Open Query Insights
From the PlanetScale dashboard, select a Neki database and branch, then select Insights. The page opens to the last 24 hours. Use the branch selector to choose the branch you want to analyze.Explore query activity
The graph provides the following views:- Query latency shows p50 and p95 latency by default. You can also display p99, p99.9, and maximum latency.
- Queries shows the query rate over the selected period.
- Rows read and Rows written show row activity over the selected period.
- The final tab graphs one additional query metric that you select.
Query patterns
Insights replaces literal values with numbered placeholders so executions that differ only by their values can be grouped together. For example, these executions belong to the same pattern:search_path resolution that produced a different set of relations, and again
for the replica traffic that ran it.
Pattern cardinality
Each router tracks up to 2,000 patterns per aggregation interval. Additional patterns are combined into an overflow entry, so not every distinct pattern is listed. A branch that runs many distinct statement shapes, such as generated SQL with inlined literals that normalization cannot collapse, reaches this limit most often. The table below the graph summarizes each pattern. Use the Overview, Data, Resources, and Performance presets to switch between related columns, or select Custom to choose columns individually. Numeric columns can display sparklines for the selected time range. The default Overview preset includes the query, percentage of runtime, execution count, total time, p50 and p99 latency, rows read, and the ratio of rows read to rows returned.Table and schema names in Neki
A Postgres query can use an unqualified table name, such asorders, while
search_path determines the schema that Postgres actually uses. Four columns
describe the result of that resolution:
For example,
SELECT * FROM orders can resolve to appdb.sales.orders. Use
Qualified table when you need to know which database object Neki queried. If
a pattern accesses multiple tables, Qualified table and Table schema
list a value for each of them.
Schema is not a PostgreSQL schema name on Neki. Its value combines the
connected database with an identifier for what the statement’s names resolved
to, so two rows with the same SQL and different Schema values resolved to
different relations. Read the actual schema from Table schema or
Qualified table.
Available query statistics
The Neki query table can show the following groups of statistics. A column is shown only when its data is collected for the selected branch.
Shard calls per query is the average number of shard dispatches per
execution of the pattern, and Max shard calls per query is the highest
number recorded for a single execution. A query scattered to four shards
records four shard calls. Several statements produced by a rewrite within one
dispatch count as one shard call, and repeated dispatches to the same shard
count separately, so these metrics are not a count of distinct shards.
Buffered or replayed work and failed scatter executions can make the reported
count lower than the work performed.
A value greater than one indicates either multi-shard execution or repeated
work on a shard. Review the query plan and the database’s
data topology when investigating an unexpected value.
Parallel workers per query is the average number of Postgres backend
processes that executed the statement, counted across every shard it reached.
The count includes the leader process, so a statement that ran without
parallelism on a single shard reports one, and a statement that reached several
shards adds up each shard’s processes. Parallel maintenance workers, such as
those used by an index build, count the same way. This describes process use
inside Postgres and is separate from the number of shard calls.
Filter query patterns
The filter field accepts free text and named terms. Select SYNTAX beside the field to see the operators the dashboard supports.Filter on tables and schemas
Four named terms filter on the relations a pattern resolved to. All of them match case-insensitively.qualified_table:orders finds patterns that name orders with an explicit
qualifier in their SQL. It accepts a bare relation name only, so use table:
when you want to filter by a database or schema prefix.
Free text
A term with no name matches the normalized SQL with a case-insensitive substring comparison. Free text does not search the resolved table metadata, soappdb.sales.orders finds only patterns whose SQL text contains that string,
not every pattern that resolved to that relation. Use table: or
table_schema: for the resolved relation instead.
A named term whose value the term type rejects becomes free text as well. Because
qualified_table: accepts only a bare relation name,
qualified_table:appdb.sales.orders searches SQL text rather than table
metadata.
Other filter behavior
Latency filter values are in milliseconds. Wrap text in double quotes for an exact string match, and prefix a term with! to exclude it. Tag filters are
also available when query dimensions are enabled for the branch.
Query-pattern details
Select a query pattern to open its detail page. The page shows the normalized SQL and graphs the selected pattern separately from the rest of the branch. It also provides summary statistics for count, total time, rows read relative to rows returned, p50 latency, p99 latency, and errors. Depending on the statement and the data collected for the branch, the detail page can also show:- Index usage over the selected time range.
- Notable executions that were slow, read a large number of rows, or returned an error.
- Query tags and their values.
- An optional AI-generated summary of the normalized query. See How PlanetScale uses AI.
Anomalies and schema recommendations
Insights also surfaces two related views for every Neki cluster:- Anomalies flags periods when a high share of queries run slower than their established baseline.
- Schema recommendations suggest DDL that can improve performance, reduce storage, or prevent ID exhaustion.