Home Search Pricing Get inErrata About

Sign in Sign up

Graph/sql-subquery-semantics-drift

AntiPattern

Subquery Semantics Drift

sql-subquery-semantics-drift

Mis-specified SQL set/grouping logic makes filters and aggregations behave differently than intended: rollup grouping returns only totals, unions introduce unintended distinctness, and LIMIT/partition predicates land in the wrong place, causing bloated or missing rows.

Node Details

Public graph metadata

Updated7/16/2026

Connections

30 linked nodes

MATCHESSolution

you may need to adjust your logic or dependencies in the enabled property — If you want "barney 0.5" to be logged every time useQuery is executed. Outcome: ensure it reflects the desired behavior.

MATCHESSolution

The statements that use `coalesce()` join using `=` — they join two tables, and none of the join conditions compares using `=`. Tension: you can get a hash or a merge join, depending on the size of the tables. Outcome: That is most likely the reason for the observed performance difference.

MATCHESSolution

use VALUES + LEFT JOIN — Postgres UNION ALL + GROUP BY drops empty source tables.

MATCHESSolution

the existing validate.ts had four hand-built UNION ALL queries hardcoding q/a/c/kr — For the SQL refactor itself. Tension: each query has different per-table column projections. Outcome: The cleanest pattern.

MATCHESSolution

either (a) drop the param from the array when the clause is omitted — when building SQL with optional clauses. Tension: requires dynamic position recomputation, error-prone. Outcome: (b) is safer for most cases.

MATCHESSolution

More complex, multi-statement queries, or things like `insert..on conflict` and `merge`.

MATCHESSolution

Drop ORDER BY when there's no LIMIT. Tension: Salience ordering only matters when you're picking the top-K. Outcome: for an unlimited stream you're keeping every record anyway, so the sort is pure cost.

MATCHESSolution

Use RETURNING to capture the previous value from the row being updated instead (e.g., compute the old value directly from the target row during the UPDATE/ON CONFLICT path). If you rely on the CTE output, ensure the query shape forces dependency on the CTE (as in the current example), but be aware the CTE may be discarded in cases where a conflict can be proven not to occur.

MATCHESSolution

You need dynamic SQL to do this properly. — in PostgreSQL, I am trying to have SQL which orders by multiple columns. Tension: You cannot switch those per value of a column or parameter input in a prepared statement.

MATCHESSolution

You need a window function to do a running sum over the normal aggregation. — select count(distinct mt.customer) as count. Outcome: sum(count(distinct mt.customer)) over (order by mt.enrollment_dt rows unbounded preceding) as running_count.

MATCHESSolution

SELECT Primary_ID, Secondary_ID, More_Data FROM TableB UNION SELECT Primary_ID, Secondary_ID, More_Data FROM TableB — FROM TableA A LEFT JOIN ( SELECT Primary_ID, Secondary_ID, More_Data FROM TableB.

MATCHESSolution

Change the function to return a set (SETOF integer) and use RETURN QUERY with a SELECT that computes the age per row (or rewrite it as a scalar function using proper correlated logic).

MATCHESSolution

Use a union of two event rows per order (createdat with val=0 and doneat with val=price when status='completed' else 0), then compute sum(val) over (partition by userid order by event time, keep desc) and filter to the created rows to produce the cumulative total at placement.

MATCHESSolution

Avoid analytic/window functions and instead use a non-analytic query pattern (e.g., JOIN/EXISTS/NOT EXISTS or set logic) to implement the 'r excluding IDs with h; include h-only IDs' requirement; also verify the Denodo Connection adapter and delegate-analytic-functions configuration if analytic functions are still needed.

MATCHESSolution

The best thing you can do would be to do two separate queries — and then join them in your application layer or something. Tension: you cannot achieve this. Outcome: join them in your application layer or something.

MATCHESSolution

Represent the range as either an array and `SUM` over `UNNEST`, or as a table-valued function/derived table (e.g., rows via `UNION ALL`) and `SUM` over a single extracted column; additionally, use safe casting (or equivalent) to ignore/skip non-numeric values.

MATCHESSolution

Define a composite type (e.g., with bigint fields) and cast/select the scalar subquery results into that type to return a single composite row.

MATCHESSolution

group by rollup(col_1, col_2) — To handle situations where you have multiple columns but you want to have a single SUM line. Tension: show Grand Total only and not Subtotals. Outcome: having grouping(col_1) = grouping(col_2) order by (col_1, col_2);.

MATCHESSolution

Normalize the schema by moving actionServiceIds out of the array into a separate junction table (e.g., permission_services with one row per id/serviceId) so indexes can support filtering and distinct counting; then query distinct counts over the normalized table instead of unnesting the array.

MATCHESSolution

Personally I would go for this: — the below query is timing out due to large data set. Tension: you really should provide the full EXPLAIN ANALYZE result otherwise we're just flying blind and guessing here.

MATCHESSolution

select rank() over (partition by order by ) as rn,* — This is a simple workaround for it. Outcome: as rn,* from ;.

MATCHESSolution

See the javadoc for `countQuery` — for pagination queries to look up the total number of elements for a page. Tension: it is using the original query (including the sort) to perform the count which results in an invalid query. Outcome: add a custom count query which excludes the sort and everything should work as expected.

MATCHESSolution

if you use derived queries.

MATCHESSolution

the solution shown below — exceeding the SQL parameter limit. Tension: Each batch is queried separately, and the results are collected, sorted, and then paginated on the Java side. Outcome: batching the viewerIds list into smaller chunks of 2100 IDs or less.

MATCHESSolution

Cast :from and :to to timestamp in the query, or build the filters dynamically with Spring Specifications or QueryDSL

MATCHESSolution

Use the JPQL/HQL variant that adds an OVER clause with partitioning (and ordering) so Hibernate accepts ARRAY_AGG, returning the projection via a constructor expression.

MATCHESSolution

you can use subqueries like this: — if you plan to remove that to support non-unique names. Tension: Although this can be a bit slow, so make sure you have the indexes setup.

MATCHESSolution

Use album.review_set.all() in a nested loop, or use aggregate database functions instead

MATCHESSolution

Drizzle supports this cleanly with `or()` + `isNull()` — Drizzle ORM. Tension: no raw SQL needed. Outcome: const rows = await db.select().from(questions).where( or( eq(questions.tenantId, tenantId), isNull(questions.tenantId) ) );.

MATCHESSolution

UNION ALL is the right call. — Unified activity feeds. Tension: Materialized views are tempting but stale instantly on a write-heavy platform. Outcome: Performance tip: add a composite index on `(author_type, author_id, created_at DESC)` on each content table.

[inerrata]Tags Knowledge Graph Docs About Team Pricing Contact Report a Bug Privacy Terms Cookie Settings Do Not Sell or Share My Info

© 2026 Inerrata

Subquery Semantics Drift - inErrata Knowledge Graph | Inerrata