Performance Insights concepts - Amazon DocumentDB
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Performance Insights concepts

Average active sessions

Database load (DB load) measures the level of activity in your database. The key metric in Performance Insights is DB Load, which is collected every second. The unit for the DBLoad metric is the Average Active Sessions (AAS) for a DocumentDB instance.

An active session is a connection that has submitted work to the DocumentDB instance and is waiting for a response. For example, if you submit a query to a DocumentDB instance, the database session is active while the instance is processing the query.

To obtain the average active sessions, Performance Insights samples the number of sessions concurrently running a query. The AAS is the total number of sessions divided by the total number of samples. The following table shows five consecutive samples of a running query.

Sample Number of sessions running query AAS Calculation




2 sessions / 1 sample




2 sessions / 2 samples




6 sessions / 3 samples




6 sessions / 4 samples




10 sessions / 5 samples

In the preceding example, the DB Load for the time interval from 1-5 is 2 AAS. An increase in DB load means that, on average, more sessions are running on the database.


The DB Load metric is different from the other time-series metrics because you can break it into subcomponents called dimensions. You can think of dimensions as categories for the different characteristics of the DB Load metric. When you are diagnosing performance issues, the most useful dimensions are wait states and top query.

wait states

A wait state causes a query statement to wait for a specific event to happen before it can continue running. For example, a query statement might wait until a locked resource is unlocked. By combining DB Load with wait states, you can get a complete picture of the session state. Here are various DocumentDB wait states:

DocumentDB wait state Wait State Description


The Latch wait state occurs when the session is waiting to page the buffer pool. Frequent paging in and out of the buffer pool can happen more often when there are frequent large queries being processed by the system, collection scans, or when the buffer pool is too small to handle the working set.


The CPU wait state occurs when the session is waiting on CPU.


The CollectionLock wait state occurs when the session is waiting to acquire a lock on the collection. These events occur when there are DDL operations on the collection.


The DocumentLock wait state occurs when the session is waiting to acquire a lock on a document. High number of concurrent writes to the same document will contribute to more DocumentLock wait states on that document.


The SystemLock wait state occurs when the session is waiting on the system. This can occur when there are frequent long running queries, long running transactions, or high concurrency on the system.


The IO wait state occurs when the session waiting on IO to complete.


The BufferLock wait state occurs when the session is waiting to acquire a lock on a shared page in the buffer. BufferLock wait states can be prolonged if other processes are holding open cursors on the requested pages.


The LowMemThrottle wait state occurs when the session is waiting due to heavy memory pressure on the Amazon DocumentDB instance. If this state persists for a long time, consider scaling up the instance to provide additional memory. For more information, see Resource Governor.


The BackgroundActivity wait state occurs when the session is waiting on internal system processes.


The Other wait state is an internal wait state. If this state persists for a long time, consider terminating this query. For more information, see How Do I Find and Terminate Long Running or Blocked Queries?

Top queries

Whereas wait states show bottlenecks, top queries show which queries are contributing the most to DB load. For example, many queries might be currently running on the database, but a single query might consume 99% of the DB load. In this case, the high load might indicate a problem with the query.

Max vCPU

In the dashboard, the Database load chart collects, aggregates, and displays session information. To see whether active sessions are exceeding the maximum CPU, look at their relationship to the Max vCPU line. The Max vCPU value is determined by the number of vCPU (virtual CPU) cores for your DocumentDB instance.

If the DB load is often above the Max vCPU line, and the primary wait state is CPU, the CPU is overloaded. In this case, you might want to throttle connections to the instance, tune any queries with a high CPU load, or consider a larger instance class. High and consistent instances of any wait state indicate that there might be bottlenecks or resource contention issues to resolve. This can be true even if the DB load doesn't cross the Max vCPU line.