fix(ecstore): preserve ILM state during pool decommission (#6369)

* fix(ecstore): migrate ILM metadata during decommission

* fix(ecstore): verify ILM metadata before decommission

* fix(ecstore): track ILM recovery across decommission

* fix(ecstore): close ILM receipt recovery gaps

* fix(ecstore): anchor decommission ILM receipts

* fix(ecstore): re-export durable ILM checkpoint

* fix(ecstore): avoid terminal receipt shadowing

* fix(ecstore): harden durable ILM cursor receipts

* fix(ecstore): repair durable ILM receipt recovery

* test(ecstore): cover durable ILM recovery boundaries

* test(ecstore): serialize multi-source ILM recovery

* test(ecstore): compile multi-source ILM recovery

* test(ecstore): isolate durable ILM scenario stack

* fix(ecstore): preserve active ILM source journals

* fix(ecstore): distinguish active ILM target cleanup

* fix(ecstore): restore decommission test imports

* fix(ecstore): drop removed decommission test import

* fix(ecstore): remove duplicate decommission error helper

* fix(ecstore): fence final decommission sweep

* test(ecstore): cover final sweep cancel fence

* fix(ecstore): fence decommission cancellation

* fix(ecstore): remove redundant clone in test

* fix(ecstore): keep manual transition progress compatible

* fix(ecstore): restore decommission worker wrapper

* fix(ecstore): restore decommission compile contracts

* test(ecstore): adapt reload worker canceler
This commit is contained in:
Zhengchao An
2026-08-23 16:43:11 +08:00
committed by GitHub
parent 450ec7f66a
commit 34bbc1adb3
12 changed files with 4272 additions and 120 deletions
+10
View File
@@ -78,6 +78,12 @@ test-group = 'embedded-test-ports'
filter = 'package(rustfs-ecstore) & test(manual_transition_page_checkpoint_persists_durable_job_progress)'
test-group = 'ecstore-serial-flaky'
# The durable ILM decommission regressions build isolated multi-pool stores and
# deliberately take source or target disks offline while checking fencing.
[[profile.default.overrides]]
filter = 'package(rustfs-ecstore) & (test(decommission_migrates_and_verifies_registered_durable_ilm_records) | test(decommission_durable_ilm_target_read_error_is_not_masked_by_peer_success) | test(decommission_durable_ilm_terminal_receipt_recovers_failed_source_cleanup) | test(decommission_durable_ilm_receipt_pagination_fails_closed_on_second_page) | test(decommission_durable_ilm_recovery_keeps_multiple_active_sources))'
test-group = 'ecstore-serial-flaky'
# Serialize the bucket-incarnation / lifecycle-fence tests. They drive
# init_bucket_metadata_sys and bucket_metadata_sys_of, i.e. process-global
# OnceLock state that serial_test's #[serial] cannot protect across nextest's
@@ -190,6 +196,10 @@ test-group = 'embedded-test-ports'
filter = 'package(rustfs-ecstore) & test(manual_transition_page_checkpoint_persists_durable_job_progress)'
test-group = 'ecstore-serial-flaky'
[[profile.ci.overrides]]
filter = 'package(rustfs-ecstore) & (test(decommission_migrates_and_verifies_registered_durable_ilm_records) | test(decommission_durable_ilm_target_read_error_is_not_masked_by_peer_success) | test(decommission_durable_ilm_terminal_receipt_recovers_failed_source_cleanup) | test(decommission_durable_ilm_receipt_pagination_fails_closed_on_second_page) | test(decommission_durable_ilm_recovery_keeps_multiple_active_sources))'
test-group = 'ecstore-serial-flaky'
# Serialize the bucket-incarnation / lifecycle-fence tests under the ci profile
# too (see the matching default-profile override near the top). No retries.
[[profile.ci.overrides]]