[gianm]

This patch adds the context parameter "sqlTimeCondition", which adds a time filter to all tables. The filter is ANDed with any other filter that the user may have provided. The goal of this feature is to enable UIs and APIs to accept user-written SQL along with having a separate control for time.

This patch also fixes two pre-existing bugs that were discovered while writing tests for this feature:

1. In some cases, WHERE on time + ORDER BY non-time over a join could
   lose the time filter. This is fixed by passing scan.getIntervals()
   to WindowOperatorQuery in DruidQuery#toScanAndSortQuery.
   testLeftJoinWithTimeFilterAndNonTimeOrderBy is a regression test.

2. The SQL planner would produce "intervals" rather than a __time filter
   for certain time filters over a join of a left-hand subquery. This
   would cause execution to fail, because execution assumes that
   "intervals" are only used for regular tables. This is fixed by
   adjusting ExecutionVertexExplorer to avoid traversing through
   noncollapsible subqueries, making such a query no longer viewed
   by the SQL planner as table-based. The regression test is in
   testJoinWithSubqueryOnLeftAndOuterTimeFilter.

[FrankChen021]

🟡 Changes recommended

sqlTimeCondition is not honored for policy-protected tables in native SQL, so supported queries can silently read outside the requested intervals. Extend interval eligibility to policy-wrapped tables and add a regression test before merging.

Reviewed 24 of 24 changed files; findings in 1 file.

Severity	Findings
P0	0
P1	1
P2	0
P3	0
Total	1

---

This is an automated review by Codex GPT-5.6-Luna(max)

After addressing the findings or replying to the comments, you can request another review from me to trigger a new automated review.

[FrankChen021]

[P1] Context is ignored for policy-wrapped tables

Finding: Native SQL applies policies before this method, representing protected tables as RestrictedDataSource. ExecutionVertex.isTableBased() returns false for that wrapper, so the early return bypasses the new timeCondition interval and the native engine reads outside the requested window.

Suggestion: Treat RestrictedDataSource as table-based for interval filtering, and add a restricted-table regression test.

[abhishekrb19]

Could this also be defined as a SQL server configuration so it applies to all queries cluster-wide, similar to druid.sql.planner.requireTimeCondition? Or is that already available through druid.query.default.context.sqlTimeCondition?

[abhishekrb19]

I also wonder how this feature would interplay with the existing requireTimeCondition feature - fixes being made in #20441.

@gianm @NishthaShah thoughts?

[FrankChen021]

The conflict is from the #20578 which migrated the sql context parameterer to the new QueryContextParameters catalog. We can follow the new API to register new context parameter.

[FrankChen021]

I also want to know if the supporting of ARRAY is needed.
Because under current design, every query context parameter can be set via the SET statement, however, the SET statement requires a literal , we CAN'T do sth like this now:

SET sqlTimeCondition = ['a/b', 'c/d'];

The only case that current simple String type can't express is the empty array case. But I'm not sure under which case we need to support this case,.

If this is a must case, I think maybe we can following existing SEGMENT_SORT_ORDER which accepts JSON array inside string, like:

SET sqlTimeCondition = ''; -- error
SET sqlTimeCondition = '[]';                          -- match nothing
SET sqlTimeCondition = '["2000-01-01/2000-02-01","2001-01-01/2001-02-01"]';  
SET sqlTimeCondition = '2000-01-01/2000-02-01,2001-01-01/2001-02-01';

[kgyrtkirk]

instead of a new argument to a lot of things; couldn't this be folded into the filter and then let the system recognize it like in other cases?

....or how about instead of passing it around transform all datasources of the queries into filtereddatasource?

I also wonder what would be the logical equiv of this:

• I think this could be viewed that all tables have logical column like: __phantom_time ; which applies to the subtree
• this condition is applies to the __phantom_time column
• the __phantom_time conditions should be pushed down to all leaf filters; regardless the intermediate nodes (rewrite shuttle) and turned into __time
• this __phantom_time column is conceptual - if it would have no value exposing it could be skipped entirely; the shuttle would be enough

[kgyrtkirk]

I think the source of a detail like this should not matter at this point