supabase / etl / 29511544140

74%

Build Type

push

github

Committed by web-flow

Commit Message

fix(bigquery): Retry stale 404s when querying tables in tests (#894)

* fix(bigquery): Retry stale 404s when querying tables in tests

A table queried right after being dropped and recreated can return a
stale NOT_FOUND while BigQuery metadata propagates. Retry 404s within
the existing query budget instead of returning None on the first one.
Tables that stay missing for the whole budget still yield None.

* fix(bigquery): Split test table queries from absence checks

query_table gets a 60 second budget with retry logging, since a view
recreated under the same name can serve stale NOT_FOUND well past the
previous budget. Absence assertions move to a new table_exists helper
backed by the tables metadata API, which answers existence consistently
and without polling.

* fix(bigquery): Assert row deletion with a no-rows wait, not existence

The six CDC tests assert that delete events emptied the table, which
the table_exists check got wrong since the table legitimately survives.
Poll for an empty query result with the short budget instead; a 404
counts as no rows visible, matching the prior behavior of these
assertions.

* fix(bigquery): Drop the unused table_exists test helper

It was introduced for absence checks that later moved to
wait_for_no_rows, leaving it without callers. Point query_table's
docs at wait_for_no_rows instead.

* fix(bigquery): Require an observed empty result in wait_for_no_rows

A single 404 previously counted as no rows visible, but the same
stale metadata that makes query_table retry NOT_FOUND can hide a
table that still has rows, so a stale 404 could pass the delete
assertion without ever confirming emptiness. Retry 404s like
non-empty results and return true only after an actual empty
response.

Coverage Stats

16 of 29 new or added lines in 1 file covered. (55.17%)

213 existing lines in 7 files now uncovered.

42774 of 58063 relevant lines covered (73.67%)

1473.24 hits per line