Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
10.5, 10.6, 10.7(EOL)
Description
A data corruption bug that affects ROW_FORMAT=COMPRESSED tables was found by RQG testing:
mysqld: /data/Server/bb-10.6-markoO/storage/innobase/row/row0purge.cc:208: bool row_purge_remove_clust_if_poss_low(purge_node_t*, ulint): Assertion `rec_get_deleted_flag(rec, rec_offs_comp(offsets))' failed.
|
The cause turned out to be corruption that had been introduced in page_zip_dir_insert() before the server had been killed and restarted. After the following fix had been applied, this assertion was not seen again:
diff --git a/storage/innobase/page/page0zip.cc b/storage/innobase/page/page0zip.cc
|
index f1da35d12ec..959256bf105 100644
|
--- a/storage/innobase/page/page0zip.cc
|
+++ b/storage/innobase/page/page0zip.cc
|
@@ -4248,8 +4248,13 @@ page_zip_dir_insert(
|
}
|
|
/* Write the entry for the inserted record.
|
- The "owned" and "deleted" flags must be zero. */
|
- mach_write_to_2(slot_rec - PAGE_ZIP_DIR_SLOT_SIZE, page_offset(rec));
|
+ The "owned" flag must be zero. */
|
+ uint16_t offs = page_offset(rec);
|
+ if (rec_get_deleted_flag(rec, true)) {
|
+ offs |= PAGE_ZIP_DIR_SLOT_DEL;
|
+ }
|
+
|
+ mach_write_to_2(slot_rec - PAGE_ZIP_DIR_SLOT_SIZE, offs);
|
mtr->zmemcpy(*cursor->block, slot_rec - page_zip->data
|
- PAGE_ZIP_DIR_SLOT_SIZE, PAGE_ZIP_DIR_SLOT_SIZE);
|
} |
It turns out that btr_cur_pessimistic_update() may invoke this function on a delete-marked record. The scenario involved ROLLBACK. I did not examine it deeper, but I suspect that a purgeable delete-marked record had been "resurrected" by an insert that was executed as delete-unmarking the record. Either the test deleted and later inserted the same record, or it updated a PRIMARY KEY value back and forth.
We failed to create a test case for this.
Attachments
Issue Links
- is caused by
-
MDEV-12353 Efficient InnoDB redo log record format
-
- Closed
-