[knudtty]

Summary

Now that we use block numbers and offsets to enforce row uniqueness, we can just use that + a PK to identify a row. There is no need to specify Body or SeverityText and waste i/o and processing time.

Testing Locally

Steps:

1. yarn dev
2. Ensure your log source table contains the settings enable_block_number_column = 1, enable_block_offset_column = 1
3. Select a row
4. Verify the query that fired only has the PK and the block number & offset in the WHERE

References

• Linear Issue: Closes HDX-4287

[changeset-bot]

🦋 Changeset detected

Latest commit: d71f347

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 4 packages

Name	Type
@hyperdx/app	Patch
@hyperdx/cli	Patch
@hyperdx/api	Patch
@hyperdx/otel-collector	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
hyperdx-oss	Ready	Preview, Comment	May 21, 2026 3:13pm

[github-actions]

🔵 Tier 2 — Low Risk

Small, isolated change with no API route or data model modifications.

Why this tier:

• Standard feature/fix — introduces new logic or modifies core functionality

Review process: AI review + quick human skim (target: 5–15 min). Reviewer validates AI assessment and checks for domain-specific concerns.
SLA: Resolve within 4 business hours.

Stats

• Production files changed: 4
• Production lines changed: 77 (+ 264 in test files, excluded from tier calculation)
• Branch: aaron/select-only-pks-in-where
• Author: knudtty

To override this classification, remove the review/tier-2 label and apply a different review/tier-* label. Manual overrides are preserved on subsequent pushes.

[github-actions]

E2E Test Results

✅ All tests passed • 178 passed • 3 skipped • 1185s

Status	Count
✅ Passed	178
❌ Failed	0
⚠️ Flaky	4
⏭️ Skipped	3

Tests ran across 4 shards in parallel.

View full report →

[github-actions]

Deep Review

🔴 P0/P1 -- must fix

• packages/app/src/components/DBRowTable.tsx:1444 -- On a Distributed source backed by ReplicatedMergeTree, TableMetadata.engine_full mirrors the local table per packages/common-utils/src/core/metadata.ts:96-108, so hasBlockColumns opens the optimization even though _block_number/_block_offset are per-shard and not globally unique; pre-PR, including Body/attributes in the WHERE accidentally disambiguated across shards, so the lookup can now silently land on a row from a different shard than the one the user clicked.
  • Fix: Detect Distributed engines via tableMetadata.engine and either disable rowKeyColumns or include a shard discriminator (_shard_num) in extraKeys and the row WHERE for those sources.
  • _adversarial

🟡 P2 -- recommended

• packages/app/src/hooks/useRowWhere.tsx:181 -- When primaryKeyColumns is provided and the row keys do not intersect it, filteredRow becomes {}, processRowToWhereClause returns '', the where: rowId ?? '0=1' guard in packages/app/src/components/DBRowDataPanel.tsx:148 doesn't catch empty string (only null/undefined), and isNonEmptyWhereExpr('') in packages/common-utils/src/core/renderChartConfig.ts then drops the WHERE entirely, so the side panel renders an arbitrary LIMIT 1 row; the CLI variant in packages/cli/src/api/eventQuery.ts:371 has no 0=1 fallback at all.
  • Fix: Either throw / return '0=1' from useRowWhere/getRowWhere when filteredRow is empty, or replace rowId ?? '0=1' with an explicit isNonEmptyWhereExpr check in both DBRowDataPanel.tsx and eventQuery.ts.
  • _{correctness, testing, adversarial, kieran-typescript}
• packages/cli/src/components/EventViewer/useEventData.ts:263 -- The CLI's only caller of buildFullRowQuery never sets primaryKeyColumns, so the parameter plumbed through packages/cli/src/api/eventQuery.ts:341 and packages/cli/src/shared/useRowWhere.ts:180 is dead and the CLI/agent surface still issues the wide WHERE clause the PR set out to eliminate.
  • Fix: Compute rowKeyColumns in the CLI (resolve primary_key/partition_key/engine_full via Metadata, derive extraKeys exactly as useConfigWithAdditionalSelect does, also append those columns through an equivalent of appendSelectWithAdditionalKeys in buildEventSearchQuery), then thread the set into buildFullRowQuery.
  • _{correctness, testing, performance, agent-native}
• packages/app/src/components/DBRowTable.tsx:1472 -- rowKeyColumns entries from splitAndTrimWithBracket(partitionKey/primaryKey) (e.g. toStartOfFiveMinutes(Timestamp), json.a.b.:Int64) are matched character-exactly against ClickHouse result column names, yet the same file at lines 1640-1644 already warns that "CH may rewrite column names in the result set (e.g. expression formatting)", so PK-expression entries can silently fail to match and the WHERE collapses to the block columns alone.
  • Fix: Build rowKeyColumns from the same source that produces the SELECT list (so column-name identity is preserved by construction), or alias the appended raw expressions so the result column name is deterministic.
  • _{correctness, adversarial}
• packages/app/src/hooks/__tests__/useRowWhere.test.tsx:611 -- New "otel_logs"/"otel_traces" tests rely on .toContain('Timestamp') / .not.toContain('Body') substring checks against the full WHERE string, which would pass even if a regression dropped every PK predicate except one, or if a column name appeared inside a value literal.
  • Fix: Replace substring assertions with toBe/toEqual against the full expected WHERE string, or assert the count and LHS of AND-joined predicates structurally.
  • _testing
• packages/cli/src/shared/useRowWhere.ts:176 -- getRowWhere and buildFullRowQuery have no test coverage in the CLI package; the new primaryKeyColumns parameter is wired through both files but no test guards the parameter plumbing, the mirror semantics versus the web hook, or future drift.
  • Fix: Add at minimum one test file under packages/cli/src/shared/__tests__/ that exercises getRowWhere with and without primaryKeyColumns, paralleling the web suite.
  • _{testing, project-standards}
• packages/app/src/components/DBRowTable.tsx:1472 -- rowKeyColumns is recomputed verbatim from splitAndTrimWithBracket(partitionKey), splitAndTrimWithBracket(primaryKey), extractColumnReferencesFromKey(partitionKey/primaryKey), and extraKeys, which is the exact union appendSelectWithAdditionalKeys already computes as allKeys at line 1376; the two will drift the next time the membership rules evolve.
  • Fix: Have appendSelectWithAdditionalKeys also return the union set (e.g. { select, additionalKeysLength, keyColumns }), or extract a single buildRowKeyColumns(partitionKey, primaryKey, extraKeys) helper used by both sites.
  • _{maintainability}
• packages/app/src/hooks/useRowWhere.tsx:135 -- The parameter is named primaryKeyColumns but the set actually contains primary-key columns, partition-key columns, raw PK/partition expressions, and _block_number/_block_offset; the misnomer is reinforced by the config-side name rowKeyColumns, which is more honest.
  • Fix: Rename the hook/function parameter (and the CLI mirror) to rowKeyColumns so both the construction site and the consumers share one accurate name.
  • _{maintainability}

🔵 P3 nitpicks (10)

• packages/app/src/components/DBRowTable.tsx:1468 -- The hasBlockColumns gate bypasses the optimization entirely for older __hdx_id-only schemas even though __hdx_id is itself row-unique.
  • Fix: Build rowKeyColumns for the __hdx_id branch too if you intend to accelerate those schemas; otherwise leave a comment recording the deliberate scope.
• packages/app/src/components/DBRowTable.tsx:1472 -- rowKeyColumns is allocated as a fresh Set on every memo run, which flips useRowWhere's useCallback (deps include primaryKeyColumns), which in turn invalidates dedupedRows in RawLogTable for every parent re-render that touches config identity.
  • Fix: Memoize rowKeyColumns on primitive deps [primaryKey, partitionKey, hasBlockColumns, extraKeys.join(',')] rather than the full config object.
• packages/app/src/components/DBRowTable.tsx:1482 -- { ...config, select, additionalKeysLength, rowKeyColumns } exposes rowKeyColumns only via inferred return type, so any future typed boundary that re-narrows to BuilderChartConfigWithDateRange will silently strip the field and degrade to the full-scan path with no compile signal.
  • Fix: Declare an explicit MergedRowTableConfig interface extending the base config with select, additionalKeysLength, and rowKeyColumns, and use it for useConfigWithAdditionalSelect's return.
• packages/app/src/hooks/useRowWhere.tsx:135 -- Accepting Set<string> directly forces every caller to memoize the set themselves or silently lose useCallback stability.
  • Fix: Accept readonly string[] (or a sorted comma-joined key) and build/memoize the Set inside the hook.
• packages/cli/src/shared/useRowWhere.ts:176 -- The CLI mirror takes primaryKeyColumns as the 4th positional argument while the web hook takes it as a named field on the options object, leaving an unflagged stylistic drift between the two @source-linked copies.
  • Fix: Either match the named-arg shape now or document the intentional divergence in the @source header for the eventual consolidation into @hyperdx/common-utils.
• packages/app/src/components/DBRowTable.tsx:1393 -- The array branch of appendSelectWithAdditionalKeys skips the dedup-against-existing-select that the string branch performs, so the larger allKeys set now produced for block-column tables can emit duplicate { valueExpression } entries.
  • Fix: Apply the same selectColumns.has(k) filter to the array branch before mapping to { valueExpression }.
• packages/app/src/components/DBRowTable.tsx:1376 -- A source whose PK or partition key contains an expensive expression (heavy hash, UDF, JSON extraction) now pays that cost per row in the list-render query because the raw expression is appended to the SELECT alongside the bare column references.
  • Fix: Skip appending a raw expression whose extracted column references are already in the SELECT, or add the raw expression only when no extracted refs cover the partitioning.
• packages/app/src/components/DBRowTable.tsx:1448 -- No unit test covers the new hasBlockColumns branch in useConfigWithAdditionalSelect; a regression that always sets or never sets rowKeyColumns would be invisible.
  • Fix: Add tests that mount useConfigWithAdditionalSelect with engine_full strings that do and do not include both enable_block_*_column = 1 settings.
• packages/app/src/hooks/__tests__/useRowWhere.test.tsx:611 -- No test covers an aliased PK column in the SELECT (Timestamp AS ts) paired with primaryKeyColumns: Set(['Timestamp']), which is the natural way the filter under-constrains the WHERE.
  • Fix: Add a test that produces such a meta/row pair and asserts the WHERE is not silently empty.
• packages/app/src/hooks/useRowWhere.tsx:181 -- Object.fromEntries(Object.entries(dbRow).filter(...)) returns { [k: string]: any }, widening the typed rest-spread; the CLI mirror uses Record<string, unknown> end-to-end and is the better template.
  • Fix: Track toward Record<string, unknown> so this branch stops being an any ingress.

---

Reviewers (10): correctness, testing, maintainability, project-standards, agent-native, learnings-researcher, kieran-typescript, performance, api-contract, adversarial.

Testing gaps:

• No CLI test exercises getRowWhere/buildFullRowQuery with or without primaryKeyColumns.
• No test covers filteredRow becoming empty after the Set intersection (the empty-WHERE cascade).
• No test covers useConfigWithAdditionalSelect's hasBlockColumns branch directly.
• No integration test asserts the Set entries built JS-side actually byte-match ClickHouse result column names for the otel schemas.