{{ message }}
Repository navigation
leaked hash_seq_search scan for hash table warning in join #15
Description
Describe the bug
If I run the join query from the readme twice in the same session I get a leaked hash_seq_search scan for hash table warning.
Environment
Ubuntu 22.04 on WSL with PostgreSQL 16.5
Reproduction Steps
SELECT cursorpage FROM documentdb_api.aggregate_cursor_first_page('documentdb', $$ {
"aggregate": "patient",
"pipeline": [
{
"$lookup": {
"from": "appointment",
"localField": "patient_id",
"foreignField": "patient_id",
"as": "appointment"
}
},
{
"$unwind":"$appointment"
},
{
"$project": {
"_id":0,
"name":1,
"appointment.doctor_name":1,
"appointment.appointment_date":1
}
}
],
"cursor": { "batchSize": 3 }
}
$$);Expected behavior
No leak warning
** Actual behavior **
NOTICE: There are open held portals. Closing them
WARNING: leaked hash_seq_search scan for hash table 0x55d1eb54fce8
Activity
diipak-bisht commentedon Jan 27, 2025
The issue is PG complains about any non-terminated hash_search at the txn commit time, and the non-terminating hash_search starts in
CreateAndDrainPersistedQueryfunction which tries to close all previously open portals withThereAreNoReadyPortals.The hash table is private and there is not much APIs to access it or freeze it before starting hash_search so that PG would not complain.
visridha commentedon Jan 28, 2025
We're actively looking at this. This should be resolved soon - and we're refactoring the code to not scan all cursors and prune them.