[Nikhi00718]

What this PR does / why we need it:

feast materialize on the DuckDB offline store crashes on DuckDB 2.0 when a FileSource holds zone-aware nanosecond timestamps, which is what pandas writes by default for a datetime64[ns, UTC] column:

duckdb.ConversionException: Conversion Error: Unimplemented type for cast (TIMESTAMPTZ_NS -> TIMESTAMP) when casting from source column event_timestamp

The duckdb extra is ibis-framework[duckdb]>=10.0.0, and ibis 12 only requires duckdb>=0.10.3, so a fresh install today gets DuckDB 2.0.0, released 2026-10-21. Feast's CI pins 1.5.6, which is why CI doesn't see this. get_historical_features keeps working, so the failure first shows up at materialization.

pull_latest_from_table_or_query_ibis and pull_all_from_table_or_query_ibis filter on table[timestamp_field].cast("timestamp"). DuckDB 2 won't cast TIMESTAMPTZ_NS to a zone-less TIMESTAMP. ibis 12 also reports that column's type as unknown, so the code can't read its zone.

The filter now goes through _comparable_timestamp:

• A zone-aware column is cast to a zone-aware timestamp at the default precision.
• A column ibis reports as unknown is cast to a UTC timestamp.
• Everything else is still cast to timestamp, as before, including strings and zone-less columns.

Which issue(s) this PR fixes:

None filed. I found no existing issue or PR for it. ibis-project/ibis#12206, "support duckdb 2.0", is open and may teach ibis this type; the unknown-type branch is harmless if it lands.

Testing

Materialize with a DuckDB-store FeatureStore, on a host at UTC+05:30, 24 hourly rows, a window that should select 4 of them, reading back through the SQLite online store:

Parquet timestamp	DuckDB 2.0.0, before	DuckDB 2.0.0, after	DuckDB 1.5.6, after
timestamp[ns, tz=UTC] (pandas default)	ConversionException	correct 4 rows	correct 4 rows
timestamp[us, tz=UTC]	correct 4 rows	correct 4 rows	correct 4 rows
timestamp[us]	correct 4 rows	correct 4 rows	correct 4 rows

New test in test_ibis.py, across both pull functions and 4 timestamp types:

• DuckDB 2.0.0: 2 failed, 8 passed before the fix, with both failures on datetime64[ns, UTC]; 10 passed after.
• DuckDB 1.5.6: 10 passed before and after, so this test won't catch a regression until CI moves to DuckDB 2.

ruff check and ruff format pass on both changed files.

On DuckDB 2.0.0 with this change, a point-in-time join against a brute-force reference also matched on all 130 entity rows, with and without a TTL, on generated data with same-timestamp rows.

Checks

• I've made sure the tests are passing. (Only test_ibis.py, from a scratch copy; see Limitations.)
• My commits are signed off (git commit -s)
• My PR title follows conventional commits format

Testing Strategy

• Unit tests
• Integration tests
• Manual tests
• Testing is not required for this change

Limitations

• I ran test_ibis.py from a scratch copy, because the repository's conftest.py needs the GCP extras, which I didn't install. I did not run the rest of the unit suite or the integration suite.
• Windows, Python 3.12, ibis 12.0.0. Not tested on other ibis backends. The zone-aware cast is the form point_in_time_join already uses.
• I didn't add an upper bound on duckdb or change the CI pin. Moving CI to DuckDB 2 is a separate decision for you.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 66.66667% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 51.33%. Comparing base (bbaca76) to head (2f46278).

Files with missing lines	Patch %	Lines
sdk/python/feast/infra/offline_stores/ibis.py	66.66%	1 Missing and 1 partial ⚠️
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #7097      +/-   ##
==========================================
+ Coverage   51.32%   51.33%   +0.01%     
==========================================
  Files         458      458              
  Lines       57441    57447       +6     
  Branches     8421     8422       +1     
==========================================
+ Hits        29482    29493      +11     
+ Misses      25950    25939      -11     
- Partials     2009     2015       +6     

Flag	Coverage Δ
go-feature-server	30.58% <ø> (ø)
python-unit	52.81% <66.66%> (+0.01%)	⬆️

Files with missing lines	Coverage Δ
sdk/python/feast/infra/offline_stores/ibis.py	68.75% <66.66%> (+3.61%)	⬆️

... and 1 file with indirect coverage changes

---

Continue to review full report in Codecov by Harness.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update bbaca76...2f46278. Read the comment docs.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.