# Flush temp directory **URL:** \ **Category:** NGINX\ **Created:** [August 14, 2013, 1:20pm UTC](https://www.ruby-forum.com/t/flush-temp-directory/236203 "2013-08-14T13:20:24Z")\ **Posts on this page:** 4\ **Page:** 1
**Author:** ![alcina](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/a/a587f6/32.png) [@alcina](https://www.ruby-forum.com/u/alcina)\ **Post date:** [August 14, 2013, 1:20pm UTC](https://www.ruby-forum.com/t/flush-temp-directory/236203/1 "2013-08-14T13:20:24Z")
Hi All, I am trying to test accelerated upload on nginx/php-fpm/php-cgi setup and comparing different scenarios e.g one where /temp is a tmpfs, one where it is a disk partition and you will also notice where in i test using php-cgi. as i need to understand which can handle file uploads faster. ``` location ~ \.php$ { include fastcgi_params; client_body_temp_path /temp; fastcgi_pass_request_body off; client_body_in_file_only on; fastcgi_param REQUEST_BODY_FILE $request_body_file; #use php-cgi #fastcgi_pass 127.0.0.1:10005; #use php-fpm fastcgi_pass 127.0.0.1:9000; fastcgi_index $dir_index; fastcgi_param DOCUMENT_ROOT $doc_root; fastcgi_split_path_info ^(.+?\.php)(/.*)$; fastcgi_param SCRIPT_FILENAME ``` $document\_root$fastcgi\_script\_name; } i encountered one issue where in the /temp is a tmpfs (size is just 1GB), after uploading a couple of files i encountered this: \*58 pwrite() “/temp/0000000053” failed (28: No space left on device) shouldn’t it be flushed once the upload is done? or do i need to add some config to flush it automatic? Im using a php script that uses curl to do the uploading, here’s the script that manages the upload (curlupload.php) \ i removed the file after it is uploaded so i can run this script in loop, the script that uploads the file: (testupload.php) \ '123456','file\_contents'=\>'@'.$file\_name\_with\_full\_path); $ch = curl\_init(); curl\_setopt($ch, CURLOPT\_URL,$target\_url); curl\_setopt($ch, CURLOPT\_POST,1); curl\_setopt($ch, CURLOPT\_POSTFIELDS, $post); $result=curl\_exec ($ch); curl\_close ($ch); echo $result; ?\> Another thing i noticed is that when not using tmpfs, I/O is high the server when this is set: client\_body\_temp\_path /temp; fastcgi\_pass\_request\_body off; client\_body\_in\_file\_only on; fastcgi\_param REQUEST\_BODY\_FILE $request\_body\_file; Thank You in advance, any help or idea would be appreciated. Regards, Ron ---
**Author:** ![ron\_ramos](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/dc4da7/32.png) [@ron\_ramos](https://www.ruby-forum.com/u/ron_ramos)\ **Post date:** [August 15, 2013, 6:12am UTC](https://www.ruby-forum.com/t/flush-temp-directory/236203/2 "2013-08-15T06:12:53Z")
oooppp sorry found the cause “client\_body\_in\_file\_only on” changed it to clean and it’s doing it’s job. but am i missing something on PHP side or config side as it seems i am getting the same response time using php-fpm without the accelerated support settings and also using php-cgi, response time is the same for all scenarios. with accelerated support using php-fpm: 10.254.12.84 - - [15/Aug/2013:10:26:05 +0800] “POST /curlupload.php HTTP/1.1” 200 359 “-” “-” “-” 33.404 1.342 . 10.254.12.84 - - [15/Aug/2013:10:26:51 +0800] “POST /curlupload.php HTTP/1.1” 200 359 “-” “-” “-” 32.168 1.216 . without accelerated support using php-fpm: 10.254.12.84 - - [15/Aug/2013:11:02:58 +0800] “POST /curlupload.php HTTP/1.1” 200 359 “-” “-” “-” 32.182 1.229 . 10.254.12.84 - - [15/Aug/2013:11:03:32 +0800] “POST /curlupload.php HTTP/1.1” 200 359 “-” “-” “-” 33.218 1.208 . using php-cgi 10.254.12.84 - - [15/Aug/2013:11:48:57 +0800] “POST /curlupload.php HTTP/1.1” 200 359 “-” “-” “-” 32.371 1.418 . 10.254.12.84 - - [15/Aug/2013:11:49:30 +0800] “POST /curlupload.php HTTP/1.1” 200 359 “-” “-” “-” 33.093 1.308 . TIA Regards, Ron ---
**Author:** ![ron\_ramos](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/dc4da7/32.png) [@ron\_ramos](https://www.ruby-forum.com/u/ron_ramos)\ **Post date:** [August 15, 2013, 11:45am UTC](https://www.ruby-forum.com/t/flush-temp-directory/236203/3 "2013-08-15T11:45:52Z")
H i, I would really wonder if you would see a real difference between using a tmpfs or not for the webserver’s tmp body location. A tmpfs is only faster, but as long as your storage has enough free IO resources and is fast enough to actual write the data, you shouldn’t notice. And keep in mind: You only use the tmpfs for the request body. But you still need to write it to disk. If your disk is limited to 120MB/s and a normal upload is about 5 MB you are only able to handle ~23 concurrent uploads. Well, you could buffer millions of request per second in your super fast RAM (if you have enough RAM :P), but your PHP worker, which will move the upload from RAM to the persistent storage will become the bottleneck. I have a problem with the way it seems you test your setup: Every system should be able to handle that kind of load. After some runs, everything should be in some kind of cache. The IOs from the uploaded files are not enough (disks also have write caches, the OS may buffer writes, too…). These IOs can be handled by every disk, also, the IOs comes in sequence, not parallel. =\> Add more load. Run tests parallel/concurrent. Increase file size to fill up any write caches, which will trigger real writes, which will block the storage in some ways you will notice. ---
**Author:** ![ron\_ramos](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/dc4da7/32.png) [@ron\_ramos](https://www.ruby-forum.com/u/ron_ramos)\ **Post date:** [August 18, 2013, 2:54am UTC](https://www.ruby-forum.com/t/flush-temp-directory/236203/4 "2013-08-18T02:54:40Z")
Thank you Igor. I was just basically looking into this: [http://php-fpm.org/wiki/Features#Accelerated\_upload\_support](http://php-fpm.org/wiki/Features#Accelerated_upload_support) so im not quite sure if i am missing something out as it has the same results enabled or disabled. I will start testing it with multiple clients and see if any difference. Thanks again. On Thu, Aug 15, 2013 at 5:45 PM, Igor Sverkos