests under MARS (see Batch Execution Environment and MARS).
  • User resource. When a thread is waiting for a resource that is potentially controlled by a user application, the resource is considered to be an external or user resource and is treated like a lock.
  • Session mutex. The tasks running in one session are interleaved, meaning that only one task can run under the session at a given time. Before the task can run, it must have exclusive access to the session mutex.
  • Transaction mutex. All tasks running in one transaction are interleaved, meaning that only one task can run under the transaction at a given time. Before the task can run, it must have exclusive access to the transaction mutex.
  In order for a task to run under MARS, it must acquire the session mutex. If the task is running under a transaction, it must then acquire the transaction mutex. This guarantees that only one task is active at one time in a given session and a given transaction. Once the required mutexes have been acquired, the task can execute. When the task finishes, or yields in the middle of the request, it will first release transaction mutex followed by the session mutex in reverse order of acquisition. However, deadlocks can occur with these resources. In the following code example, two tasks, user request U1 and user request U2, are running in the same session.
  U1:    Rs1=Command1.Execute("insert sometable EXEC usp_someproc");
  U2:    Rs2=Command2.Execute("select colA from sometable");
  

  The stored procedure executing from user request U1 has acquired the session mutex. If the stored procedure takes a long time to execute, it is assumed by the Database Engine that the stored procedure is waiting for input from the user. User request U2 is waiting for the session mutex while the user is waiting for the result set from U2, and U1 is waiting for a user resource. This is deadlock state logically illustrated as:

Deadlock Detection

All of the resources listed in the section above participate in the Database Engine deadlock detection scheme. Deadlock detection is performed by a lock monitor thread that periodically initiates a search through all of the tasks in an instance of the Database Engine. The following points describe the search process:

• The default interval is 5 seconds.
• If the lock monitor thread finds deadlocks, the deadlock detection interval will drop from 5 seconds to as low as 100 milliseconds depending on the frequency of deadlocks.
• If the lock monitor thread stops finding deadlocks, the Database Engine increases the intervals between searches to 5 seconds.
• If a deadlock has just been detected, it is assumed that the next threads that must wait for a lock are entering the deadlock cycle. The first couple of lock waits after a deadlock has been detected will immediately trigger a deadlock search rather than wait for the next deadlock detection interval. For example, if the current interval is 5 seconds, and a deadlock was just detected, the next lock wait will kick off the deadlock detector immediately. If this lock wait is part of a deadlock, it will be detected right away rather than during next deadlock search.