Find slow database queries

See which database statements your app spends its time on, how often each runs, its slowest calls, and where it runs as an N+1 loop.

Last updated 20 November 2026View as Markdown

The statements come from the db spans of your sampled transactions, so tracing must be on (see View traces and transactions). The SDK's integration for your database client (Postgres, MySQL, Prisma and others) creates those spans. To see every statement a Deutsche Cloud PostgreSQL database runs, not only those of sampled requests, and with no traces at all, connect the database to the project (see Connect a database to a project): the Queries tab then marks each statement Database + traces, Database only or Traces only. This page describes the tab without a connected database. Replace project-id with the project's id and fingerprint with a statement's fingerprint (sqlfp1: and 16 hexadecimal characters).

Prerequisites

  • Permission for errors:GetEvents on the project. Without it, the Queries tab says "You can see this project but not its queries."
  • In the console, also errors:ListProjects on the project to open it (without it the project is not found).

List the statements

Statements are grouped by their text without values, so WHERE id = 7 and WHERE id = 8 are one statement, and the Queries tab keeps and shows only that text. The db spans of the stored transactions keep the statement as the SDK sent it.

  1. Open the project, then the Queries tab.
  2. Choose 24 h or 7 days.

Two charts sit above the table. Calls per minute shows the calls of sampled requests, or, with a database connected, every call the database measured. Duration shows the p95 of the slowest call of each sampled request (a statement run once per request is exact; an N+1 reads high), and with a database connected also the database's mean per call as a dashed line. Each line's legend says which of the two it is.

Each row shows the Statement, its Time share of all database time in the project, the Calls/min (calls divided by the minutes of the range; marked sampled when only traces counted them), the Total, the Calls, the Mean and the Slowest call, and, when it ran as an N+1, in how many transactions.

An N+1 is a transaction that ran the same statement five or more times one after another under the same parent span. That is usually a loop that should be one query: a join, or WHERE id IN (…).

Open one statement

  1. On the Queries tab, choose a statement.

The page shows the statement's calls, calls per minute, mean, p95 and N+1 count, the same two charts for this statement alone (with a database connected, Calls per minute draws both lines: the gap between them is your sampling rate), and then:

  • Measured by the database: for each connected database you may read with rds:DescribeDBQueryInsights that ran it, what pg_stat_statements counted in the range: calls, calls per minute, mean, total, rows per call, cache hit and temporary data written, by database and role. Max and Std. deviation are since pg_stat_statements was last reset, not only the range. A link opens the statement in the database's Query insights.
  • The database at time: the database in the minute of the slowest request, or of the request you chose with Database at that minute (the minute in the middle of its slowest call, so a call that waited a minute on a lock shows the minute it waited in): connections in use against max_connections, active and idle-in-transaction sessions, sessions waiting on a lock, the longest open transaction, the cache hit, its calls and statement time, and the five statements that took the most time in that minute. Each number is compared with the median of the hour around it ("usually 3"). Sessions and locks are counted once, when the minute is sampled, so a lock shorter than a minute can fall between two counts. A minute the database was not sampled in says "No snapshot for this minute". Time a statement waits for a table lock before it starts is not in pg_stat_statements: a call the trace shows taking seconds can have a mean of under a millisecond in the database's numbers, and the lock waits of its minute are the place to look.
  • Slowest requests, Samples (the 20 most recent sampled requests that ran it) and N+1; each opens its trace.
  • By project and Where it ran, with a link to the database when it is one of yours.

A statement only the database ran (a background job, a migration, another app) has a page too: it says "No sampled request ran this statement." A statement no transaction ran and no database measured in the range says "This statement was not seen in the last 24 hours." with a link to Show 7 days. In a trace, a db span's details link to The database at this request's minute.

Did this page help you?