4e4d8002e5 extended locks to have 64-bit
keys in most cases, but some were left as 32-bit. However, code using
these 32-bit locks assumed that the significant bytes of the key's long
value started from lck_string[0], which is false on big-endian
architectures. This commit adds Lock::getKeyString(), which gets a
pointer to the first used byte of lck_string, and should be used in
place of accessing lck_string directly.
2) Re-enabled checkouts in PIO routines, this resolves CORE-4179. Windows build may be broken, please validate.
3) Implemented faster reaction on asynchronous cancellation/shutdown requests.
4) Avoided locking dbb_sync in AST. This change also solves races during massive attachment shutdown.
fix for bug CORE-4444 : Engine could hung and block all attachments in out of disk space condition during physical backup,
improvement CORE-4445 : Extend main database file faster when physical backup state changed from stalled to merge
1. Frontported fix for CORE-3935 and CORE-3993.
2. Added debugging support for mutexes and rwlocks in Vulcan style.
Unfortunately after last 'svn up' build asserts in JrdStatement:71.
2) Reworked the lock manager synchronization: added the RAII mutex holders and avoided dirty reads for the shared memory.
3) Encapsulated lck_length into the Lock class + small related refactoring.
4) Front ported recent fixes from the v2.x branches.
5) Misc cleanup.
Also some cleanups, the most important are:
- meaningful ctor on Jrd::Lock, helping to avoid code dup
- avoid unneeded h-file dependencies, making boot build engine dependent
1) Fixed CORE-1696 - "Deadlock in LM by using nbackup utility" with priority rating of lock requesting for LCK_backup_database and LCK_bdb.
2) A lot of stability fixes were made.
3) The key thing is using new GlobalRWLock for removing LCK_set_owner_handle function and speed up performance.
4) The code is still ready for using both in shared cache architecture and in claster architecture.
2. Fixed writing page while nbackup merge state is stalled.
3. Fixed using dead dbb in AST handler while somebody is changing cached lock of the GlobalRWLock.
