fix(sessions): stop DatabaseSessionService.update_time onupdate from masking explicit writes

[chelsealong]

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):

• Fixes: DatabaseSessionService raises false StaleSessionError when two events share a timestamp #7276

Problem:

DatabaseSessionService.append_event raises StaleSessionError when there is
only one writer, whenever an event shares the same microsecond timestamp as
the previous event and carries a state change.

append_event always sets StorageSession.update_time explicitly from the
event's timestamp (database_session_service.py), and the column is also
declared with onupdate=func.now() (schemas/v1.py). When the new timestamp
equals the value already stored, SQLAlchemy sees no change to update_time
and drops it from the UPDATE statement's explicit values. The UPDATE still
runs because another column (state) changed, so onupdate fires and the
database's own clock is written instead — whole seconds on SQLite, the
transaction start time on Postgres. The in-memory revision marker, read
before commit, no longer matches what storage actually holds, so the next
append_event on that session is incorrectly rejected as stale.

Equal timestamps are common with coarser clocks (e.g. Python 3.12 on Windows,
15.6ms GetSystemTimeAsFileTime ticks), so a workflow emitting several events
per step hits this regularly.

Solution:

Drop onupdate=func.now() from StorageSession.update_time
(src/google/adk/sessions/schemas/v1.py). create_session already sets a
default for inserts, and append_event is the only code path that updates
this column on an existing row — and it always sets it explicitly. No other
column relies on this onupdate (verified StorageAppState/StorageUserState
are untouched, since they still rely on their own onupdate and are never set
explicitly).

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

Added test_append_event_same_timestamp_single_writer_not_stale to
tests/unittests/sessions/test_session_service.py, reproducing the issue's
repro script against the DATABASE (SQLite) backend: two events with an
identical timestamp, the second carrying a state delta, followed by a third
event from the same session object.

$ git checkout HEAD~1 -- src/google/adk/sessions/schemas/v1.py
$ pytest tests/unittests/sessions/test_session_service.py -k test_append_event_same_timestamp_single_writer_not_stale
FAILED ... StaleSessionError: The session has been modified in storage since it was loaded.

$ git checkout HEAD -- src/google/adk/sessions/schemas/v1.py
$ pytest tests/unittests/sessions/test_session_service.py -k test_append_event_same_timestamp_single_writer_not_stale
1 passed

$ pytest tests/unittests/sessions -q
502 passed, 2 xfailed

$ pytest tests/unittests -n auto -q
16177 passed, 82 skipped, 26 xfailed, 2 xpassed

pre-commit run --files src/google/adk/sessions/schemas/v1.py tests/unittests/sessions/test_session_service.py passes (isort, pyink, ruff,
addlicense, compliance checks, codespell). The one hook failure it reports
(check-new-py-prefix on src/google/adk/memory/_sqlite_memory_service.py)
is pre-existing on main/unmodified HEAD and unrelated to this change.

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have added tests that prove my fix is effective.
• New and existing unit tests pass locally with my changes.

AI assistance disclosure

This change (analysis, fix, and test) was developed with the assistance of
Claude Code (Anthropic).