@AndrewSavinykh They can be imported in Java.

@Dennis apparently different packages are installed by default in DO and vultr.

on vultr nano is installed by default

on DO vim-minimal is installed by default

but not other way around

@Dennis what are you using darcs for?

@Dennis and what is the actual purpose behind chrony? Is not fedora has ntp syncronization out of the box?

@Dennis so I'm testing on DO now, dnf update subjectively runs slower than on vultr

@AndrewSavinykh That's weird. vi is part of the POSIX standard.

@AndrewSavinykh The download or the installation?

@AndrewSavinykh I don't recall. Did I install that manually?

@AndrewSavinykh Define out-of-the-box. All Linux distros need some ntp daemon, and RedHat prefers cronyd over ntpd.

It's supposed to sync faster.

@Dennis yep

@Dennis ah right. I thought that it would be installed along with linux, not separately

@Dennis both ;)

again I have not exactly timed, but on vultre the whole arena setup took about 30 minutes (before I had to compile factor, it probably added another 5 minutes) on DO it's been running for an hour and has not been finished yet

just finisihed

so took about an hour

I know that install time does not worry you, but I think it's indicative of overall performance

Those extra 256 MiB of RAM should be at least partly responsible for that.

@AndrewSavinykh With Linux distros, the line between part of the system and just another package is quite blurry. There are almost no parts of the system that cannot be uninstalled and/or replaced with something else. This includes the kernel.

I like how DO/vultr charge you by the minute so you can quickly try something out, tear it down and pay just a couple of cents for that

Yeah, I did that with a FreeBSD droplet a few days ago. Had to test a code golf answer.

The $5 vultr VPS seems like a sweet deal, but I don't think 15 GiB of disk space are enough anymore. I only have 2.7 GiB left in the arena, although I could probably free part of the disk space.

@Dennis yep their storage nodes are 512 for 5$

/dev/vda1 20575868 11859172 7830108 61% /

this is DO

/dev/vda1 15420800 13640280 977776 94% /

oh-oh

this is vultr

@Dennis i have a feeling that half of the package is texlive which something installs as deps. haskell perhaps

they must take quite a bit of space

I got away with texlive-dummy for a while, but some package required the "real" texlive. Don't remember which.

Hm, the interpreters in /opt occupy only 5.2 GiB.

And there's 2.2 GiB in logs. Lord.

@Dennis mine is 2.6GB

And you installed all of them?

@Dennis yes

How is that possible?

@Dennis it may be my linux incompentence

what command is supposed to be used to find out the size

I used du -hs /opt

is this wrong?

No, I did the same.

if you know the command to list size per subdirectory send it to me and your output, I'll compare

The disk hogs:

5.2G    /opt
1.6G    /swapfile
6.4G    /usr
2.5G    /var

inside /opt

Sorted by size.

Sorted alphabetically.

@Dennis Why the hell is Haskell so huge?

Yowza, trying to view the logs froze up the arena.

@Pavel It's the whole Haskell platform. Not really sure what's included.

@Dennis So I did not install haskell inside opt

this accounts for the most of differnce

@Dennis tio.run/nexus/bash#RYkxDoMwEAS/…

These are the folders that I do not have

Right, a few of those were created but not used.

@Dennis you are running haskell from /usr/local/bin/runhaskell how come it's in /opt?

It refused to install into any other directory. I should be able to delete the folder in opt though.

Alright, down to 13 GiB used.

Without removing the Haskell folder. I'll try that later.

@AndrewSavinykh I needed darcs to clone some interpreter that I ultimately wasn't able to install.

@Dennis fair enough

This one.

Right, I'm guessing if you could not do this even with ais523 help, it's impossible ;)

The problem is that its tokenizer requires some packages that do not exist for Fedora.

Or they do, but they were hidden really well.

@Dennis I'm guessing this is something that is bound to happen with any systems. At least you have working selinux ;)

That's pretty much the sole reason I went with Fedora anyway. Anything else should be fixable or a reasonable compromise.

I guess CentOS would be a viable alternative.

@Dennis so got arena running on DO but going to tear it down soon. The main blockers was that some packages were missing (installed by default on vultr) that dnf does not freaking working on DO until you do dnf update, did not have such problem on vultr and then because of memory before you mount swap file pip won't even update

but those all are minors, I'll use this info when I'm restructuring the scripts to be more sane

@Dennis why cheddar is in /opt? you are calling it from elsewhere?

Geez, the /opt/haskell folder contained the tarball I downloaded, the tarball inside that tarball, all the extracted files, and all the files from building... And none if that is needed after installing!

@AndrewSavinykh Downgoat wanted me to compile Cheddar from source, but that required too much memory.

@Dennis that's bizare I deleted all of that and my net total is still 13GB

/o\ I installed Julia 0.3 and wrote a wrapper, but never added it to the JSON file.

Well, that has been fixed.

@AndrewSavinykh Without my Trash folder, I'm down to 10.6 GiB. That's without a swapfile and reduced logs.

I've added MaxRetentionSec=1week to /etc/systemd/journald.conf and restarted the service to keep only the last week's log on the arena server. That should be more than enough.

@Dennis well my 13GB includes a swap file of 1.5GB

@Dennis why would you even need 0.3 if you have 0.4 and 0.5? this time you cannot say that this is so that you don't break the permalinks ;)

I added the swapfile to try building Cheddar from source but never remembered to. With too much traffic, it probably does more harm than good.

@AndrewSavinykh Golfing. Julia's versions are incredibly incompatible and get less golfy with every release. If I could get them from somewhere, I'd add 0.1 and 0.2 as well.

Why did you need swap to build Cheddar? How much RAM does the droplet have?

512 MiB

Some npm thing wouldn't compile.

What' the limit on how much memory a single program can allocate?

I'm guessing I can't just execute malloc for 60 seconds straight.

256 MiB, but if memory gets full, the OOM killer will nuke the process anyway.

All limits.

The number of open files is 192 because PowerShell.

That's a little silly, a must admit.

My favorite is the System.OutOfMemoryException

Can you get it to print a trace? Really intrested in why that happens.

That would require breaking PowerShell temporarily. I'd rather not.

Fair enough.

@Dennis ulimit?

ah, saw the link

yes ulimit ;)

ulimit sets some limitations for resource usage. It's fairly important because fork bombs could bring the arena down otherwise. Let me find the configuration file.

@Dennis cheers. so are you saying you had to decrease them?

There are no limits by default.

# cat /etc/security/limits.d/99-tio.conf
runner soft cpu             2
runner hard cpu             2
runner soft core            0
runner hard core            0
runner soft fsize        4096
runner hard fsize        4096
runner soft nofile        192
runner hard nofile        192
runner soft nproc         128
runner hard nproc         128
runner soft rss        262144
runner hard rss        262144
runner soft sigpending    128
runner hard sigpending    128

@Dennis on ubuntu there are. When I tried to open 65000 ports it did not work and I had to use ulimits command

There are some limits by default:

core file size          (blocks, -c) 0
data seg size           (kbytes, -d) unlimited
scheduling priority             (-e) 0
file size               (blocks, -f) unlimited
pending signals                 (-i) 31570
max locked memory       (kbytes, -l) 64
max memory size         (kbytes, -m) unlimited
open files                      (-n) 1024
pipe size            (512 bytes, -p) 8
POSIX message queues     (bytes, -q) 819200
real-time priority              (-r) 0
stack size              (kbytes, -s) 8192

That depends on your distro.

@Dennis what do I need to make it work, just to dump the file into the directory?

But I stand corrected, Fedora comes with fairly sane limits by default.

@AndrewSavinykh Yep, that's enough.

This one is also important btw.

# cat ~apache/.ssh/config
Host arena.tryitonline.net
ControlMaster auto
ControlPersist yes
ControlPath /tmp/%r@%h

It keeps SSH sessions to the arena open since creating a new one is fairly slow.

That file belongs on the main server btw.

@Dennis erm

why did not I notice that it is slow?

Good question. Maybe I don't need it after all.

Against my DO arena without it the whole test suite is 7 minutes 3 seconds. Running now with it

same

@Dennis can't observe a difference

Interesting.

@Dennis I certaintly notice, there's a pretty big difference for simple programs.

But programs where it's noticable are generaly small enough that it's not an issue anyway.

@AndrewSavinykh Are you using the private network?

No

I am. I wonder if that makes a difference.

@Dennis does it make a difference

I did it mostly since there are no transfer limits on the private network. Not sure if it impacts speed.

@Dennis but you don't know your bandwidth usage?

Not even a clue. I think the limits on DO are nominal and not really enforced at this point.

@Dennis on DO there is no /var/log/messages Where do syslog go? On vultr there is this file

Geez, looks like vultr uses a heavily modified Fedora image.

journalctl shows the logs.

It reads from /var/log/journal, which stores the logs in compressed form.

@Dennis and where do I get messages from that I write with logger

I can't find nothing in journalctl output

I'm not familiar with logger.

@Dennis did you ever need to write anything in syslog?

the logger thingy I've figured out, just wondering if you had to use anything similar before

So far, /var/log/httpd on the main server contained all the information I needed.

@Dennis I also think that I do not have pypy-devel and SDL-devel

I installed pypy-devel for pip iirc, which I needed for Mathics. Mathics on PyPy is a terrible idea though.

Did you change it twice?

No. You probably got an old version from your browser's cache.

But I did Ctrl-F5 if I remember well...

Even then, if you go back in history (for example), you could get served an old version. I' not saying this is what happened, but I'm out of ideas if it's not that.

Well, it's 04:34, I'd recommend getting some sleep for now. You might get an idea if you wake up fresh and after your morning coffee...

Because I don't think a hard refresh is supposed to load from cache, I think it's supposed to replace cache.

As long as it doesn't happen again, I'll just blame gremlins.

The cache for a website as a whole and the cache of a previously opened tab are different. The latter still contains the same values in textareas, etc.

@AndrewSavinykh I have installed Cinnamon Gum, Binary Lambda Calculus, and PowerShell on the same day I installed SDL. I can't think of a reason why the first three would require it, so it's likely this was part of the trial and error of the PowerShell install.

If it ever happens again, you can be sure gremlins are not to blame. I just hard-refreshed all of my TIO tabs.

@Dennis thank you

Alright, kill script is done, now I just have to call it from the frontend.

@Dennis after these commands

can I rm sandbox_extra.* ?

or are these files required?

sorry files/ should not be there

@Dennis I was reviewing some old code-golf and found that this solution no longer produces output. Locally, it still prints out continuously. The question was Pythagorean triples

@Dennis tio.run/nexus/… produces a bunch of junk output, do you know why?

@AndrewSavinykh That should be safe. Once an SE module is installed, it stays installed unless there are naming conflicts.

@miles Yeah, infinite output still needs some work. I thought I had fixed that already...

@Pavel No clue. I'll try to figure it out.

@Dennis I'm looking for good guesses. In a script I install with dnf glibc-devel and then install python modules with pip. In Vultr it works on DO the very same script fails complaining that gcc cannot find gnu/stubs-64.h When the script finished running the stubs-64.h is in place and if I re-run pip it works. Any idea why that could be? I'm not at least consciously run anything in parallel, so I don't see how this can be

Huh, glibc-devel is the sole provider of /usr/include/gnu/stubs-64.h.

Your droplet is clearly haunted.

Not helpful, sorry, but that is my best guess for the time being.

@Dennis that's all right it's very difficult to suggest something under circumstance. I'm working through it

@Dennis github.com/TryItOnline/beeswax/blob/master/build why there is a julia part in the end after exit 0? Does it get invoked somehow?

@Dennis this script fails when run without $Home from a service on stratup ;) I'm trying to "fix" it

@AndrewSavinykh Line 8 writes the Julia code into a file.