View Astra DB Serverless metrics
You can view Astra DB Serverless health metrics to get insight into database performance and workload distribution.
Organizations on paid plans can also export metrics and scrape metrics to third-party services.
View metrics for Serverless (vector) databases
-
In the Astra Portal, click the name of the database that you want to inspect.
If you have both Serverless (vector) and Serverless (non-vector) databases, be aware that there are different metrics dashboards for each deployment type. The Serverless icon denotes Serverless (non-vector) databases.
-
On the database’s Overview tab, view the Key metrics:
- Total Latency by Percentile
-
Indicates the processing time for read and write requests as histogram quantiles of 50 or 99.
p50 Reads/Writes reports median latency, which means that half of the requests were processed faster than this value. p99 Reads/Writes reports 99th percentile latency, which means that only 1% of requests were slower this value.
Read and write latencies display in nanoseconds.
- Total Throughput
-
The number of requests processed per second, calculated as a sum of read and write throughput. This widget also reports the average throughput per second.
Gaps in read/write metrics indicate periods when there were no requests.
-
To view more granular metrics, hover over a specific time in the graph.
-
To change the reporting period, use the schedule Time Frame Section menu. The default time range is the last 10 minutes.
-
For multi-region databases, use the Region menu to inspect data for a different region.
For more metrics, see Scrape Astra DB Serverless metrics in exposition format and Export Astra DB Serverless metrics.
For databases in PCU groups, see Monitor PCU groups in Astra.
View metrics for Serverless (non-vector) databases
-
In the Astra Portal, click the name of the database that you want to inspect.
If you have both Serverless (vector) and Serverless (non-vector) databases, be aware that there are different metrics dashboards for each deployment type. The Serverless icon denotes Serverless (non-vector) databases.
-
Click the Health tab to explore database metrics in the embedded Grafana viewer.
To view granular metrics for time-series graphs, hover over a specific time in the graph.
Gaps in read/write metrics indicate periods when there were no requests.
For more information about specific widgets or metrics, see Serverless (non-vector) metrics widgets and Astra DB Serverless metrics reference. For databases in PCU groups, see Monitor PCU groups in Astra.
-
To sort and filter the data, do the following:
-
For multi-region databases, use the Region menu to inspect data for a different region.
-
To change the reporting period, use the schedule Time Frame Section menu. The default time range is the last 30 minutes. For more information about time controls, see the Grafana documentation.
-
You can click Refresh to refresh the dashboard. The default refresh interval is one minute.
-
-
To drill down into a widget’s data, hover over the widget, click More, and then select an option:
-
Inspect: Inspect timestamped data or raw JSON data. You can also export the data as a CSV.
-
View: View the widget in full screen mode. To navigate back to the dashboard view, use the breadcrumbs in the header of the Grafana viewer.
-
-
To set up alerts or switch dashboards, click Toggle menu, and then click Dashboards or Alerting.
Serverless (non-vector) metrics widgets
When viewing a Serverless (non-vector) database, the Health tab includes the following sections and widgets.
- Request Overview section
-
-
Requests Combined: Displays request rates (requests per second) for different types of requests. The rates are summed over all database coordinators.
-
Request Errors: Displays the request error rates (requests per second) for different types of requests. The rates are summed over all database coordinators.
Timeouts indicate that the queries are taking too long to complete.
Unavailables indicate that the coordinator did not have enough alive data nodes to work with.
Failures can be caused by queries violating certain guardrails or other error conditions. For more information, see Astra DB Serverless database limits.
-
Average Request Complexity: This widget provides a high-level view of the average complexity of database requests. It measures the ratio of the average request units to the average request rate. A higher complexity score indicates that a request is more resource-intensive and expensive in terms of read/write activity. This metric combines write operations, including regular writes, counter writes, and index writes, to provide a comprehensive view of the write request complexity.
-
- Writes section
-
-
Write Latency: Displays coordinator write request latency quantiles on the left y-axis and the total write request rate (requests per second) on the right y-axis.
-
Write Size Distribution: Displays different write request mutation size quantiles. Large mutations can cause performance problems and might even be rejected.
-
- Reads section
-
-
Read Latency: Displays coordinator read request latency quantiles on the left y-axis and the total read request rate (requests per second) on the right y-axis.
-
Range Latency: Displays coordinator range request latency quantiles on the left y-axis and the total range request rate (requests per second) on the right y-axis.
-
- Stargate section
-
-
Connected Clients: Reports the number of CQL connections for the database.
-
- Lightweight Transactions (LWTs) section
-
-
Compare and Set (CAS) Write Latency: Displays coordinator CAS write request latency quantiles on the left y-axis and the total CAS write request rate (requests per second) on the right y-axis.
-
CAS Write Contention: Displays coordinator CAS write request contention quantiles on the left y-axis and the number of unfinished commits on the right y-axis.
A high number of contended requests can negatively affect request latency and cause timeouts. Reduce the number of concurrent requests to the same partition. Unfinished commits cause increased latency. Reducing contention can help reduce the number of unfinished commits.
-
CAS Read Latency: Displays coordinator CAS read request latency quantiles on the left y-axis and the total CAS read request rate (requests per second) on the right y-axis.
-
CAS Read Contention: Displays coordinator CAS read request contention quantiles on the left y-axis and the number of unfinished commits on the right y-axis.
A high number of contended requests can negatively affect request latency and cause timeouts. Reduce the number of concurrent requests to the same partition. Unfinished commits cause increased latency. Reducing contention can help reduce the number of unfinished commits.
-
- Tombstones section
-
-
Tombstones Scanned per second: Displays the number of tombstones being scanned per keyspace, table, and second. A large number of tombstones can cause increased latency or query failures.
-
Tombstone Guardrail Warnings per second: Displays the number of queries exceeding the tombstone guardrail warning threshold per keyspace, table, and second. For more information, see Astra DB Serverless database limits.
-
Tombstone Guardrail Failures per second: Displays the number of queries exceeding the tombstone guardrail failure threshold per keyspace, table, and second. For more information, see Astra DB Serverless database limits.
-
