ISAR/libmdbx
3
0
Fork 0
mirror of https://github.com/isar/libmdbx.git synced 2025-01-10 04:34:13 +08:00
Code Issues Packages Projects Releases Wiki Activity

mdbx-windows: backport - always susppend local threads while resize DB (workaround for Windows kernel bug).

Browse Source 
We should not concern about performance on Windows platform,
it just unreasonable. Therefore just always suspend the local
threads to avoid this issue.

This resolves https://github.com/leo-yuriev/libmdbx/issues/48

Change-Id: I6e652692794b8c4c0d41625be62f2051b63c033a
This commit is contained in:
Leo Yuriev 2018-11-22 14:19:49 +03:00 committed by Leonid Yuriev
parent 850fe8408e
commit f183cef7d7
1 changed files with 15 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
src/mdbx.c
View File

@ -1983,11 +1983,14 @@ static int mdbx_mapresize(MDBX_env *env, const pgno_t size_pgno,
env->me_dxb_mmap.current == env->me_dxb_mmap.filesize)
goto bailout;
if ((env->me_flags & MDBX_RDONLY) || limit_bytes != env->me_dxb_mmap.length ||
size_bytes < env->me_dxb_mmap.current) {
/* Windows allows only extending a read-write section, but not a
/* 1) Windows allows only extending a read-write section, but not a
* corresponing mapped view. Therefore in other cases we must suspend
* the local threads for safe remap. */
* the local threads for safe remap.
* 2) At least on Windows 10 1803 the entire mapped section is unavailable
* for short time during NtExtendSection() or VirtualAlloc() execution.
*
* THEREFORE LOCAL THREADS SUSPENDING IS ALWAYS REQUIRED!
*/
array_onstack.limit = ARRAY_LENGTH(array_onstack.handles);
array_onstack.count = 0;
suspended = &array_onstack;
@ -1996,7 +1999,6 @@ static int mdbx_mapresize(MDBX_env *env, const pgno_t size_pgno,
mdbx_error("failed suspend-for-remap: errcode %d", rc);
goto bailout;
}
}
#else
/* Acquire guard to avoid collision between read and write txns
* around env->me_dbgeo */