Merge pull request #710 from orozery/disk_space_reservation

add disk space reservation
This commit is contained in:
Takeshi Nakatani 2018-03-04 13:27:25 +09:00 committed by GitHub
commit 45c7ea9194
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 102 additions and 56 deletions

View File

@ -183,7 +183,7 @@ number of one part size in multipart uploading request.
The default size is 10MB(10485760byte), minimum value is 5MB(5242880byte). The default size is 10MB(10485760byte), minimum value is 5MB(5242880byte).
Specify number of MB and over 5(MB). Specify number of MB and over 5(MB).
.TP .TP
\fB\-o\fR ensure_diskfree(default the same as multipart_size value) \fB\-o\fR ensure_diskfree(default 0)
sets MB to ensure disk free space. This option means the threshold of free space size on disk which is used for the cache file by s3fs. sets MB to ensure disk free space. This option means the threshold of free space size on disk which is used for the cache file by s3fs.
s3fs makes file for downloading, and uploading and caching files. s3fs makes file for downloading, and uploading and caching files.
If the disk free space is smaller than this value, s3fs do not use diskspace as possible in exchange for the performance. If the disk free space is smaller than this value, s3fs do not use diskspace as possible in exchange for the performance.

View File

@ -1429,7 +1429,7 @@ int FdEntity::NoCacheCompleteMultipartPost(void)
int FdEntity::RowFlush(const char* tpath, bool force_sync) int FdEntity::RowFlush(const char* tpath, bool force_sync)
{ {
int result; int result = 0;
S3FS_PRN_INFO3("[tpath=%s][path=%s][fd=%d]", SAFESTRPTR(tpath), path.c_str(), fd); S3FS_PRN_INFO3("[tpath=%s][path=%s][fd=%d]", SAFESTRPTR(tpath), path.c_str(), fd);
@ -1448,10 +1448,12 @@ int FdEntity::RowFlush(const char* tpath, bool force_sync)
if(0 < restsize){ if(0 < restsize){
if(0 == upload_id.length()){ if(0 == upload_id.length()){
// check disk space // check disk space
if(FdManager::IsSafeDiskSpace(NULL, restsize)){ if(ReserveDiskSpace(restsize)){
// enough disk space // enough disk space
// Load all uninitialized area // Load all uninitialized area
if(0 != (result = Load())){ result = Load();
FdManager::get()->FreeReservedDiskSpace(restsize);
if(0 != result){
S3FS_PRN_ERR("failed to upload all area(errno=%d)", result); S3FS_PRN_ERR("failed to upload all area(errno=%d)", result);
return static_cast<ssize_t>(result); return static_cast<ssize_t>(result);
} }
@ -1554,6 +1556,32 @@ int FdEntity::RowFlush(const char* tpath, bool force_sync)
return result; return result;
} }
// [NOTICE]
// Need to lock before calling this method.
bool FdEntity::ReserveDiskSpace(size_t size)
{
if(FdManager::get()->ReserveDiskSpace(size)){
return true;
}
if(!is_modify){
// try to clear all cache for this fd.
pagelist.Init(pagelist.Size(), false);
if(-1 == ftruncate(fd, 0) || -1 == ftruncate(fd, pagelist.Size())){
S3FS_PRN_ERR("failed to truncate temporary file(%d).", fd);
return false;
}
if(FdManager::get()->ReserveDiskSpace(size)){
return true;
}
}
FdManager::get()->CleanupCacheDir();
return FdManager::get()->ReserveDiskSpace(size);
}
ssize_t FdEntity::Read(char* bytes, off_t start, size_t size, bool force_load) ssize_t FdEntity::Read(char* bytes, off_t start, size_t size, bool force_load)
{ {
S3FS_PRN_DBG("[path=%s][fd=%d][offset=%jd][size=%zu]", path.c_str(), fd, (intmax_t)start, size); S3FS_PRN_DBG("[path=%s][fd=%d][offset=%jd][size=%zu]", path.c_str(), fd, (intmax_t)start, size);
@ -1561,10 +1589,6 @@ ssize_t FdEntity::Read(char* bytes, off_t start, size_t size, bool force_load)
if(-1 == fd){ if(-1 == fd){
return -EBADF; return -EBADF;
} }
// check if not enough disk space left BEFORE locking fd
if(FdManager::IsCacheDir() && !FdManager::IsSafeDiskSpace(NULL, size)){
FdManager::get()->CleanupCacheDir();
}
AutoLock auto_lock(&fdent_lock); AutoLock auto_lock(&fdent_lock);
if(force_load){ if(force_load){
@ -1575,23 +1599,6 @@ ssize_t FdEntity::Read(char* bytes, off_t start, size_t size, bool force_load)
// check disk space // check disk space
if(0 < pagelist.GetTotalUnloadedPageSize(start, size)){ if(0 < pagelist.GetTotalUnloadedPageSize(start, size)){
if(!FdManager::IsSafeDiskSpace(NULL, size)){
// [NOTE]
// If the area of this entity fd used can be released, try to do it.
// But If file data is updated, we can not even release of fd.
// Fundamentally, this method will fail as long as the disk capacity
// is not ensured.
//
if(!is_modify){
// try to clear all cache for this fd.
pagelist.Init(pagelist.Size(), false);
if(-1 == ftruncate(fd, 0) || -1 == ftruncate(fd, pagelist.Size())){
S3FS_PRN_ERR("failed to truncate temporary file(%d).", fd);
return -ENOSPC;
}
}
}
// load size(for prefetch) // load size(for prefetch)
size_t load_size = size; size_t load_size = size;
if(static_cast<size_t>(start + size) < pagelist.Size()){ if(static_cast<size_t>(start + size) < pagelist.Size()){
@ -1603,9 +1610,25 @@ ssize_t FdEntity::Read(char* bytes, off_t start, size_t size, bool force_load)
load_size = static_cast<size_t>(pagelist.Size() - start); load_size = static_cast<size_t>(pagelist.Size() - start);
} }
} }
if(!ReserveDiskSpace(load_size)){
S3FS_PRN_WARN("could not reserve disk space for pre-fetch download");
load_size = size;
if(!ReserveDiskSpace(load_size)){
S3FS_PRN_ERR("could not reserve disk space for pre-fetch download");
return -ENOSPC;
}
}
// Loading // Loading
int result; int result = 0;
if(0 < size && 0 != (result = Load(start, load_size))){ if(0 < size){
result = Load(start, load_size);
}
FdManager::get()->FreeReservedDiskSpace(load_size);
if(0 != result){
S3FS_PRN_ERR("could not download. start(%jd), size(%zu), errno(%d)", (intmax_t)start, size, result); S3FS_PRN_ERR("could not download. start(%jd), size(%zu), errno(%d)", (intmax_t)start, size, result);
return -EIO; return -EIO;
} }
@ -1642,17 +1665,21 @@ ssize_t FdEntity::Write(const char* bytes, off_t start, size_t size)
pagelist.SetPageLoadedStatus(static_cast<off_t>(pagelist.Size()), static_cast<size_t>(start) - pagelist.Size(), false); pagelist.SetPageLoadedStatus(static_cast<off_t>(pagelist.Size()), static_cast<size_t>(start) - pagelist.Size(), false);
} }
int result; int result = 0;
ssize_t wsize; ssize_t wsize;
if(0 == upload_id.length()){ if(0 == upload_id.length()){
// check disk space // check disk space
size_t restsize = pagelist.GetTotalUnloadedPageSize(0, start) + size; size_t restsize = pagelist.GetTotalUnloadedPageSize(0, start) + size;
if(FdManager::IsSafeDiskSpace(NULL, restsize)){ if(ReserveDiskSpace(restsize)){
// enough disk space // enough disk space
// Load uninitialized area which starts from 0 to (start + size) before writing. // Load uninitialized area which starts from 0 to (start + size) before writing.
if(0 < start && 0 != (result = Load(0, static_cast<size_t>(start)))){ if(0 < start){
result = Load(0, static_cast<size_t>(start));
}
FdManager::get()->FreeReservedDiskSpace(restsize);
if(0 != result){
S3FS_PRN_ERR("failed to load uninitialized area before writing(errno=%d)", result); S3FS_PRN_ERR("failed to load uninitialized area before writing(errno=%d)", result);
return static_cast<ssize_t>(result); return static_cast<ssize_t>(result);
} }
@ -1750,6 +1777,7 @@ void FdEntity::CleanupCache()
FdManager FdManager::singleton; FdManager FdManager::singleton;
pthread_mutex_t FdManager::fd_manager_lock; pthread_mutex_t FdManager::fd_manager_lock;
pthread_mutex_t FdManager::cache_cleanup_lock; pthread_mutex_t FdManager::cache_cleanup_lock;
pthread_mutex_t FdManager::reserved_diskspace_lock;
bool FdManager::is_lock_init(false); bool FdManager::is_lock_init(false);
string FdManager::cache_dir(""); string FdManager::cache_dir("");
bool FdManager::check_cache_dir_exist(false); bool FdManager::check_cache_dir_exist(false);
@ -1901,19 +1929,7 @@ bool FdManager::CheckCacheDirExist(void)
size_t FdManager::SetEnsureFreeDiskSpace(size_t size) size_t FdManager::SetEnsureFreeDiskSpace(size_t size)
{ {
size_t old = FdManager::free_disk_space; size_t old = FdManager::free_disk_space;
if(0 == size){
if(0 == FdManager::free_disk_space){
FdManager::free_disk_space = static_cast<size_t>(S3fsCurl::GetMultipartSize() * S3fsCurl::GetMaxParallelCount());
}
}else{
if(0 == FdManager::free_disk_space){
FdManager::free_disk_space = max(size, static_cast<size_t>(S3fsCurl::GetMultipartSize() * S3fsCurl::GetMaxParallelCount()));
}else{
if(static_cast<size_t>(S3fsCurl::GetMultipartSize() * S3fsCurl::GetMaxParallelCount()) <= size){
FdManager::free_disk_space = size; FdManager::free_disk_space = size;
}
}
}
return old; return old;
} }
@ -1957,6 +1973,7 @@ FdManager::FdManager()
try{ try{
pthread_mutex_init(&FdManager::fd_manager_lock, NULL); pthread_mutex_init(&FdManager::fd_manager_lock, NULL);
pthread_mutex_init(&FdManager::cache_cleanup_lock, NULL); pthread_mutex_init(&FdManager::cache_cleanup_lock, NULL);
pthread_mutex_init(&FdManager::reserved_diskspace_lock, NULL);
FdManager::is_lock_init = true; FdManager::is_lock_init = true;
}catch(exception& e){ }catch(exception& e){
FdManager::is_lock_init = false; FdManager::is_lock_init = false;
@ -1980,6 +1997,7 @@ FdManager::~FdManager()
try{ try{
pthread_mutex_destroy(&FdManager::fd_manager_lock); pthread_mutex_destroy(&FdManager::fd_manager_lock);
pthread_mutex_destroy(&FdManager::cache_cleanup_lock); pthread_mutex_destroy(&FdManager::cache_cleanup_lock);
pthread_mutex_destroy(&FdManager::reserved_diskspace_lock);
}catch(exception& e){ }catch(exception& e){
S3FS_PRN_CRIT("failed to init mutex"); S3FS_PRN_CRIT("failed to init mutex");
} }
@ -2181,17 +2199,22 @@ bool FdManager::ChangeEntityToTempPath(FdEntity* ent, const char* path)
void FdManager::CleanupCacheDir() void FdManager::CleanupCacheDir()
{ {
S3FS_PRN_INFO("cache cleanup requested");
if(!FdManager::IsCacheDir()){ if(!FdManager::IsCacheDir()){
return; return;
} }
AutoLock auto_lock(&FdManager::cache_cleanup_lock, true); AutoLock auto_lock_no_wait(&FdManager::cache_cleanup_lock, true);
if (!auto_lock.isLockAcquired()) {
return;
}
if(auto_lock_no_wait.isLockAcquired()){
S3FS_PRN_INFO("cache cleanup started");
CleanupCacheDirInternal(""); CleanupCacheDirInternal("");
S3FS_PRN_INFO("cache cleanup ended");
}else{
// wait for other thread to finish cache cleanup
AutoLock auto_lock(&FdManager::cache_cleanup_lock);
}
} }
void FdManager::CleanupCacheDirInternal(const std::string &path) void FdManager::CleanupCacheDirInternal(const std::string &path)
@ -2224,16 +2247,38 @@ void FdManager::CleanupCacheDirInternal(const std::string &path)
}else{ }else{
FdEntity* ent; FdEntity* ent;
if(NULL == (ent = FdManager::get()->Open(next_path.c_str(), NULL, -1, -1, false, true, true))){ if(NULL == (ent = FdManager::get()->Open(next_path.c_str(), NULL, -1, -1, false, true, true))){
S3FS_PRN_DBG("skipping locked file: %s", next_path.c_str());
continue; continue;
} }
if(ent->IsMultiOpened()){
S3FS_PRN_DBG("skipping opened file: %s", next_path.c_str());
}else{
ent->CleanupCache(); ent->CleanupCache();
S3FS_PRN_DBG("cleaned up: %s", next_path.c_str());
}
Close(ent); Close(ent);
} }
} }
closedir(dp); closedir(dp);
} }
bool FdManager::ReserveDiskSpace(size_t size)
{
AutoLock auto_lock(&FdManager::reserved_diskspace_lock);
if(IsSafeDiskSpace(NULL, size)){
free_disk_space += size;
return true;
}
return false;
}
void FdManager::FreeReservedDiskSpace(size_t size)
{
AutoLock auto_lock(&FdManager::reserved_diskspace_lock);
free_disk_space -= size;
}
/* /*
* Local variables: * Local variables:
* tab-width: 4 * tab-width: 4

View File

@ -144,6 +144,7 @@ class FdEntity
void Close(void); void Close(void);
bool IsOpen(void) const { return (-1 != fd); } bool IsOpen(void) const { return (-1 != fd); }
bool IsMultiOpened(void) const { return refcnt > 1; }
int Open(headers_t* pmeta = NULL, ssize_t size = -1, time_t time = -1, bool no_fd_lock_wait = false); int Open(headers_t* pmeta = NULL, ssize_t size = -1, time_t time = -1, bool no_fd_lock_wait = false);
bool OpenAndLoadAll(headers_t* pmeta = NULL, size_t* size = NULL, bool force_load = false); bool OpenAndLoadAll(headers_t* pmeta = NULL, size_t* size = NULL, bool force_load = false);
int Dup(bool no_fd_lock_wait = false); int Dup(bool no_fd_lock_wait = false);
@ -173,6 +174,7 @@ class FdEntity
ssize_t Read(char* bytes, off_t start, size_t size, bool force_load = false); ssize_t Read(char* bytes, off_t start, size_t size, bool force_load = false);
ssize_t Write(const char* bytes, off_t start, size_t size); ssize_t Write(const char* bytes, off_t start, size_t size);
bool ReserveDiskSpace(size_t size);
void CleanupCache(); void CleanupCache();
}; };
typedef std::map<std::string, class FdEntity*> fdent_map_t; // key=path, value=FdEntity* typedef std::map<std::string, class FdEntity*> fdent_map_t; // key=path, value=FdEntity*
@ -186,6 +188,7 @@ class FdManager
static FdManager singleton; static FdManager singleton;
static pthread_mutex_t fd_manager_lock; static pthread_mutex_t fd_manager_lock;
static pthread_mutex_t cache_cleanup_lock; static pthread_mutex_t cache_cleanup_lock;
static pthread_mutex_t reserved_diskspace_lock;
static bool is_lock_init; static bool is_lock_init;
static std::string cache_dir; static std::string cache_dir;
static bool check_cache_dir_exist; static bool check_cache_dir_exist;
@ -217,8 +220,9 @@ class FdManager
static size_t GetEnsureFreeDiskSpace(void) { return FdManager::free_disk_space; } static size_t GetEnsureFreeDiskSpace(void) { return FdManager::free_disk_space; }
static size_t SetEnsureFreeDiskSpace(size_t size); static size_t SetEnsureFreeDiskSpace(size_t size);
static size_t InitEnsureFreeDiskSpace(void) { return SetEnsureFreeDiskSpace(0); }
static bool IsSafeDiskSpace(const char* path, size_t size); static bool IsSafeDiskSpace(const char* path, size_t size);
static void FreeReservedDiskSpace(size_t size);
bool ReserveDiskSpace(size_t size);
FdEntity* GetFdEntity(const char* path, int existfd = -1); FdEntity* GetFdEntity(const char* path, int existfd = -1);
FdEntity* Open(const char* path, headers_t* pmeta = NULL, ssize_t size = -1, time_t time = -1, bool force_tmpfile = false, bool is_create = true, bool no_fd_lock_wait = false); FdEntity* Open(const char* path, headers_t* pmeta = NULL, ssize_t size = -1, time_t time = -1, bool force_tmpfile = false, bool is_create = true, bool no_fd_lock_wait = false);

View File

@ -4671,8 +4671,6 @@ static int my_fuse_opt_proc(void* data, const char* arg, int key, struct fuse_ar
S3FS_PRN_EXIT("multipart_size option must be at least 5 MB."); S3FS_PRN_EXIT("multipart_size option must be at least 5 MB.");
return -1; return -1;
} }
// update ensure free disk space if it is not set.
FdManager::InitEnsureFreeDiskSpace();
return 0; return 0;
} }
if(0 == STR2NCMP(arg, "ensure_diskfree=")){ if(0 == STR2NCMP(arg, "ensure_diskfree=")){
@ -5034,8 +5032,7 @@ int main(int argc, char* argv[])
} }
// check free disk space // check free disk space
FdManager::InitEnsureFreeDiskSpace(); if(!FdManager::IsSafeDiskSpace(NULL, S3fsCurl::GetMultipartSize() * S3fsCurl::GetMaxParallelCount())){
if(!FdManager::IsSafeDiskSpace(NULL, S3fsCurl::GetMultipartSize())){
S3FS_PRN_EXIT("There is no enough disk space for used as cache(or temporary) directory by s3fs."); S3FS_PRN_EXIT("There is no enough disk space for used as cache(or temporary) directory by s3fs.");
exit(EXIT_FAILURE); exit(EXIT_FAILURE);
} }

View File

@ -1125,7 +1125,7 @@ void show_help (void)
" multipart_size (default=\"10\")\n" " multipart_size (default=\"10\")\n"
" - part size, in MB, for each multipart request.\n" " - part size, in MB, for each multipart request.\n"
"\n" "\n"
" ensure_diskfree (default same multipart_size value)\n" " ensure_diskfree (default 0)\n"
" - sets MB to ensure disk free space. s3fs makes file for\n" " - sets MB to ensure disk free space. s3fs makes file for\n"
" downloading, uploading and caching files. If the disk free\n" " downloading, uploading and caching files. If the disk free\n"
" space is smaller than this value, s3fs do not use diskspace\n" " space is smaller than this value, s3fs do not use diskspace\n"