Overview of introspection tools

Introspection tools for Cloud Spanner enable you to investigate issues with your database. They consist of a set of built-in tables that you can query to gain more insight about queries, transactions, reads and more. Not sure which tool to use for a particular problem? The following list summarizes each tool and the kinds of questions they can help answer.

Query statistics

When investigating issues in your database, it is helpful to know which queries are expensive, run frequently or scan a lot of data.

Query statistics are aggregated statistics for queries, gathered in 1, 10, and 60 minute intervals. Statistics are collected for queries that completed successfully as well as queries that failed, timed out, or were canceled by the user.

The statistics include highest CPU usage, total query execution counts, average latency, most data scanned, and additional basic query statistics. Use these statistics to help identify expensive, frequently run or data-intensive queries.

Oldest active queries

Sometimes you want to look at the current workload on the system by examining running queries. Use the Oldest active queries tool to investigate long running queries that may be having an impact on database performance. This tool tells you what the queries are, when they started running and to which session they belong.

Read statistics

Read statistics can be used to investigate the most common and most resource-consuming reads on your database using the Cloud Spanner Reads API. These statistics are collected and stored in 3 different time intervals - minute, 10 minutes and an hour. For each time interval, Cloud Spanner tracks the reads that are using the most resources.

Use read statistics to find out the combined resource usage by all reads, find the most CPU consuming reads, and find out how a specific read's frequency changes over time.

Transaction statistics

Transaction statistics can be used to investigate transaction-related issues. For example, you can check for slow-running transactions that might be causing contention or identify changes in transaction shapes that are leading to performance regressions. Each row contains statistics of all transactions executed over the database during 1, 10, and 60 minute intervals.

Lock statistics

Lock statistics can be used to investigate lock conflicts in your database. Used with transactions statistics, you can find transactions that are causing lock conflicts by trying to acquire locks on the same cells at the same time.

API methods included in each tool

In Cloud Spanner there is some overlap between transactions, reads and queries. Therefore, it might not be clear which API methods are included when compiling results for each introspection tool. The following table lists the main API methods and their relationship to each tool.

API Methods Transaction Modes Query statistics Oldest active queries Read statistics Transaction statistics Lock statistics
Read, StreamingRead Read-only transaction1 No No Yes No No
Read-write transaction No No Yes Yes Yes
ExecuteSql, ExecuteStreamingSql Read-only transaction1 Yes Yes No No No
Read-write transaction Yes Yes No Yes Yes
ExecuteBatchDml Read-write transactions No Yes4 No Yes Yes
Commit Read-write transactions (DML2, Mutations3) No No No Yes Yes

Notes:

1 Read-only transactions are not included in transaction statistics or lock statistics. Only read-write transactions are included in transaction statistics and lock statistics.

2 Uncommitted DML operations are not included in transaction statistics.

3 Empty mutations that are effectively no-op, are not included in transaction statistics.

4 Statement(s) within the batch will appear in oldest active queries, rather than the entire batch.