perf(otel-sqlite): add stored session_id column to eliminate full-table scans

[makroumi]

Problem

Every session-scoped query in OTelSqliteStore performed a full table scan on spans.

Root causes confirmed with EXPLAIN QUERY PLAN:

1. COALESCE(conversation_id, chat_session_id) in WHERE and GROUP BY; SQLite cannot use an index on a function-wrapped column. Both idx_spans_conversation and idx_spans_chat_session were defined but unreachable for these queries.

2. OR across two columns (conversation_id = ? OR chat_session_id = ?): the query planner cannot satisfy this with a single index seek.

Affected: getSpansByConversationId, getTraceIds, the sessions view (used by getSessions and getSessionsSince), and both cleanup subqueries in _cleanupOnStartup.

Solution

Add session_id TEXT GENERATED ALWAYS AS (COALESCE(conversation_id, chat_session_id)) STORED to spans. SQLite computes and persists this at write time. No changes to INSERT paths or call sites.

Add idx_spans_session ON spans(session_id, start_time_ms) — composite index covering the equality lookup and the ORDER BY start_time_ms used in getSpansByConversationId.

Changes

Location	Before	After
getSpansByConversationId	WHERE conversation_id = ? OR chat_session_id = ?	WHERE session_id = ?
getTraceIds	WHERE conversation_id = ? OR chat_session_id = ?	WHERE session_id = ?
sessions view	GROUP BY COALESCE(conversation_id, chat_session_id)	GROUP BY session_id
_cleanupOnStartup cutoff	GROUP BY COALESCE(...)	GROUP BY session_id
_cleanupOnStartup DELETE	NOT IN subquery	NOT EXISTS — NULL-safe by structure
SCHEMA_VERSION	1	2

Migration

v1 → v2 uses the standard SQLite table rebuild pattern: rename existing table → recreate with generated column → INSERT INTO ... SELECT with explicit 24-column list (session_id excluded, it is GENERATED ALWAYS) → drop old table → recreate indexes → drop and recreate sessions view. Required because SQLite does not support ALTER TABLE ADD COLUMN for generated columns.

Verification

EXPLAIN QUERY PLAN
SELECT * FROM spans WHERE session_id = 'x';
→ SEARCH spans USING INDEX idx_spans_session (session_id=?)

EXPLAIN QUERY PLAN
SELECT session_id, MIN(start_time_ms) FROM spans
WHERE session_id IS NOT NULL GROUP BY session_id;
→ SEARCH spans USING INDEX idx_spans_session (session_id=?)

Migration verified: row count preserved, session_id correctly resolves to conversation_id when present, chat_session_id as fallback, null when both are absent. sessions view aggregates correct post-migration.

[Copilot]

Pull request overview

This PR improves performance of the Copilot extension’s OTelSqliteStore SQLite queries by introducing a persisted session_id generated column and a composite index, allowing session-scoped lookups and aggregations to use indexed seeks instead of full-table scans.

Changes:

• Bump schema to v2 and add session_id TEXT GENERATED ALWAYS AS (COALESCE(conversation_id, chat_session_id)) STORED plus idx_spans_session(session_id, start_time_ms).
• Rewrite session-scoped queries (getSpansByConversationId, getTraceIds) and the sessions view to use session_id.
• Update startup cleanup logic to group by session_id and switch the retention delete to NOT EXISTS.

[makroumi]

@microsoft-github-policy-service agree