ABFSS Select From glob not recognising * wildcards

[daryl-lynch-bzy]

What Happens?

The * wildcards that I use in a glob ( 'ABFSS/*/*.json) search have stopped working....

Since Tue 3rd Feb 2026, I have run into the following error:

I am running duckdb 1.4.4 within a Microsoft Fabric Python (3.11) Notebook. I am trying to read the files in delta using the pattern described by @djouallah here Attach_Lakehouse_v2.ipynb

I process has successfully worked for several month and started using 1.4.4 between 27-Jan to 2-Feb. Here is the code that is failing:

## The following code is derived from Mim Djouallah (https://github.com/djouallah/Fabric_Notebooks_Demo/blob/main/Attach_LH/Attach_Lakehouse_v2.ipynb)

def duckdb_attach_lakehouse(token, wks, lh, path):
    con = duckdb.connect(f'temp_{time.time_ns()}.duckdb')
    con.sql('SET enable_object_cache=true')
    con.sql(f""" CREATE or replace SECRET onelake ( TYPE AZURE, PROVIDER ACCESS_TOKEN, ACCESS_TOKEN '{token}')   """)
    
    sql_schema     = set()
    sql_statements = set()
    
    # modified to select only the ListOperatons Table
    con.sql(f""" SELECT  * FROM glob ("{path}/*") """).df()['file'].tolist()
    list_tables = con.sql(f""" SELECT  distinct(split_part(file, '_delta_log', 1)) as tables FROM glob ("{path}*/_delta_log/*.json") """).df()['tables'].tolist()
    for table_path in list_tables:
            parts = table_path.strip("/").split("/")
            schema = parts[-2]
            table = parts[-1]
            sql_schema.add(f"CREATE SCHEMA IF NOT EXISTS {schema};")
            sql_statements.add(f"""CREATE OR REPLACE view {schema}.{table} AS SELECT * 
                                FROM delta_scan('abfss://{wks}@onelake.dfs.fabric.microsoft.com/{lh}/Tables/{schema}/{table}');""")
    con.sql(" ".join(sql_schema))
    con.sql(" ".join(sql_statements))
    con.sql("SHOW ALL TABLES").show(max_width=150)
    con.sql('CHECKPOINT')
    return con

The list_tables step is now returning an empty list [] rather than the list of tables.

Fixes attempted

In the previous step in had the following syntax which also returns an empty list.

con.sql(f""" SELECT * FROM glob ("{path}/*") """).df()['file'].tolist()

when the wildcard is replaced with ** is works.

con.sql(f""" SELECT * FROM glob ("{path}**") """).df()['file'].tolist()

if I replace this path with a single table is works:

abfss://onelake.dfs.fabric.microsoft.com/{wks_id}/{wks_id}/{lkh_id}/Tables/Bronze/{Table Name}/_delta_log/*.json

So can't run the following:

abfss://onelake.dfs.fabric.microsoft.com/{wks_id}/{wks_id}/{lkh_id}/Tables/Bronze/*/_delta_log/*.json

Next Steps

I can refactor the code, but I was hoping I could understand why it might have suddenly stopped?

[djouallah]

@benfleis any glob behaviour change you made in azure maybe between 1.4.3 and 1.4.4 ?

[benfleis]

It's likely a side effect of several small+related changes I made to support writes, combined with the known brittleness of globbing in the abfss protocol. (See several older issues calling out the glob inconsitencies: https://github.com/duckdb/duckdb-delta/issues/ )

It would take a deeper dive to get to this one -- there is a set of related changes -- FileExists, a small fix to Glob behavior and other things, which all interplay in the multi-file scans (which include their own heuristics as well). A quick naive test of these globs (alone) worked in my setup, but there's probably an interplay here with the multi file scanner.