Hey team 👋 — we’re running into a consistent and cluster-wide

slot_id not found

issue on our StarRocks 3.4.2 setup, and need help confirming if this is a known bug or version-specific issue. Cluster setup: • StarRocks version: 3.4.2-c15ba7c • 3 nodes (each runs both FE + BE) • FEs: 10175 (Leader), 10002 & 10174 (Followers) • BEs: 10174, 10002, 10175 • Queries are routed via a load balancer (round-robin across all nodes). Error (100% reproducible):

ERROR 1064 (HY000): Expr evaluate meet error: slot_id <XXXX> not found: BE:10175

• Happens on every node — not limited to one backend. • Occurs on any query execution through the LB. • Persists after full cluster restart and after rebalancing replicas. What we’ve verified: • Cluster healthy: all FEs/BEs Alive:true (

SHOW PROC '/frontends'

,

/backends'

). • No schema or version mismatches. • Network between nodes is clean (no packet loss or RPC timeout). • Same behavior whether FE leader or follower handles query. 🙏 Appreciate any insights or fixes from the team. This issue is reproducible 100% of the time.

Did this happen to all of your tables, or on a single table? Also, did you run any

alter table

recently?

Hello Robert, thanks for your prompt response! Adding some context — it looks like the issue might be tied to metadata or statistics desynchronization after recent ALTER TABLE operations. The affected tables were altered (adding/dropping columns), and soon after that we started seeing both: •

Expr evaluate meet error: slot_id <id> not found

•

only found column statistics: {73: datakey, 102: count}, but missing statistic of col: 101: sum

This points toward a possible column descriptor or slot-mapping mismatch between FE metadata and BE tablet descriptors — especially since the

slot_id

and column stat IDs appear misaligned. We tried a few things already: 1. ANALYZE TABLE on the affected tables → didn’t resolve the slot/stat mismatch. 2. DROP STATS and re-ANALYZE → stats refreshed successfully but errors still persist. 3. Cluster otherwise healthy (no missing replicas, all Alive:true). It seems like BE metadata for altered columns may not have been fully synchronized or re-built post-alter, leading to missing slot descriptors or invalid stat references.

Can you please describe your table structure and the schema change (what you did to get to this point)? Would dropping and creating new column fix the issue?

Hello! After further investigation, here’s what we found: 1️⃣ “only found column statistics” issue • This was happening because we had a

DISTINCT COUNT

check on the primary key column in our query. • It appears that the statistics subsystem couldn’t locate all column stats when distinct aggregation was applied to the PK field. 2️⃣ “slot_id not found” issue • This occurs consistently in queries that include a

WHERE

clause with a

NULL

check combined with an

IN

condition. • For example:

(region IN ('Bay Area', 'SoCal', 'LA Area', 'Rocky Mountains', 'Midwest', 'NorCal') OR region IS NULL)

• The error surfaces during expression evaluation on the BE side, suggesting a possible mismatch or invalid slot assignment for the column involved in these logical checks. Let us know if this pattern aligns with any known issues around expression handling or predicate pushdown — we can provide table DDL and sample queries if that helps confirm the root cause

This seems like a bug. Can you please raise a github issue and provide a reproducible case?

Hello! I’ve submitted a GitHub issue for the

slot_id not found

and

column statistics missing

errors we’ve been encountering. Issue link -

<https://github.com/StarRocks/starrocks/issues/63840>

Please let us know of any additional details. Thank you!