Race condition between client disconnect and worker assignment causes worker process crash on TLS connections
Details
| Detail name | Value |
|---|---|
| Changelog Number | 31828 |
| Type | Bug |
| Status | Resolved |
| Affected Versions | Exasol 7.0.0 |
| Fix Versions | Exasol 2025.1.17 |
| Resolution Date | 2026-11-12 |
Description
A race condition exists between a client disconnecting and the database dispatching that client's query to a worker process. Under high concurrency, if a client is forcefully disconnected (e.g. network drop, TCP reset, or abrupt application termination) at the exact moment a worker process is being assigned to handle the client's query, the worker process could crash with an internal error. The affected session receives an Internal server error, transaction was rolled back response. Other sessions and ongoing queries are not affected; the worker slot is automatically recycled.
The race is specific to TLS-encrypted connections and requires the disconnect to coincide within a millisecond-scale window during worker assignment.
Workaround
There is no workaround.
Fix
The race condition has been resolved. The connection server now correctly detects when a client connection becomes invalid during the worker assignment handoff and aborts the dispatch cleanly without involving the worker process. The worker slot is recycled and the session is terminated with a standard connection-lost error, with no crash occurring.