hyperknot's questions

I have the following nginx config:

server {
    location ~ ^/(\d+)/(\d+)/(\d+)\.png$ {
        alias /mount/cache/c/$1/$2/$3.png;
        try_files $uri @proxy;
        add_header x-source file;
    }

    location @proxy {
        proxy_pass http://127.0.0.1:28000;
        add_header x-source proxy;
    }
}

What I'm trying to achieve is to try to load the files from disk, and if they are not found only then ask the upstream for it.

My problem is that try_files doesn't seem to work, it always requests @proxy (x-source header is always proxy). If I comment out try_files then it loads the files from the disk correctly, but then the proxy part doesn't work.

I'd like to delete an nginx cache directory, which I quickly purged by:

mv cache cache.bak
mkdir cache
service nginx restart

Now I have a cache.bak folder which has 2 million files. I'd like to delete it, without disturbing the server.

A simple rm -rf cache.bak trashes the server, even the simplest HTTP response takes 16 seconds while rm is running, so I cannot do that.

I tried ionice -c3 rm -rf cache.bak, but it didn't help. The server has an HDD, not an SSD, probably on an SSD these might not be a problem.

I believe the best solution would be some kind of throttling, like how nginx's built in cache manager does.

How would you solve this? Is there any tool which can do exactly this?

ext4 on Ubuntu 16.04

From experimenting, it seems that even when using ps aux | grep ... the output depends on the terminal size. This seems seriously weird for me.

Isn't a pipe supposed to make command's output behave in a predictable way?

For example ls does all fancy formatting magic when run on its own, but when run like ls | less it always returns the same output styling.

But unlike ls, the output from ps aux | less cuts the lines to terminal width on FreeBSD (on bash at least). From my brief experimenting, the same command always outputs full length lines in Linux when used with pipes.

Why is this happening and how can I fix it?

Is this just a really weird FreeBSD default? From my understanding this "feature" would make it almost impossible to write reliable bash scripts on FreeBSD as the output of simple piped commands would not be reliable. Isn't this a problem?

Update: Thanks to Bert, I've tried it with ps aux > somefile and the lines inside the file are cut exactly like how it was with grep and less!

Info: FreeBSD 10.2, bash 4.3.42, same when connecting over SSH or when logged in locally via KVM.

I made a website, behind CloudFlare, and sometimes I get an early terminated (nginx 499) request, which is really hard to work around in the codebase.

This happens in about 1 out of 4000 requests.

What I would like to ask is if this is normal or not? CloudFlare support told me that 499 happens when something between CF and the server drops the connection. I'm not on AWS, not on a load balancer, but on a simple VPS with direct IP connection (no virtual routers, etc.)

CloudFlare has a great option in PageRules to set up URL forwarding. I can set it to *.example.com/* to redirect everything to a different domain.

My problem is that while this works perfectly, this needs valid A / CNAME records in the DNS page. For example a server or an S3 bucket. I don't get why is this needed, since no requests will ever hit the server via this forwarding.

My question is: is there a way to set up URL forwarding via CloudFlare which doesn't require running a server?

I am using a dedicated server, and unlike on VPS providers or VM machines, getting back to a clean state (untouched, after install) is quite a laborious thing to do (lots of clicking around in a web control panel and launching Java KVM viewers + doing the installation manually).

My question is that given ZFS support for snapshots, would it be possible to get back to a clean state just by issuing a single command?

My ZFS partition table would be the "Auto (ZFS)" offer by the installer, which looks like this on my server (using FreeBSD 10.2):

NAME                 USED  AVAIL  REFER  MOUNTPOINT
zroot               34.8G   176G    96K  /zroot
zroot/ROOT          1.46G   176G    96K  none
zroot/ROOT/default  1.46G   176G  1.46G  /
zroot/tmp           76.8M   176G  76.8M  /tmp
zroot/usr           33.2G   176G    96K  /usr
zroot/usr/home      32.6G   176G  32.6G  /usr/home
zroot/usr/ports      631M   176G   631M  /usr/ports
zroot/usr/src         96K   176G    96K  /usr/src
zroot/var            692K   176G    96K  /var
zroot/var/audit       96K   176G    96K  /var/audit
zroot/var/crash       96K   176G    96K  /var/crash
zroot/var/log        204K   176G   204K  /var/log
zroot/var/mail        96K   176G    96K  /var/mail
zroot/var/tmp        104K   176G   104K  /var/tmp

My questions are the following:

1. Is it possible to get back to this clean state using ZFS snapshots? What if I changed the kernel, deleted half the file-system, etc. Can a snapshot revert back everything like if it was just after installation?

2. If yes, how can I do that? What command do I need to create the snapshot and what do I need to revert to one?

In FreeBSD, I'd like to set the timezone to UTC in an unattended way. If I run tzsetup, and select "yes" and "a" in the dialogs, for choosing UTC, I see that there are no /etc/localtime files present.

So if I want to set up UTC on a server in an unattended way, can I just rm /etc/localtime instead of running tzsetup?

I'd like to set values in php-fpm conf files what are normally set in php.ini. I'm using nginx.

I've created the following setting, but I'm not sure if this would work.

php_value[memory_limit] = 96M
php_value[max_execution_time] = 120
php_value[max_input_time] = 300
php_value[php_post_max_size] = 25M
php_value[upload_max_filesize] = 25M

Do you think if this is OK like this?

What happens when a value is both set in php.ini and in php-fpm conf files? The php-fpm overrides the ini one?

Finally, isn't it a problem that this way I can set different values for all virtual hosts? I mean php.ini seems like a global setting, while this is host dependent. Can different hosts run with different memory-limits, etc?

If I want to run the command

md5deep -rb . | find /I ".jpg" | sort > local.md5

and I run it from the command prompt, it shows exactly like how I typed, but when I put it in a .bat file and run that, this is what I see:

md5deep -rb .   | find /I ".jpg"   | sort  1>local.md5

It seems that it inserts tabs before | and it turns > into a 1>. Why? First I thought that it's something to do with the encoding, but the file is just a simple ANSI file.

Does it have anything to do with @@ turning @ inside .bat files?

I would like to create an MD5 file of a directory tree, containing only the bare names of the files. md5deep exactly does that, but it cannot filter for files when in recursive mode.

md5deep -rb . > md5.txt

Is there any way to filter for (*.jpg) before outputting it to md5.txt? In Linux it would be just a |, but how do I do it in Windows?