mirror of
https://github.com/rustfs/rustfs.git
synced 2026-08-04 04:17:44 +00:00
houseme
6d0583872b
Three tests share the global list-objects mutation-sequence state (the LIST_OBJECTS_MUTATION_SEQUENCE atomic and the per-bucket sequence map) and all call reset_list_objects_mutation_sequences_for_test(), but only observed_mutation_persists_namespace_journal_high_water was marked #[serial_test::serial]. The serial lock only excludes tests that also carry the attribute, so the unmarked tests still ran concurrently with it: the journal test's persist of "bucket" -> 11 could land after the recovery test's reset, making it observe 11 instead of the expected 9 (flaky under the default multi-threaded test runner, green with --test-threads=1). Add #[serial_test::serial] to the two unmarked tests: - persistent_key_only_index_load_recovers_mutation_sequence_from_journal - list_objects_mutation_sequence_advances_per_bucket Unique bucket names alone would not fix this: the per-bucket assertions in list_objects_mutation_sequence_advances_per_bucket check exact values produced by the single global atomic counter, which any concurrent observe call perturbs, and every reset zeroes that global state for whichever peer is mid-flight. Co-authored-by: heihutu <heihutu@gmail.com>