Both Windows and MMAP capable implementations fell back to a
non-MEM_LARGE_PAGES/HugeTLB allocation with the large page implementaion
failed. These can can be freed by the corresponding function.
Prior to this, if we fell back to a conventional memory, than will
results in deallocation using munmap/VirtualFree on a memory allocated
using my_malloc_lock. At worst this could succeed and the
my_malloc_lock looses its memory without knowing about it.
Changed wording in error messages from MySQL to MariaDB. In
cases where the word server could be used instead it was done.
Tests that have these errors recorded were updated.
In a multi-batch recovery, we must ensure that INIT_PAGE and
especially the MDEV-15528 FREE_PAGE records will be taken
properly into account.
Writing a FREE_PAGE record gives the server permission to omit
a page write. If recovery insists on applying log to a page
whose page flush has been omitted, then the consistency checks
in the application of high-level redo log records (appending
an undo log record, inserting or deleting an index record)
will likely fail.
mlog_init_t::add(): Return whether the state was changed.
mlog_init_t::will_avoid_read(): Determine whether a page read
will be avoided and whether older log records can be safely
skipped.
recv_sys_t::parse(): Even if store==STORE_NO, process the records
INIT_PAGE and FREE_PAGE. While processing them, we can delete older
redo log records for the page. If store!=STORE_NO, we can directly
skip redo log recods of other types if mlog_init indicates that the
page will be freed or initialized by at a later LSN.
This fix was developed in cooperation with
Thirunarayanan Balathandayuthapani.
There is a 4 MB hard limit on Travis-CI and build output needs to be less
than that. Silencing the 'make install' step gets rid of a lot of
"Installing.." and "Missing.." and removing all mysql-test files will
make the dh_missing warnings much shorter.
ADD default history partitions generates wrong partition name,
f.ex. p2 instead of p1. Gap in sequence of partition names leads to
ha_partition::open_read_partitions() fail on inexistent name.
Manual fixing such broken table requires:
1. create empty table by any name (t_empty) with correct number
of partitions;
2. stop the server;
3. rename data files (.myd, .myi or .ibd) of broken table to t_empty
fixing the partition sequence (#p2 to #p1, #p3 to #p2);
4. start the server;
5. drop the broken table;
6. rename t_empty to correct table name.
The package libcurl4-dev is a virtual package, and no package with that
name actually exists. Depend on libcurl4-openssl-dev as the primary
option in case no libcurl4-dev providing package has been installed, and
if some of them (libcurl4-gnutls-dev, libcurl4-nss-dev,
libcurl4-openssl-dev ) are installed, they also satisfy the dependency
via the virtual package.
The Travis-CI still runs on libcurl3-dev as Travis-CI also runs Ubuntu
Xenial which does not yet have libcurl4 available.
This complements commit 2e0a40bdf0.
The package libdbd-mariadb-perl has been available in Debian September
2018, so it is already included since Debian Buster and Ubuntu Disco.
Use it as the primary Recommends for MariaDB client binaries, but keep
the MySQL variant as a fall-back in case this package gets backported to
older releases that did not yet have it.
This change was done in downstream Debian packaging in
bb4ad78bc1
This fixes Lintian errors and warnings:
E: mariadb-client-10.5: manpage-not-compressed usr/share/man/man1/mariadb-analyze.1
E: mariadb-client-10.5: manpage-not-compressed usr/share/man/man1/mariadb-optimize.1
E: mariadb-client-10.5: manpage-not-compressed usr/share/man/man1/mariadb-repair.1
E: mariadb-client-10.5: manpage-not-compressed usr/share/man/man1/mysqlanalyze.1
E: mariadb-client-10.5: manpage-not-compressed usr/share/man/man1/mysqloptimize.1
E: mariadb-client-10.5: manpage-not-compressed usr/share/man/man1/mysqlrepair.1
W: mariadb-client-10.5: binary-without-manpage usr/bin/mariadb-conv
W: libmariadb-dev: binary-without-manpage usr/bin/mariadb_config
If the binary name is mariadb_config, the manpage needs to be mariadb_config.1.
Also run 'wrap-and-sort -a -v' to get lists in correct order.
In addition, sort .gitignore and remove duplicate lines.
The dpkg-buildpackage has by default sensible values for --jobs and
--try-jobs, and it also inherits whatever 'parallel' is set in
DEB_BUILD_OPTIONS. Overriding it here should not be needed and removing
it is better, since it allows build systems to control the level or
parallel builds via DEB_BUILD_OPTIONS if they want, which they often do
(e.g. to avoid CPU congestion or overheating laptops).
Make package have sensible control file relations to mysql-8.0.
Also some cleanup:
- Remove obsolete versioned control relationships. When the package
name already has a version older than current package, the extra
version is obsolete.
- Remove obsolete postinst script that was empty
- Remove priority "extra" which is deprecated, use "optional" that
applies for the whole source package instead.
The current debian/control file and other packaging of 10.5 targets the
latest Debian unstable. In autobake-deb.sh there are automatic build
adaptations for older Debian/Ubuntu releases.
Some adaptations were for Debian versions prior to Jessie and Ubuntu
versions prior to Trusty. Those distro releases are no longer supported
anyway, so these should be cleaned away now in the 10.5 release cycle.
Summary:
- Debian Jessie has libcrack2 release 2.9.2-1, no need to have adaptations
for any older environments
- Debian Jessie has libpcre3-dev release 2:8.35-3.3, no need to consider
older versions
- Systemd has been available since Debian Jessie, no need to consider
that a Debian/Ubuntu environment would not have it available
adaptations for environments for anything older
Turn read cache off for periodic update.
Like 498a96a4 says:
Aria with row_format=fixed uses IO_CACHE of type READ_CACHE for
sequential read in update loop. When history row is inserted inside
this loop the cache misses it and fails with error.
This applicable to any additional row inserts on UPDATE. In this case
it was initiated by UPDATE FOR PORTION.
Related to MDEV-20441.
- Fixed mysql_prepare_create_table() constraint duplicate checking;
- Refactored period constraint handling in mysql_prepare_alter_table():
* No need to allocate new objects;
* Keep old constraint name but exclude it from dup checking by automatic_name;
- Some minor memory leaks fixed;
- Some conceptual TODOs.
This bug was introduced by MDEV-15528
commit a35b4ae898.
In the case that I analyzed, we failed to apply
an EXTENDED,INSERT_REUSE_REDUNDANT redo log record whose
preceding record points to unallocated area after PAGE_HEAP_TOP.
Had we properly written the FREE_PAGE record for the page,
recovery would have processed it, because during the checkpoint,
the log had been completely written past the LSN of the missed write.
fseg_free_page_low(): Always invoke mtr_t::free().
The other call is in fsp_free_page().
TDC_RT_REMOVE_ALL -> tdc_remove_table(). Some occurrences replaced with
TDC_element::flush() (whenver TABLE_SHARE is available).
TDC_RT_REMOVE_NOT_OWN[_KEEP_SHARE] -> TDC_element::flush(). These modes
assume that current thread owns TABLE_SHARE reference, which means we can
avoid hash lookup and flush unused TABLE instances directly.
TDC_RT_REMOVE_UNUSED -> TDC_element::flush_unused(). Only [ab]used by
mysql_admin_table() currently. Should be removed eventually.
Part of MDEV-17882 - Cleanup refresh version
