Inspections and recommendations in Yandex MPP Analytics for PostgreSQL
Managed database clusters regularly undergo diagnostics to detect possible issues, increase the cluster's reliability, and improve its performance. The results of such checks are displayed as inspections under Recommendations. You can see notifications about successful checks and recommendations on eliminating discovered risks. The responsibility to troubleshoot any detected issues lies within the Yandex Cloud user's remit.
All checks have a severity level:
- High level: Criteria of high cluster availability, significant risks of reduced performance, data loss risks. Such checks warrant special attention and require following the recommendations provided.
- Moderate level: Possible risks of reduced performance, suboptimal memory and disk space usage.
- Low level: Potential risks and cluster operation limitations.
The list of recommendations is updated regularly. For each recommendation, the following timestamps are fixed: the date of the risk's first detection and the date of the most recent status update on this issue. If the recommendation seems to be excessive or incorrect, you can hide it, specifying a reason. Once the hiding period expires, the recommendation will automatically become available again if the issue persists.
Recommendations are available at the cluster, folder, and cloud levels and provide tips for all your resources. However, the absence of recommendations does not mean that your cluster is optimized: the list of checks gets continuously appended but still remains incomprehensive and cannot replace monitoring, since it is targeted at detecting patterns rather than specific issues. You can additionally run cluster performance diagnostics and analyze monitoring metrics.
To manage recommendations in Yandex MPP Analytics for PostgreSQL, you need the managed-greenplum.editor role or higher.
Inspections available in Yandex MPP Analytics for PostgreSQL
| Category | Check | Risk | Severity |
|---|---|---|---|
| Performance | CPU usage | Risk of degraded performance | High |
| High availability | Memory allocation | Lack of RAM | Moderate |
| High availability | ZSTD memory accounting | Risk of exceeding resource manager memory limits | Moderate |
Risk of reduced performance
Description
The host continuously uses all CPU resources, which may cause delays in query processing. Check the load and optimize your queries (which may include using the WebSQL AI assistant) or increase computing resources of the cluster.
Action
To increase the cluster's computing resources:
- Navigate to Yandex MPP Analytics for PostgreSQL.
- Select your cluster and click Edit.
- Under Resources, select a host class with the required amount of vCPUs.
- Click Save changes.
Not enough RAM
Description
The cluster is running out of RAM, which leads to slow performance or emergency shutdowns. Increase the cluster's RAM.
Action
To increase the amount of RAM:
- Navigate to Yandex MPP Analytics for PostgreSQL.
- Select your cluster and click Edit.
- Under Resources, select a host class with the required amount of RAM.
- Click Save changes.
Risk of exceeding resource manager memory limits
Description
The gp_enable_zstd_memory_accounting parameter controls memory allocation for the ZSTD algorithm. Enabling it prevents the system from exceeding the resource manager memory limits by allocating ZSTD to a separate zstd_context, which significantly reduces the risk of cluster failure due to OOM. We recommend enabling this parameter.
Action
To enable the gp_enable_zstd_memory_accounting parameter:
- Navigate to Yandex MPP Analytics for PostgreSQL.
- Select your cluster and click Edit.
- Click Settings under DBMS settings and enable
gp_enable_zstd_memory_accounting.