Skip to content

:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.

fix(mcp-analytics): count users by distinct_id to avoid person join oom (#121984)

Georgis Andonis9eb904c6c36b
diff --git a/products/mcp_analytics/backend/dashboard_templates.py b/products/mcp_analytics/backend/dashboard_templates.py
index 238d7c99f016..bf7bdacd3d42 100644
--- a/products/mcp_analytics/backend/dashboard_templates.py
+++ b/products/mcp_analytics/backend/dashboard_templates.py
@@ -59,7 +59,7 @@ def get_mcp_analytics_default_template() -> DashboardTemplate:
"name": "$mcp_initialize",
"custom_name": "users",
"math": "hogql",
- "math_hogql": "count(DISTINCT person_id)",
+ "math_hogql": "count(DISTINCT distinct_id)",
},
{
"kind": "EventsNode",
diff --git a/products/mcp_analytics/backend/tests/test_dashboard_templates.py b/products/mcp_analytics/backend/tests/test_dashboard_templates.py
index 3262e6812355..df25cda44900 100644
--- a/products/mcp_analytics/backend/tests/test_dashboard_templates.py
+++ b/products/mcp_analytics/backend/tests/test_dashboard_templates.py
@@ -39,7 +39,7 @@ def test_mcp_analytics_default_template_includes_users_and_sessions() -> None:
"name": "$mcp_initialize",
"custom_name": "users",
"math": "hogql",
- "math_hogql": "count(DISTINCT person_id)",
+ "math_hogql": "count(DISTINCT distinct_id)",
},
{
"kind": "EventsNode",
diff --git a/products/mcp_analytics/frontend/leaderboardHome/mcpLeaderboardHomeLogic.ts b/products/mcp_analytics/frontend/leaderboardHome/mcpLeaderboardHomeLogic.ts
index 16d2b278e0a4..5cb189ecd3da 100644
--- a/products/mcp_analytics/frontend/leaderboardHome/mcpLeaderboardHomeLogic.ts
+++ b/products/mcp_analytics/frontend/leaderboardHome/mcpLeaderboardHomeLogic.ts
@@ -42,6 +42,10 @@ const TOOL_CALL_WHERE = `event = '$mcp_tool_call'
const BUCKETED_ROW_LIMIT = 10000
const WINDOW_ROW_LIMIT = 1000
+// The users counts below use distinct_id, not person_id, for the same reason as USERS_QUERY in
+// mcpDashboardOverviewLogic.ts: the person_id join over every person override runs out of memory
+// on large projects.
+
const labelOf = (valueSql: string): string => `coalesce(nullIf(trim(${valueSql}), ''), 'Unknown')`
const propertyLabel = (property: string): string => labelOf(`toString(properties.${property})`)
@@ -71,7 +75,7 @@ const windowFacetQuery = (label: string, onlyErrors = false): string => `
SELECT
${label} AS label,
count() AS calls,
- uniq(person_id) AS users,
+ uniq(distinct_id) AS users,
countIf(toBool(properties.$mcp_is_error)) AS errors
FROM events
WHERE ${TOOL_CALL_WHERE}
@@ -84,14 +88,14 @@ LIMIT ${WINDOW_ROW_LIMIT}
const MODEL_LABEL = propertyLabel('$mcp_llm_model')
const labUsersQuery = `
-SELECT ${labSqlExpression(MODEL_LABEL)} AS lab, uniq(person_id) AS users
+SELECT ${labSqlExpression(MODEL_LABEL)} AS lab, uniq(distinct_id) AS users
FROM events
WHERE ${TOOL_CALL_WHERE}
GROUP BY lab
`
const namedModelUsersQuery = `
-SELECT uniqIf(person_id, ${MODEL_LABEL} != 'Unknown') AS users
+SELECT uniqIf(distinct_id, ${MODEL_LABEL} != 'Unknown') AS users
FROM events
WHERE ${TOOL_CALL_WHERE}
`
diff --git a/products/mcp_analytics/frontend/mcpDashboardOverviewLogic.ts b/products/mcp_analytics/frontend/mcpDashboardOverviewLogic.ts
index 56bb3d274bfa..d013c01d9d49 100644
--- a/products/mcp_analytics/frontend/mcpDashboardOverviewLogic.ts
+++ b/products/mcp_analytics/frontend/mcpDashboardOverviewLogic.ts
@@ -63,13 +63,16 @@ ORDER BY bucket
// Distinct MCP users for the "Users" tile — how many distinct people made tool calls.
// Counted over the doubled window like the KPI query, then split into the selected period
// and its equal-length predecessor with a single conditional uniq so the comparison is a
-// true distinct-person count (summing per-bucket distinct users would over-count anyone
+// true distinct-user count (summing per-bucket distinct users would over-count anyone
// active on more than one day). `__CUR_START__` is the selected-period boundary, injected
// as a timezone-aware toDateTime at call time.
+// Users are counted by distinct_id, not person_id. HogQL resolves person_id through a join
+// over every person override of the project, and on large projects that join runs out of
+// memory. A person with several distinct_ids counts once per distinct_id.
const USERS_QUERY = `
SELECT
- uniqIf(person_id, timestamp >= __CUR_START__) AS current_users,
- uniqIf(person_id, timestamp < __CUR_START__) AS prior_users
+ uniqIf(distinct_id, timestamp >= __CUR_START__) AS current_users,
+ uniqIf(distinct_id, timestamp < __CUR_START__) AS prior_users
FROM events
WHERE event = '$mcp_tool_call'
AND properties.$mcp_tool_name IS NOT NULL