Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
10.2(EOL), 10.3(EOL), 10.4(EOL)
-
None
Description
mariabackup.mdev-14447 test case is in buildbot recently
Root cause is that mariabackup validation of first page fails with checksum mis-match error. (It can be due to partial write or zero filled page).
In xb_load_single_table_tablespace(), mariabackup allows the corrupted page 0
if (err != DB_SUCCESS && err != DB_CORRUPTION && xtrabackup_backup) {
|
/* allow corrupted first page for xtrabackup, it could be just
|
zero-filled page, which we restore from redo log later */
|
die("Failed to not validate first page of the file %s, error %d",name, (int)err);
|
}
|
Test case for it:
call mtr.add_suppression("InnoDB: New log files created");
|
|
|
let $basedir=$MYSQLTEST_VARDIR/tmp/backup;
|
let $incremental_dir=$MYSQLTEST_VARDIR/tmp/backup_inc1;
|
|
|
CREATE TABLE t(a varchar(40) PRIMARY KEY, b varchar(40), c varchar(40), d varchar(40), index(b,c,d)) ENGINE INNODB;
|
|
|
set global innodb_log_checkpoint_now = true;
|
|
|
let $backuplog=$MYSQLTEST_VARDIR/tmp/backup.log;
|
echo # Create full backup , modify table, then create incremental/differential backup;
|
--disable_result_log
|
|
|
--echo # dbug execute makes checksum mismatch while reading page 0 of table t.
|
|
|
exec $XTRABACKUP --defaults-file=$MYSQLTEST_VARDIR/my.cnf --backup --target-dir=$basedir --dbug=+d,page_intermittent_checksum_mismatch >$backuplog;
|
--enable_result_log
|
|
|
--disable_result_log
|
echo # Prepare full backup, apply incremental one;
|
exec $XTRABACKUP --prepare --verbose --apply-log-only --target-dir=$basedir;
|
|
|
echo # Restore and check results;
|
let $targetdir=$basedir;
|
-- source include/restart_and_restore.inc
|
|
|
|
|
--enable_result_log
|
SELECT count(*) FROM t;
|
DROP TABLE t;
|
|
|
# Cleanup
|
rmdir $basedir;
|
debug execute if code for validation fails:
|
|
diff --git a/storage/innobase/buf/buf0buf.cc b/storage/innobase/buf/buf0buf.cc
|
index 8dc2fd2..8f00c83 100644
|
--- a/storage/innobase/buf/buf0buf.cc
|
+++ b/storage/innobase/buf/buf0buf.cc
|
@@ -1103,6 +1103,17 @@ buf_page_is_corrupted(
|
|
if (srv_checksum_algorithm
|
== SRV_CHECKSUM_ALGORITHM_CRC32) {
|
+
|
+ DBUG_EXECUTE_IF(
|
+ "page_intermittent_checksum_mismatch", {
|
+ static int page_counter;
|
+ if (page_counter++ == 4) {
|
+ fprintf(stderr, "mismatch\n");
|
+ checksum_field2++;
|
+ }
|
+ });
|
+
|
+
|
crc32 = buf_page_check_crc32(read_buf,
|
Attachments
Issue Links
- relates to
-
-
- Closed
-
Activity
For a couple of lines like this, I'd prefer a loop over goto (similar to below, like it was already done for system tablespace elsewhere in this file) . I've no big concerns about performance, usually looping would be unnecessary.
for (int i= 0; i < 10; i++) {
|
err = file->validate_first_page(&flush_lsn);
|
if (err != DB_CORRUPTION) break;
|
my_sleep(1000);
|
}
|
|
|
|
diff --git a/extra/mariabackup/xtrabackup.cc b/extra/mariabackup/xtrabackup.ccindex 87f8822..1a45b1b 100644--- a/extra/mariabackup/xtrabackup.cc+++ b/extra/mariabackup/xtrabackup.cc@@ -3092,8 +3092,19 @@ xb_load_single_table_tablespace(die("Can't open datafile %s", name);}+ ulint retry_count = 10;+retry_read:err = file->validate_first_page(&flush_lsn);+ if (err == DB_CORRUPTION) {+ retry_count--;++ if (retry_count != 0) {+ os_thread_sleep(10000);+ goto retry_read;+ }+ }+if (err == DB_SUCCESS && file->space_id() != SRV_TMP_SPACE_ID) {os_offset_t node_size = os_file_get_size(file->handle());os_offset_t n_pages;@@ -3124,9 +3135,7 @@ xb_load_single_table_tablespace(delete file;- if (err != DB_SUCCESS && err != DB_CORRUPTION && xtrabackup_backup) {- /* allow corrupted first page for xtrabackup, it could be just- zero-filled page, which we restore from redo log later */+ if (err != DB_SUCCESS && xtrabackup_backup) {die("Failed to not validate first page of the file %s, error %d",name, (int)err);}}This could solve the problem. But whether the delay can bring down the performance