Monitoring Memory Resources

Monitoring the Memory

GaussDB(DWS) provides a view for monitoring the memory usage of the entire cluster.

Query the pgxc_total_memory_detail view as a user with sysadmin permissions.

If the following error message is returned during the query, enable the memory management function.

To enable memory management, you can set enable_memory_limit and max_process_memory on the GaussDB(DWS) console as follows:

Log in to the GaussDB(DWS) management console.
In the navigation pane on the left, click Clusters.
In the cluster list, find the target cluster and click its name. The Basic Information page is displayed.
Click the Parameter Modification tab, change the value of enable_memory_limit to on, and click Save to save the file.
Change the value of max_process_memory to a proper one. For details about the modification suggestions, see max_process_memory. After it is done, click Save.
In the Modification Preview dialog box, confirm the modifications and click Save. After the modification, restart the cluster for the modification to take effect.

Monitoring the Shared Memory

You can query the context information about the shared memory on the pg_shared_memory_detail view.

This view lists the context name of the memory, level, the upper-layer memory context, and the total size of the shared memory.

In the database, GUC parameter memory_tracking_mode is used to configure the memory statistics collecting mode, including the following options:

none: The memory statistics collecting function is not enabled.
normal: Only memory statistics is collected in real time and no file is generated.
executor: The statistics file is generated, containing the context information about all allocated memory used on the execution layer.
When the parameter is set to executor, cvs files are generated under the pg_log directory of the DN process. The file names are in the format of memory_track_<DN name>_query_<queryid>.csv. The information about the operators executed by the postgres thread of the executor and all stream threads are input in this file during task execution.

The instance is built with a file content similar to the following:
```
0, 0, ExecutorState, 0, PortalHeapMemory, 0, 40K, 602K, 23
1, 3, CStoreScan_29360131_25, 0, ExecutorState, 1, 265K, 554K, 23
2, 128, cstore scan per scan memory context, 1, CStoreScan_29360131_25, 2, 24K, 24K, 23
3, 127, cstore scan memory context, 1, CStoreScan_29360131_25, 2, 264K, 264K, 23
4, 7, InitPartitionMapTmpMemoryContext, 1, CStoreScan_29360131_25, 2, 31K, 31K, 23
5, 2, VecPartIterator_29360131_24, 0, ExecutorState, 1, 16K, 16K, 23
0, 0, ExecutorState, 0, PortalHeapMemory, 0, 24K, 1163K, 20
1, 3, CStoreScan_29360131_22, 0, ExecutorState, 1, 390K, 1122K, 20
2, 20, cstore scan per scan memory context, 1, CStoreScan_29360131_22, 2, 476K, 476K, 20
3, 19, cstore scan memory context, 1, CStoreScan_29360131_22, 2, 264K, 264K, 20
4, 7, InitPartitionMapTmpMemoryContext, 1, CStoreScan_29360131_22, 2, 23K, 23K, 20
5, 2, VecPartIterator_29360131_21, 0, ExecutorState, 1, 16K, 16K, 20
```
The fields include the output SN, SN of the memory allocation context within the thread, name of the current memory context, output SN of the parent memory context, name of the parent memory context, tree layer No. of the memory context, peak memory used by the current memory context, peak memory used by the current memory context and all its child memory contexts, and plan node ID of the query where the thread is executed.

In this example, the record "1, 3, CStoreScan_29360131_22, 0, ExecutorState, 1, 390K, 1122K, 20" represents the following information about Explain Analyze:
- CstoreScan_29360131_22 indicates the CstoreScan operator.
- 1122K indicates the peak memory used by the CstoreScan operator.
fullexec: The generated file includes the information about all memory contexts requested by the execution layer.
If the parameter is set to fullexec, the output information will be similar to that for executor, except that some memory context allocation information may be returned because the information about all memory applications (no matter succeeded or not) is printed. As only the memory application information is recorded, the peak memory used by the memory context is recorded as 0.