[MDEV-11556] InnoDB redo log apply fails to adjust data file sizes - Jira

Details

Type: Bug
Status: Closed (View Workflow)
Priority: Major
Resolution: Fixed
Affects Version/s: 10.1.20
Fix Version/s: 10.1.21, 10.2.4
Component/s: Backup, Storage Engine - InnoDB
Labels:
None

Sprint:
10.2.4-5

Description

Prepare of partial backup (with --tables option) sporadically fails with following error :

InnoDB: Error: tablespace size stored in header is 4864 pages, but
InnoDB: the sum of data file sizes is only 768 pages
InnoDB: Cannot start InnoDB. The tail of the system tablespace is
InnoDB: missing.

Attached are:
test_partial_uncommitted.sh - test which reliably demonstrates the problem in my environment
test_partial_uncommitted.log - corresponding log
my.cnf - config used
partial_uncommitted.test 'identical' mtr test, which DOES NOT demonstrate the problem (for unknown reasons (generated xtrabackup_info and backup-my.cnf are identical to .sh test).

Interesting observation is that all tables are backed up - this should be another bug eventually if not fixed as part of this.

Attachments

- Sort By Name
- Sort By Date
- Ascending
- Descending
- Thumbnails
- List
- Download All

my.cnf
0.3 kB
2016-12-13 10:52
partial_uncommitted.test
0.5 kB
2016-12-13 10:53
problem.tar.gz
8.89 MB
2016-12-13 14:25
test_partial_uncommitted.log
25 kB
2016-12-13 10:51
test_partial_uncommitted.sh
0.7 kB
2016-12-13 10:51

Issue Links

causes

MDEV-12027 posix_fallocate returned OS error 201.

Closed

MDEV-12075 innodb_use_fallocate does not work in MariaDB Server 10.1.21

Closed

MDEV-22501 Various issues when using --innodb-data-file-size-debug=-1 | SIGABRT in fil_mutex_enter_and_prepare_for_io

Closed

relates to

MDEV-11690 Remove UNIV_HOTBACKUP

Closed

MDEV-11915 Detect InnoDB system tablespace size mismatch early

Closed

MDEV-12022 InnoDB wrongly ignores the end of an .ibd file

Closed

MDEV-13822 Mariabackup --prepare --incremental sets file size incorrectly

Closed

MDEV-14082 Enforcing innodb_open_files leads to fil_system->mutex problem

Closed

MDEV-18194 Incremental prepare tries to access page which is out of tablespace bounds.

Closed

MDEV-30069 InnoDB: Trying to write ... bytes at ... outside the bounds of the file ...

Closed

MDEV-11968 With innodb_page_size=8K crash with 'Error 17' after "Tried to read 65536 bytes at offset 0. Was only able to read 49152."

Closed

MDEV-12893 innodb.log_data_file_size failed in buildbot with InnoDB: Database page corruption

Closed

MDEV-14447 mariabackup --incremental --prepare fails to extend last file when applying ibdata1.delta

Closed

MDEV-23190 Assertion `id.page_no() < space.size' during recovery

Closed

MDEV-26068 Server crashes when innodb-data-file-size-debug set to 1

Closed

(10 relates to)

Activity

Ascending order - Click to sort in descending order

View 11 older comments

Marko Mäkelä added a comment - 2016-12-29 12:10

The remaining problem was solved by removing some inappropriate rounding of the InnoDB system tablespace size:

diff --git a/storage/xtradb/srv/srv0start.cc b/storage/xtradb/srv/srv0start.cc

index 8b04b51ce0b..34fb1f87bdf 100644

--- a/storage/xtradb/srv/srv0start.cc

+++ b/storage/xtradb/srv/srv0start.cc

@@ -192,9 +192,6 @@ static const ulint SRV_UNDO_TABLESPACE_SIZE_IN_PAGES =

 #define SRV_N_PENDING_IOS_PER_THREAD	OS_AIO_N_PENDING_IOS_PER_THREAD

 #define SRV_MAX_N_PENDING_SYNC_IOS	100

-/** The round off to MB is similar as done in srv_parse_megabytes() */

-#define CALC_NUMBER_OF_PAGES(size)  ((size) / (1024 * 1024)) * \

-				  ((1024 * 1024) / (UNIV_PAGE_SIZE))

 #ifdef UNIV_PFS_THREAD

 /* Keys to register InnoDB threads with performance schema */

 UNIV_INTERN mysql_pfs_key_t	io_handler_thread_key;

@@ -1029,15 +1026,12 @@ open_or_create_data_files(

 			size = os_file_get_size(files[i]);

 			ut_a(size != (os_offset_t) -1);

-			/* Under some error conditions like disk full

-			narios or file size reaching filesystem

-			limit the data file could contain an incomplete

-			extent at the end. When we extend a data file

-			and if some failure happens, then also the data

-			file could contain an incomplete extent.  So we

-			need to round the size downward to a megabyte.*/

+			/* If InnoDB encountered an error or was killed

+			while extending the data file, the last page

+			could be incomplete. */

-			rounded_size_pages = (ulint) CALC_NUMBER_OF_PAGES(size);

+			rounded_size_pages = static_cast<ulint>(

+				size >> UNIV_PAGE_SIZE_SHIFT);

 			if (i == srv_n_data_files - 1

 			    && srv_auto_extend_last_data_file) {

Marko Mäkelä added a comment - 2016-12-29 12:10 The remaining problem was solved by removing some inappropriate rounding of the InnoDB system tablespace size: diff --git a/storage/xtradb/srv/srv0start.cc b/storage/xtradb/srv/srv0start.cc index 8b04b51ce0b..34fb1f87bdf 100644 --- a/storage/xtradb/srv/srv0start.cc +++ b/storage/xtradb/srv/srv0start.cc @@ -192,9 +192,6 @@ static const ulint SRV_UNDO_TABLESPACE_SIZE_IN_PAGES = #define SRV_N_PENDING_IOS_PER_THREAD OS_AIO_N_PENDING_IOS_PER_THREAD #define SRV_MAX_N_PENDING_SYNC_IOS 100 -/** The round off to MB is similar as done in srv_parse_megabytes() */ -#define CALC_NUMBER_OF_PAGES(size) ((size) / (1024 * 1024)) * \ - ((1024 * 1024) / (UNIV_PAGE_SIZE)) #ifdef UNIV_PFS_THREAD /* Keys to register InnoDB threads with performance schema */ UNIV_INTERN mysql_pfs_key_t io_handler_thread_key; @@ -1029,15 +1026,12 @@ open_or_create_data_files( size = os_file_get_size(files[i]); ut_a(size != (os_offset_t) -1); - /* Under some error conditions like disk full - narios or file size reaching filesystem - limit the data file could contain an incomplete - extent at the end. When we extend a data file - and if some failure happens, then also the data - file could contain an incomplete extent. So we - need to round the size downward to a megabyte.*/ + /* If InnoDB encountered an error or was killed + while extending the data file, the last page + could be incomplete. */ - rounded_size_pages = (ulint) CALC_NUMBER_OF_PAGES(size); + rounded_size_pages = static_cast<ulint>( + size >> UNIV_PAGE_SIZE_SHIFT); if (i == srv_n_data_files - 1 && srv_auto_extend_last_data_file) {

Jan Lindström (Inactive) added a comment - 2016-12-29 14:37

ok to push.

Jan Lindström (Inactive) added a comment - 2016-12-29 14:37 ok to push.

Marko Mäkelä added a comment - 2016-12-29 16:00

I repushed once more to bb-10.1-mdev-11556. There was a bug in my test file that caused it to fail when run against a non-debug server.

Marko Mäkelä added a comment - 2016-12-29 16:00 I repushed once more to bb-10.1-mdev-11556. There was a bug in my test file that caused it to fail when run against a non-debug server.

Marko Mäkelä added a comment - 2016-12-30 08:07

I pushed this to 10.1 and am now working on the merge to 10.2.

Marko Mäkelä added a comment - 2016-12-30 08:07 I pushed this to 10.1 and am now working on the merge to 10.2.

Marko Mäkelä added a comment - 2019-01-24 16:55

I did not merge this correctly to MariaDB 10.2. A follow-up adjustment will fix ~~MDEV-18194~~.

Marko Mäkelä added a comment - 2019-01-24 16:55 I did not merge this correctly to MariaDB 10.2. A follow-up adjustment will fix MDEV-18194 .

People

Assignee:: Marko Mäkelä

Reporter:: Andrii Nikitin (Inactive)

Votes:: 0 Vote for this issue

Watchers:: 5 Start watching this issue

Dates

Created:: 2016-12-13 10:51

Updated:: 2022-11-22 09:00

Resolved:: 2016-12-30 11:49

Git Integration

Error rendering 'com.xiplink.jira.git.jira_git_plugin:git-issue-webpanel'. Please contact your Jira administrators.