Bug was that ReplSemiSyncMaster::commitTrx() was waiting on a condition
for state to change, but didn't take into account that one could have
disabled semi-sync during the wait.
Problem is that FLUSH TABLES WITH READ LOCK first blocks threads from
starting new commits, then waits for running commits to complete. But
in-order parallel replication needs commits to happen in a particular
order, so this can easily deadlock.
To fix this problem, this patch introduces a way to temporarily pause
the parallel replication worker threads. Before starting FTWRL, we let
all worker threads complete in-progress transactions, and then
wait. Then we proceed to take the global read lock. Once the lock is
obtained, we unpause the worker threads. Now commits are blocked from
starting by the global read lock, so the deadlock will no longer occur.
Before, the Seconds_behind_master was updated already when an event
was queued for a worker thread to execute later. This might lead users
to interpret a low value as the slave being almost up to date with the
master, while in reality there might still be lots and lots of events
still queued up waiting to be applied by the slave.
See https://lists.launchpad.net/maria-developers/msg08958.html for
more detailed discussions.
Patch backported from MariaDB 10.1
- Ensure that we wait with cleanup() until slave thread has stopped.
- Added signal_thd_deleted() to signal close_connections() that all THD's has been freed.
Other things
- Removed not needed calls to THD_CHECK_SENTRY() when we are calling 'delete thd'.
Patch from Daniel Black:
- Change the charset of mysql.column_stats.{min_value, max_value} from
utf8_bin varchar to varbinary
- Adjust the code that saves/reads the data accordingly.
- Also provide upgrade statement in mysql_system_tables_fix.sql
pre/CMakeLists.txt defines CMAKE_DEBUG_POSTFIX which causes a different
library name on Windows debug build (pcred.lib rather than pre.lib)
However MERGE_LIBRARIES macro that is used to create static embedded
library (out of other static libraries), can not handle per-configuration
library names. Thus the build fails with "pre.lib not found"
Fix is to remove unnecessary CMAKE_DEBUG_POSTFIX
Fix build failures caused by new C runtime library
- isnan, snprintf, struct timespec are now defined, attempt to
redefine them leads
- P_tmpdir, tzname are no more defined
- lfind() and lsearch() in lf_hash.c had to be renamed, declaration
conflicts with some C runtime functions with the same name declared in
a header included by stdlib.h
Also fix couple of annoying warnings :
- remove #define NOMINMAX from config.h to avoid "redefined" compiler
warnings(NOMINMAX is already in compile flags)
- disable incremental linker in Debug as well (feature not used much
and compiler crashes often)
Also simplify package building with Wix, require Wix 3.9 or later
(VS2015 is not compatible with old Wix 3.5/3.6)
When compiled with "-Wl,-Bsymbolic-functions" flags
(e.g. when building a .deb package on Ubuntu) with TokuDB and jemalloc,
mysqld crashed in toku_get_processor_frequency_cpuinfo() when
free()-ing a buffer returned by getline().
getline() uses libc malloc() internally, while free() is aliased
to jemalloc's free() in this configuration.
Fixing not to use getline(). Using a static buffer instead.
Make unique table check after setup_fields of update because unique table can materialize table and we do not need field resolving after materialization.
The regression is caused by change bind-address server parameter
in MDEV-8083, so now server listens on IPv4 only by default.
The problem however is that on Windows, connection to server on localhost
appears to be much faster, if server listens on IPv6/dual stack.
mysql_real_connect() would try to connect to IPv6 loopback first,
and if this fails, the failing connect() call takes several seconds.
To fix, use bind-address=* on Windows, and 127.0.0.1 elsewhere
find_item_in_list() now recognize view fields like a fields even if they rever to an expression.
The problem of schema name do not taken into account for field with it and
derived table fixed.
Duplicating code removed
A few tests assumes that the CYCLE timer is always available,
which is not true on some platforms (e.g. ARM).
Fixing the tests not to reply on the CYCLE availability.
ha_partition::init_record_priority_queue()
Cherry-pick rev.6b0ee0c795499cee7f9deb649fb010801e0be4c2 from mysql-5.6.
Bug #18305270 BACKPORT BUG#18694052 FIX
FOR ASSERTION `!M_ORDERED_REC_BUFFER'
FAILED TO 5.6
PROBLEM
-------
Missed to remove record priority queue if
init_index failed for a partition which
was causing the crash.
FIX
---
Remove priority queue if init_index fails
for partition.
Analysis: Current implementation will write and read at least one block
(sort_buffer_size bytes) from disk / index even if that block does not
contain any records.
Fix: Avoid writing / reading empty blocks to temporary files (disk).
Since MariaDB packages have absolute paths, they are marked as not relocatable
by setting CPACK_RPM_PACKAGE_RELOCATABLE. According to logics of recent CPackRPM
it is not enough: one needs to set CPACK_PACKAGE_RELOCATABLE additionally.
mysql_secure_installation used incorrect path while looking up for "mysql"
client tool: $basedir/$basedir/bin instead of $basedir/bin.
This patch adapts "my_print_defaults" lookup algorithm for "mysql" client tool.
Assertion `inited==INDEX' failed in int handler::ha_index_first(uchar*)
The crash was because errors from init_read_record_idx() was not taken care of.
