Inconsistent datetime offsets in charts on master – anyone seeing this on 6.0.0?

[Alice-608]

On latest master, I'm seeing inconsistent datetime offsets in bar/scatter charts (eg: some points shift hours, others days), while raw SQL Lab queries return correct timestamps.

Before rolling back, has anyone experienced this on 6.0.0 with a similar setup (PostgreSQL/MySQL)? Just need to know if this is a master regression or also present in stable. Thanks!

[dosubot[bot]]

This appears to be a master-only regression. There's a known issue (#37925) tracking exactly what you're describing—non-uniform datetime shifts in Bar/Scatter charts while SQL Lab returns correct values.

The root cause is that chart queries apply datetime normalization via normalize_dttm_col() which SQL Lab bypasses. This normalization can apply up to three offset layers (datasource offset, query time shift, time comparison offsets) conditionally per row, causing the inconsistent shifts you're seeing.

Several related fixes were merged in Feb 2026 on master:

• PR #37979: Fixed datetime_to_epoch to use astimezone(pytz.utc) instead of replace(tzinfo=pytz.utc)
• PR #35587: Added utc=True to PyArrow datetime conversion

Temporary workaround: Try setting the dataset offset to 0 (no offset) if viable for your use case.

Since these fixes landed recently on master and the issue was reported against master, 6.0.0 stable should not have this particular regression. If you're on master, you may want to ensure you have the latest commits including the above fixes.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}