supabase / etl / 42613095517

81%

Build Type

push

github

Committed by web-flow

Commit Message

test: crash-replay property for destinations (#1221)

* test: crash-replay property for destinations

Run the random CDC workload through a real pipeline into ClickHouse
(both engines), DuckLake, BigQuery, and Snowflake, and inject one
fault per case: a crash and restart, a lost write acknowledgement,
or a rejected write, on either the first copy write or a streaming
write. Afterwards each table in the destination must match the
source, with no lost rows and no duplicate keys.

- Move the workload generator to etl::test_utils::cdc_workload so
  destination tests can use it.
- WorkloadShape keeps each destination to changes it supports.
- ClickHouse turns on insert block dedup after the initial copy, as
  Replicated and Shared MergeTree do.
- DuckLake uses the production pool size.
- ClickHouse and DuckLake run on every PR. BigQuery runs in a new
  nightly workflow and Snowflake in the existing one.

* test: return slot read errors as TestCaseError

wait_for_sync_slot_removal returns a TestCaseError on timeout but
panicked on a failed slot read. Propagate the read error the same
way so proptest reports it as a case failure.

* ci: limit BigQuery nightly to integration tests

The nightly filter matched the lib binary too, so --run-ignored only
also ran three known-defect BigQuery unit tests that fail by design.
Restrict it to the etl-destinations::main integration binary.

* test: document ClickHouse crash-replay row fields

* test: treat missing tables as crash-replay failures

BigQuery and DuckLake read a missing destination table as zero rows,
so a case whose source table ended empty passed even when the
destination lost or never recreated the table. Table sync writes every
table at least once, so a missing table is now an error. BigQuery
retries it within the settle window to ride out stale NOT_FOUND.

* test: order crash-replay helpers before callers

Move-only change to follow the define-helpers-before-use rule.

Coverage Stats

103 of 129 new or added lines in 3 files covered. (79.84%)

21 existing lines in 2 files now uncovered.

64432 of 79105 relevant lines covered (81.45%)

32892.01 hits per line