Changes codes

1) Changes codes for performance and request's costs
    s3fs gets object's  attributes by using HEAD request.
    Directory objects is following 4 type:
      a) name type is "dir", with meta information
      b) name type is "dir", without meta information(but has files in itself)
      c) name type is "dir/", with(out) meta information
      d) name type is "dir_$folder$", with(out) meta information
    The codes is changed to order checking directory object.
    So that, s3fs decreases requests for checking objects.

    Before version has a bug, that is s3fs can not be recognizable type-b) 
    definitely when s3fs check the object directly.(but s3fs can, when s3fs 
    check the object by listing)
    This change fixes this bug.

2) Adds "multireq_max" option
    Added "multireq_max" option is maximum number of parallel request 
    for listing objects.
    This changes is possible to solve CURLE_COULDNT_CONNECT.
    If this option can not solve it, this option will be useful for tuning 
    performance by each.




git-svn-id: http://s3fs.googlecode.com/svn/trunk@449 df820570-a93a-0410-bd06-b72b767a4274
This commit is contained in:
ggtakec@gmail.com 2013-06-19 14:53:58 +00:00
parent b65c3e195e
commit 45950044f7
4 changed files with 128 additions and 57 deletions

View File

@ -92,6 +92,9 @@ You can specify this option for performance, s3fs memorizes in stat cache that t
\fB\-o\fR nodnscache - disable dns cache.
s3fs is always using dns cache, this option make dns cache disable.
.TP
\fB\-o\fR multireq_max (default="500")
maximum number of parallel request for listing objects.
.TP
\fB\-o\fR url (default="http://s3.amazonaws.com")
sets the url to use to access Amazon S3. If you want to use HTTPS, then you can set url=https://s3.amazonaws.com
.TP

View File

@ -54,7 +54,7 @@ using namespace std;
//-------------------------------------------------------------------
// Define
//-------------------------------------------------------------------
#define MAX_MULTI_HEADREQ 500 // max request count in readdir curl_multi.
#define MAX_MULTI_HEADREQ 500 // default: max request count in readdir curl_multi.
#define DIRTYPE_UNKNOWN -1
#define DIRTYPE_NEW 0
#define DIRTYPE_OLD 1
@ -112,6 +112,7 @@ static gid_t s3fs_gid = 0; // default = root.
static bool is_s3fs_umask = false;// default does not set.
static mode_t s3fs_umask = 0;
static bool dns_cache = true; // default = true
static int multireq_maxcnt = MAX_MULTI_HEADREQ;
// if .size()==0 then local file cache is disabled
static std::string use_cache;
@ -307,6 +308,12 @@ static int chk_dir_object_type(const char *path, string& newpath, string& nowpat
// Get object attributes with stat cache.
// This function is base for s3fs_getattr().
//
// [NOTICE]
// Checking order is changed following list because of reducing the number of the requests.
// 1) "dir"
// 2) "dir/"
// 3) "dir_$folder$"
//
static int get_object_attribute(const char *path, struct stat *pstbuf, headers_t* pmeta, bool overcheck, bool* pisforce)
{
int result = -1;
@ -321,6 +328,10 @@ static int get_object_attribute(const char *path, struct stat *pstbuf, headers_t
//FGPRINT(" get_object_attribute[path=%s]\n", path);
if(!path || '\0' == path[0]){
return -ENOENT;
}
memset(pstat, 0, sizeof(struct stat));
if(0 == strcmp(path, "/") || 0 == strcmp(path, ".")){
pstat->st_nlink = 1; // see fuse faq
@ -330,13 +341,10 @@ static int get_object_attribute(const char *path, struct stat *pstbuf, headers_t
// Check cache.
strpath = path;
if(overcheck){
Pos = strpath.find("_$folder$", 0);
if(string::npos != Pos){
if(overcheck && string::npos != (Pos = strpath.find("_$folder$", 0))){
strpath = strpath.substr(0, Pos);
strpath += "/";
}
}
if(pisforce){
(*pisforce) = false;
}
@ -348,63 +356,80 @@ static int get_object_attribute(const char *path, struct stat *pstbuf, headers_t
return -ENOENT;
}
// At first, check "object/".
// At first, check path
strpath = path;
if(overcheck && 0 < strpath.length() && '/' != strpath[strpath.length() - 1]){
s3_realpath = get_realpath(strpath.c_str());
result = curl_get_headers(s3_realpath.c_str(), (*pheader));
// overcheck
if(overcheck && 0 != result){
if('/' != strpath[strpath.length() - 1] && string::npos == strpath.find("_$folder$", 0)){
// path is "object", check "object/" for overcheck
strpath += "/";
s3_realpath = get_realpath(strpath.c_str());
if(0 != (result = curl_get_headers(s3_realpath.c_str(), (*pheader)))){
string strSp= path;
strSp += "_$folder$";
s3_realpath = get_realpath(strSp.c_str());
result = curl_get_headers(s3_realpath.c_str(), (*pheader));
}
}
if(0 != result){
// not found "object/", check "_$folder$"
strpath = path;
s3_realpath = get_realpath(strpath.c_str());
if(0 != (result = curl_get_headers(s3_realpath.c_str(), (*pheader)))){
// Not found --> check( if overcheck )
if(overcheck){
if(string::npos == strpath.find("_$folder$", 0)){
// path doesn't have "_$folder$" --> check for s3fox etc
if('/' == strpath[strpath.length() - 1]){
strpath = strpath.substr(0, strpath.length() - 1);
}
strpath += "_$folder$";
s3_realpath = get_realpath(strpath.c_str());
result = curl_get_headers(s3_realpath.c_str(), (*pheader));
}
if(0 == result){
// found "_$folder$" object.
strpath = path; // reset original
strpath += "/";
}else{
// path does not have "_$folder$" --> check "no dir obejct".
if(-ENOTEMPTY == directory_empty(path)){
}
if(0 != result){
// not found "object/" and "object_$folder$", check no dir object.
strpath = path;
if(string::npos == strpath.find("_$folder$", 0)){
if('/' == strpath[strpath.length() - 1]){
strpath = strpath.substr(0, strpath.length() - 1);
}
if(-ENOTEMPTY == directory_empty(strpath.c_str())){
// found "no dir obejct".
forcedir = true;
strpath = path; // reset original
strpath += "/";
forcedir = true;
if(pisforce){
(*pisforce) = true;
}
result = 0;
}
}
}
}else{
// Add no object cache.
// found "path" object.
if('/' != strpath[strpath.length() - 1]){
// check a case of that "object" does not have attribute and "object" is possible to be directory.
if(is_need_check_obj_detail(*pheader)){
if(-ENOTEMPTY == directory_empty(strpath.c_str())){
strpath += "/";
forcedir = true;
if(pisforce){
(*pisforce) = true;
}
result = 0;
}
}
}
}
if(0 != result){
// finally, "path" object did not find. Add no object cache.
strpath = path; // reset original
StatCache::getStatCacheData()->AddNoObjectCache(strpath);
return result;
}
}
}else{
return result;
}
}else{
// if path has "_$folder$", need to cut it.
Pos = strpath.find("_$folder$", 0);
if(string::npos != Pos){
if(string::npos != (Pos = strpath.find("_$folder$", 0))){
strpath = strpath.substr(0, Pos);
strpath += "/";
}
}
}
// Set into cache
if(0 != StatCache::getStatCacheData()->GetCacheSize()){
// add into stat cache
if(!StatCache::getStatCacheData()->AddStat(strpath, (*pheader), forcedir)){
@ -2963,7 +2988,7 @@ static int readdir_multi_head(const char *path, S3ObjList& head)
mh = curl_multi_init();
// Make single head request.
for(liter = headlist.begin(), cnt = 0; headlist.end() != liter && cnt < MAX_MULTI_HEADREQ; ){
for(liter = headlist.begin(), cnt = 0; headlist.end() != liter && cnt < multireq_maxcnt; ){
string fullpath = path + (*liter);
string fullorg = path + head.GetOrgName((*liter).c_str());
string etag = head.GetETag((*liter).c_str());
@ -2991,7 +3016,7 @@ static int readdir_multi_head(const char *path, S3ObjList& head)
return -EIO;
}
liter++;
cnt++; // max request count in multi-request is MAX_MULTI_HEADREQ.
cnt++; // max request count in multi-request is multireq_maxcnt.
}
// Send multi request.
@ -4208,6 +4233,10 @@ static int my_fuse_opt_proc(void *data, const char *arg, int key, struct fuse_ar
use_cache = strchr(arg, '=') + 1;
return 0;
}
if (strstr(arg, "multireq_max=") != 0) {
multireq_maxcnt = atoi(strchr(arg, '=') + 1);
return 0;
}
if(strstr(arg, "nonempty") != 0) {
nonempty = true;
// need to continue for fuse.

View File

@ -637,6 +637,41 @@ time_t get_lastmodified(headers_t& meta)
return get_lastmodified((*iter).second.c_str());
}
//
// Returns it whether it is an object with need checking in detail.
// If this function returns true, the object is possible to be directory
// and is needed checking detail(searching sub object).
//
bool is_need_check_obj_detail(headers_t& meta)
{
headers_t::const_iterator iter;
// directory object is Content-Length as 0.
if(0 != get_size(meta)){
return false;
}
// if the object has x-amz-meta information, checking is no more.
if(meta.end() != meta.find("x-amz-meta-mode") ||
meta.end() != meta.find("x-amz-meta-mtime") ||
meta.end() != meta.find("x-amz-meta-uid") ||
meta.end() != meta.find("x-amz-meta-gid") ||
meta.end() != meta.find("x-amz-meta-owner") ||
meta.end() != meta.find("x-amz-meta-group") ||
meta.end() != meta.find("x-amz-meta-permissions") )
{
return false;
}
// if there is not Content-Type, or Content-Type is "x-directory",
// checking is no more.
if(meta.end() == (iter = meta.find("Content-Type"))){
return false;
}
if("application/x-directory" == (*iter).second){
return false;
}
return true;
}
//-------------------------------------------------------------------
// Help
//-------------------------------------------------------------------
@ -711,6 +746,9 @@ void show_help (void)
" nodnscache - disable dns cache\n"
" - s3fs is always using dns cache, this option make dns cache disable.\n"
"\n"
" multireq_max (default=\"500\")\n"
" - maximum number of parallel request for listing objects.\n"
"\n"
" url (default=\"http://s3.amazonaws.com\")\n"
" - sets the url to use to access amazon s3\n"
"\n"

View File

@ -93,6 +93,7 @@ gid_t get_gid(headers_t& meta);
blkcnt_t get_blocks(off_t size);
time_t get_lastmodified(const char* s);
time_t get_lastmodified(headers_t& meta);
bool is_need_check_obj_detail(headers_t& meta);
void show_usage(void);
void show_help(void);