When using the [Fail-on-Conflict](../concurrency-control/#fail-on-conflict) concurrency control policy, transactions are assigned priorities that help decide which transactions should be aborted in case of conflict.
There are two priority buckets, each having a priority range of [reals](https://www.postgresql.org/docs/15/datatype.html) in [0, 1] as follows:
1. `High-priority` bucket: if the first statement in a transaction takes a `FOR UPDATE/ FOR SHARE/ FOR NO KEY UPDATE` explicit row lock using SELECT, it will be assigned a priority from this bucket.
2. `Normal-priority` bucket: all other transactions are assigned a priority from this bucket.
Note that a transaction with any priority P1 from the high-priority bucket can abort a transaction with any priority P2 from the normal-priority bucket. For example, a transaction with priority 0.1 from the high-priority bucket can abort a transaction with priority 0.9 from the normal-priority bucket.
Priorities are randomly chosen from the applicable bucket. However, you can use the following two YSQL parameters to control the priority assigned to transactions in a specific session:
- `yb_transaction_priority_lower_bound`
- `yb_transaction_priority_upper_bound`
These parameters help set lower and upper bounds on the randomly-assigned priority that a transaction should receive from the applicable bucket. These parameters accept a value of `real` datatype in the range [0, 1]. Also note that the same bounds apply to both buckets.
{{< note title="All single shard transactions have a priority of 1 in the normal-priority bucket." >}}
{{}}
The `yb_get_current_transaction_priority` function can be used to fetch the transaction priority of the current active transaction. It outputs a pair ` (bucket)`, where `` is of a real datatype between [0, 1] with 9 decimal units of precision, and `` is either `Normal` or `High`.
{{< note title="Note">}}
As an exception, if a transaction is assigned the highest priority possible, that is, a priority of 1 in the high-priority bucket, the function returns `highest priority transaction` without any real value.
{{}}
A transaction's priority is `0.000000000 (normal-priority transaction)` until a transaction is really started.
## Examples
The following examples demonstrate how to set priorities for your transactions and get the current transaction priority.
1. Create a table and insert some data.
```sql
CREATE TABLE test (k INT PRIMARY KEY, v INT);
INSERT INTO test VALUES (1, 1);
```
1. Set the lower and upper bound values for your transactions as follows:
```sql
SET yb_transaction_priority_lower_bound = 0.4;
SET yb_transaction_priority_upper_bound = 0.6;
```
1. Create a transaction in the normal-priority bucket as follows:
```sql
BEGIN TRANSACTION ISOLATION LEVEL REPEATABLE READ;
SELECT yb_get_current_transaction_priority(); -- 0 due to an optimization which doesn't really start a real transaction internally unless a write occurs
```
```output
yb_get_current_transaction_priority
-------------------------------------------
0.000000000 (Normal priority transaction)
(1 row)
```
```sql
SELECT * FROM test;
```
```output
k | v
---+---
1 | 1
(1 row)
```
```sql
SELECT yb_get_current_transaction_priority(); -- still 0 due to the optimization which doesn't really start a real transaction internally unless a write occurs
```
```output
yb_get_current_transaction_priority
-------------------------------------------
0.000000000 (Normal priority transaction)
(1 row)
```
```sql
INSERT INTO test VALUES (2, '2'); -- perform a write which starts a real transaction
SELECT yb_get_current_transaction_priority(); -- non-zero now
```
```output
yb_get_current_transaction_priority
-------------------------------------------
0.537144608 (Normal priority transaction)
(1 row)
```
```sql
COMMIT;
```
1. Create a transaction in the high-priority bucket as follows:
```sql
BEGIN TRANSACTION ISOLATION LEVEL REPEATABLE READ;
SELECT * FROM test WHERE k = 1 FOR UPDATE; -- starts a transaction in a high-priority bucket
```
```output
k | v
---+---
1 | 1
(1 row)
```
```sql
SELECT yb_get_current_transaction_priority();
```
```output
yb_get_current_transaction_priority
-----------------------------------------
0.412004009 (High priority transaction)
(1 row)
```
```sql
COMMIT;
```
1. Create a transaction with the highest priority
```sql
SET yb_transaction_priority_upper_bound = 1;
SET yb_transaction_priority_lower_bound = 1;
BEGIN TRANSACTION ISOLATION LEVEL REPEATABLE READ;
SELECT * FROM test WHERE k = 1 FOR UPDATE;
```
```output
k | v
---+---
1 | 1
(1 row)
```
```sql
SELECT yb_get_current_transaction_priority();
```
```output
yb_get_current_transaction_priority
-------------------------------------
Highest priority transaction
(1 row)
```
```sql
COMMIT;
```
{{< note title="Internal representation of priorities" >}}
Internally, both the normal and high-priority buckets are mapped to a `uint64_t` space. The 64 bit range is used by the two priority buckets as follows:
1. Normal-priority bucket: `[yb::kRegularTxnLowerBound, yb::kRegularTxnUpperBound]`, that is, 0 to `uint32_t_max`-1
1. High-priority bucket: `[yb::kHighPriTxnLowerBound, yb::kHighPriTxnUpperBound]`, that is, `uint32_t_max` to `uint64_t_max`
For ease of use, the bounds are expressed as a [0, 1] real range for each bucket in the lower or upper bound YSQL parameters and the `yb_get_current_transaction_priority` function. The [0, 1] real range map proportionally to the integer ranges for both buckets. In other words, the [0, 1] range in the normal-priority bucket maps to `[0, uint32_t_max-1]` and the [0, 1] range in the high-priority bucket maps to `[uint32_t_max, uint64_t_max]`.
{{< /note >}}