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
erikgrinaker opened this issue
Jun 25, 2022
· 0 comments
· Fixed by #84603
Assignees
Labels
A-storageRelating to our storage engine (Pebble) on-disk storage.C-bugCode not up to spec/doc, specs & docs deemed correct. Solution expected to change code/behavior.T-storageStorage Team
As seen in cockroachdb/pebble#1786, it appears that table and block property collectors do not record Pebble range tombstones. Since these typically have bounds without timestamps they are ignored by the collectors, but conceptually they write at all timestamps. This can cause the TBI filtering to omit SSTs that contain these range tombstones, thus surfacing keys that should have been removed.
This isn't currently believed to be a correctness problem, because all TBI callers only use them as a performance optimization (to possibly skip irrelevant keys) and fall back to regular non-TBI iterators for processing. However, future callers may begin relying on them.
The text was updated successfully, but these errors were encountered:
erikgrinaker
added
C-bug
Code not up to spec/doc, specs & docs deemed correct. Solution expected to change code/behavior.
A-storage
Relating to our storage engine (Pebble) on-disk storage.
T-storage
Storage Team
labels
Jun 25, 2022
erikgrinaker
changed the title
storage: table/block property collectors do not handle range tombstones
storage: table/block property collectors ignore Pebble range tombstones
Jun 25, 2022
87916: ui: fix sql activity % of all runtime when filtered r=j82w a=j82w
Before this shows the % value was 2.5%, but after a filter applied it was changed to 100%.
https://loom.com/share/5b627ca2c0524653ae2ad6840a32e0a3
After this shows the % value stays the same after the filter was applied.
https://loom.com/share/338fde861349447f8725f991d01fd231closes#87415
Release justification: Category 2: Bug fixes and
low-risk updates to new functionality
Release note: (ui change): Fixed the % of all
runtime when filtered. It previously used only
the filtered data to calculate the value instead
of all the data in that time period.
88020: pkg/storage: unskip TestPebbleMVCCTimeIntervalWithRangeClears r=jbowens,erikgrinaker a=nicktrav
Unskips a test that was skipped due to #83376, which has since been fixed.
Release note: None.
Co-authored-by: j82w <[email protected]>
Co-authored-by: Nick Travers <[email protected]>
A-storageRelating to our storage engine (Pebble) on-disk storage.C-bugCode not up to spec/doc, specs & docs deemed correct. Solution expected to change code/behavior.T-storageStorage Team
As seen in cockroachdb/pebble#1786, it appears that table and block property collectors do not record Pebble range tombstones. Since these typically have bounds without timestamps they are ignored by the collectors, but conceptually they write at all timestamps. This can cause the TBI filtering to omit SSTs that contain these range tombstones, thus surfacing keys that should have been removed.
This isn't currently believed to be a correctness problem, because all TBI callers only use them as a performance optimization (to possibly skip irrelevant keys) and fall back to regular non-TBI iterators for processing. However, future callers may begin relying on them.
Jira issue: CRDB-17032
Epic CRDB-2624
The text was updated successfully, but these errors were encountered: