π΄ Required Information
Describe the Bug:
migrate_from_sqlalchemy_sqlite.migrate skips any event row whose to_event() raises, logs a per-row warning, and then reports Migrated {len(events)} events. β the number of rows in the source table, not the number inserted. The transaction is committed and the run ends with Migration completed successfully., so a migration that permanently dropped events reports the full count. The pickle migration in the same package already counts successful inserts (num_rows in migrate_from_sqlalchemy_pickle.py).
Steps to Reproduce:
- Create a v0 SQLite database with one session and two event rows.
- Make one row unconvertible, e.g. by making
StorageEvent.to_event() raise for it.
- Run
python -m google.adk.sessions.migration.migrate_from_sqlalchemy_sqlite --source_db_url sqlite:///source.db --dest_db_path dest.db.
- Read the log.
Expected Behavior:
The summary counts the rows that were actually migrated, and the rows that were skipped are named, so the operator can tell the destination is incomplete before following the script's own "rename your database" instruction.
Observed Behavior:
WARNING ... Failed to migrate event event2: unreadable event payload
INFO ... Migrated 2 events.
INFO ... Migration completed successfully.
dest.db contains one event row (event1).
Environment Details:
- ADK Library Version: main @ 044a1ec
- Desktop OS: macOS
- Python Version: 3.14.5
Model Information:
- Are you using LiteLLM: No
- Which model is being used: N/A (offline migration script)
π‘ Optional Information
Regression:
The count has been the source row count since the script was added.
Logs:
INFO ... Migrating events...
WARNING ... Failed to migrate event event2: unreadable event payload
INFO ... Migrated 2 events.
INFO ... Migration completed successfully.
Additional Context:
I hit this while auditing the two migration scripts against each other. A regression test that builds a two-event source DB, fails one conversion, and asserts the log reports Migrated 1 events. fails on main today. I have a fix ready that counts successful inserts and adds one warning naming the skipped ids; happy to open the PR, or to keep it even smaller if you would rather only fix the count.
Minimal Reproduction Code:
The test in the PR (tests/unittests/sessions/migration/test_migration.py::test_sqlite_migration_reports_only_migrated_events) is self-contained: it builds the v0 database, patches to_event for one row, runs migrate(), and asserts both the destination contents and the log.
How often has this issue occurred?:
- Always (100%) whenever a row fails to convert
π΄ Required Information
Describe the Bug:
migrate_from_sqlalchemy_sqlite.migrateskips any event row whoseto_event()raises, logs a per-row warning, and then reportsMigrated {len(events)} events.β the number of rows in the source table, not the number inserted. The transaction is committed and the run ends withMigration completed successfully., so a migration that permanently dropped events reports the full count. The pickle migration in the same package already counts successful inserts (num_rowsinmigrate_from_sqlalchemy_pickle.py).Steps to Reproduce:
StorageEvent.to_event()raise for it.python -m google.adk.sessions.migration.migrate_from_sqlalchemy_sqlite --source_db_url sqlite:///source.db --dest_db_path dest.db.Expected Behavior:
The summary counts the rows that were actually migrated, and the rows that were skipped are named, so the operator can tell the destination is incomplete before following the script's own "rename your database" instruction.
Observed Behavior:
dest.dbcontains one event row (event1).Environment Details:
Model Information:
π‘ Optional Information
Regression:
The count has been the source row count since the script was added.
Logs:
Additional Context:
I hit this while auditing the two migration scripts against each other. A regression test that builds a two-event source DB, fails one conversion, and asserts the log reports
Migrated 1 events.fails onmaintoday. I have a fix ready that counts successful inserts and adds one warning naming the skipped ids; happy to open the PR, or to keep it even smaller if you would rather only fix the count.Minimal Reproduction Code:
The test in the PR (
tests/unittests/sessions/migration/test_migration.py::test_sqlite_migration_reports_only_migrated_events) is self-contained: it builds the v0 database, patchesto_eventfor one row, runsmigrate(), and asserts both the destination contents and the log.How often has this issue occurred?: