fix(warehouse): surface actionable error on postgres incremental field type drift

[posthog]

Problem

A Postgres warehouse source configured for incremental sync on a column whose type
was altered after setup (for example a timestamp column later changed to text)
gets permanently, silently stuck. The stored incremental_field_type no longer
matches the live column, so _build_query emits WHERE "col" > '<timestamp>' and
Postgres rejects it with operator does not exist: text > timestamp without time zone.
Because that message contains does not exist, it's classified non-retryable and the
sync gives up — the customer only sees the raw Postgres error with no guidance on how
to fix it, and the sync stays halted until someone reconfigures the incremental field.

Changes

• Re-validate at sync setup. Before building any query, compare the persisted
  incremental field type against the live column type (_validate_incremental_field_type_matches_column).
  On drift, fail fast with an actionable IncrementalFieldTypeMismatchError telling the
  user to reconfigure the incremental field. Compatible cross-type-same-category cases
  Postgres handles fine (e.g. a date column with a timestamp cursor) are not flagged;
  xmin (a synthetic cursor) is excluded.
• Map the reactive case too. Add an operator does not exist entry to
  PostgresSource.get_non_retryable_errors, ordered before the generic does not exist
  key so its actionable message wins, as a fallback for any comparison that still reaches
  Postgres.

How did you test this code?

Automated tests only (added to test_postgres.py); no manual sync run was performed.

• TestValidateIncrementalFieldTypeMatchesColumn — parameterized over compatible
  types (that must not raise, including date↔timestamp and smallint/bigint↔integer)
  and incompatible drift (text/varchar vs a temporal/numeric cursor, cross-category
  swaps), plus the missing-column and xmin skip paths. Catches a regression where drift
  detection either fails to fire on a text-vs-timestamp column or false-positives on a
  still-compatible column.
• TestPostgresSourceNonRetryableErrors additions — assert the mismatch (both raw and
  Temporal-wrapped forms) is non-retryable, and that the operator does not exist key is
  ordered ahead of does not exist so production's friendly_errors[0] surfaces the
  actionable message rather than None.

Ran the two classes locally: 115 passed.

Automatic notifications

• Publish to changelog?
• Alert Sales and Marketing teams?

Docs update

🤖 Agent context

Autonomy: Fully autonomous

Authored by Claude Code (Opus 4.8) from a PostHog inbox report. Invoked the
/writing-tests skill before adding tests to apply the value gate (each test group
names the regression it catches). Considered two approaches from the report — a mapped
user-facing message for the operator does not exist case, and proactive re-validation
at setup — and implemented both: the early validation prevents the mismatched query from
ever reaching Postgres, and the mapped message is a defense-in-depth fallback. Category
grouping (temporal/numeric) was chosen over exact-type equality so legitimate implicit
casts Postgres already accepts (date vs timestamp) don't get falsely rejected.

---

Created with PostHog Code from an inbox report.

[scheduled-actions-posthog]

This PR hasn't seen activity in a week! Should it be merged, closed, or further worked on? If you want to keep it open, please remove the stale label – otherwise this will be closed in another week. If you want to permanently keep it open, use the waiting label.

[scheduled-actions-posthog]

This PR was closed due to lack of activity. Feel free to reopen if it's still relevant.