You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[WITHDRAWN] Add removed_fragments to the fragment-reuse index format
#7402
Withdrawn. This vote was opened on an incorrect premise — apologies for the noise.
The observable corruption in #7374's reproduction (a take ... fragment does not exist error on an IVF_PQ index after a deferred compaction) was a stale PQ-storage row-id bug affecting compacted fragments, and it was already fixed by #7315 ("refresh PQ storage row ids after fragment-reuse remap"). Bisecting confirms it: the commit before #7315 errors on a compacted fragment; #7315 makes the same reproduction return correct results.
The remaining "fully-deleted fragment" concern in the issue text is a real but latent invariant gap: the fragment-reuse index's remap_row_id does pass a deleted fragment's address through unchanged, but that is masked end-to-end by the query-time deletion mask (create_restricted_deletion_mask blocks fragments missing from the manifest), and on current code it could not be reproduced as observable corruption via any search path. A persisted format field isn't justified for a defect that is already masked everywhere observable.
Withdrawing pending a genuine, unmasked reproduction. If one turns up, this can be re-opened framed honestly as hardening that invariant.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Withdrawn. This vote was opened on an incorrect premise — apologies for the noise.
The observable corruption in #7374's reproduction (a
take ... fragment does not existerror on an IVF_PQ index after a deferred compaction) was a stale PQ-storage row-id bug affecting compacted fragments, and it was already fixed by #7315 ("refresh PQ storage row ids after fragment-reuse remap"). Bisecting confirms it: the commit before #7315 errors on a compacted fragment; #7315 makes the same reproduction return correct results.The remaining "fully-deleted fragment" concern in the issue text is a real but latent invariant gap: the fragment-reuse index's
remap_row_iddoes pass a deleted fragment's address through unchanged, but that is masked end-to-end by the query-time deletion mask (create_restricted_deletion_maskblocks fragments missing from the manifest), and on current code it could not be reproduced as observable corruption via any search path. A persisted format field isn't justified for a defect that is already masked everywhere observable.Withdrawing pending a genuine, unmasked reproduction. If one turns up, this can be re-opened framed honestly as hardening that invariant.
All reactions