DWS_2000000032 Number of Database Deadlocks in a GaussDB(DWS) Cluster Exceeds the Threshold

Updated on 2025-03-03 GMT+08:00

View PDF

Alarm Description

If the number of deadlocks in the cluster database exceeds the threshold within a specific time frame and the suppression conditions are not met, the DMS alarm module will generate an alarm. The alarm will be cleared once the DMS alarm module detects that the number of deadlocks in the cluster database is below the threshold.

Alarm Attributes

Alarm ID	Alarm Category	Alarm Severity	Alarm Type	Service Type	Auto Cleared
DWS_2000000032	Management plane alarm	> 10 (Critical); > 1 (Major)	Service alarm	GaussDB(DWS)	Yes

Alarm Changes

Change Type	Change Version	Description	Reason for Change
New	8.2.1.230	New alarm	New alarm

Alarm Parameters

Type	Parameter	Description
Fault Location	Cluster name	Cluster for which the alarm is generated.
	Tenant name	Name of the tenant to which the cluster belongs.
	Alarm level	Severity of the alarm.
Additional Information	Resource ID	ID of the cluster for which the alarm is generated.
	Resource name	Cluster for which the alarm is generated.
	Database name	Name of the database for which the alarm is generated.
	First_alarm_time	First occurrence event of an alarm, including the alarm threshold and current value.

Impact on the System

The connection pool is unable to allocate more connections to service requests because a large number of lock requests are causing connections to be unresponsive.

Possible Causes

Resource contention and lock actions are mutually exclusive.

Procedure

Log in to the GaussDB(DWS) console.
On the Alarm page, view the alarms generated in the last seven days.
Use gsql to connect to the cluster based on the alarm information. For details, see Using the CLI to Connect to a GaussDB(DWS) Cluster.
Connect to the cluster and run the SQL statement to query the current lock conflict statement.
```
select * from pgxc_lock_conflicts;
```
Decide whether to terminate the lock based on the statement content. To terminate the lock, run the following statement. pid and nodename are obtained from the previous step.
```
execute direct on (nodename) 'SELECT PG_TERMINATE_BACKEND(pid)';
```