XMPP Service Operators - 2021-04-24

ext-to-btrfs just sounds like a disastrous idea

just buy or rent a hard drive/cloud storage and migrate the data

or better use, use zfs

it's unstable immature crap like that pushed into the "stable" releases of distributions that made me stay the hell away from btrfs

back in debian jessie, btrfs didn't even warn you when it encountered a corrupted chunk

not even something in dmesg

running btrfs is a minefield of this

» has been broken for years, it will immediately and irrevocable corrupt your partition

» did that on a laptop 2 years ago and it worked without any issues typical redhat (IBM) mentality. Works on my machine!

therefor, good enough to sip

ship

"we only use ssds so we removed that feature"

zfs-on-linux is the beta never-gonna-be-good stuff :)

btrfs is in the kernel, it works today and you can assume it'll still work tommorow

btrfs is *extremely* stable when used in the single, dup, or raid1 modes

file system conversions will never be 100% perfect

I use btrfs in raid1 and single depending on the machine, but I use it everywhere

zfs on linux is the main dev repo now. It *is* the good stuff

but it'll never be merged and is often broken by upstream linux changes right ?

brtfs vs. zfs 🍿️

main dev repo? mainline or of zfs?

the only thing I hate about zfs is that cache RAM usage shows as used, not buffer

zfs for the win :p

» <octagon> btrfs is *extremely* stable when used in the single, dup, or raid1 modes yeah but that's not impressive and doesn't offer much over an lvm or hardraid

moparisthebest: ZoL uses the same zfs codebase as all the other openzfses

hwraid doesn't have checksumming, and hwraid/lvm has no compression, dedupe, or snapshots

mdraid is tedious to maintain

including FreeBSD, which is switching to OpenZFS because that's where the good stuff is happening

» <octagon> hwraid doesn't have checksumming, yes it does

It's called Type 2 protection and can be had with HBA controllers too

yea hardware raid is an immediate hard no

it's still useful for zfs in that it can prevent some stupid errors relating to the sas cable plug being loose from happening

btrfs and zfs both have wins and losses over each other, depends what you want really

I just don't think so

but you should still use btrfs or zfs wherever possible, they are modern file systems no reason to not use them

there's no scenario i can think of that btrfs does that zfs or some other solution does a whole lot better

tom, can I setup a raid1 array with 2x2TB drives today and then add a 3TB drive tommorow and then replace one of the 2TB with a 4TB the next day with ZFS? can with btrfs

I'd still rather do XFS w/ checksumming on a lvm raid1

» <moparisthebest> tom, can I setup a raid1 array with 2x2TB drives today and then add a 3TB drive tommorow and then replace one of the 2TB with a 4TB the next day with ZFS? can with btrfs you can append vdevs to a pool

why not? you can add devices to a mirror in zfs

adding vdevs is a hack

I'm under the impression you can't "change shape" with zfs

reconfigure your topology on the fly is a hack

what about changing raid levels ?

until you issue a 'rebalance' your data is in a very unsafe state

like going from single to raid1

and won't fix itself until you do rebalance manually

but it's not like it's going to warn you that some of your blocks are only partially replicated

> like going from single to raid1 zfs can do that

in-place without downtime ?

» <moparisthebest> I'm under the impression you can't "change shape" with zfs there are some restrictions with regard to Zn and vdev types

not sure about the growing/"changing shape"

yes

yes, you can add a device to make a mirror on the fly

and the good thing is you only sync what is occupied, unlike mdraid

you just can't reconfigure the topology of a raidZn vdev within a pool

you can grow it and add/remove other vdevs tough

*though

you can add multiple raidZn vdevs to a pool

so you can't grow a zpool?

yes you can grow a zpool

there are many ways to do it

you can grow the vdevs and/or add more storage type vdevs to a pool

this is how mine looks

pool: tank state: ONLINE scan: scrub repaired 0B in 04:02:08 with 0 errors on Thu Apr 8 02:19:10 2021 config: NAME STATE READ WRITE CKSUM SLOW tank ONLINE 0 0 0 - raidz1-0 ONLINE 0 0 0 - scsi-35000c5004247a387 ONLINE 0 0 0 0 scsi-35000c50042488edf ONLINE 0 0 0 0 scsi-35000c5004248a547 ONLINE 0 0 0 0 scsi-35000c500560c720f ONLINE 0 0 0 0 scsi-35000c500565cbc4b ONLINE 0 0 0 0 special mirror-2 ONLINE 0 0 0 - ata-Samsung_SSD_850_EVO_M.2_1TB_S33ENX0H902345P-part3 ONLINE 0 0 0 0 ata-HFS500G32TND-N1A2A_FI64N09421020524Z ONLINE 0 0 0 0 errors: No known data errors

I plan to replace the special mirror-2 vdev with 12GB/s SAS ssds when I can afford more hardware

until then I had a m.2 ssd with extra space and a sata ssd i threw in

i plan to replace the drives within the mirror-2 vdev and then grow that vdev to accomodate more space, then adjust some of the datasets so they are fully accelerated

currently, just tiny io and metadata are stored within solid state

because spinning rust is terrible at performance with tiny io

or random io

so i just send the tiny and random to the ssds, which they are good at

best of both worlds!

and I get a little bit of hardware error-correction across the full path so even a a faulty cable won't cause an error which the software needs to correct and slow things down via the Type 2 protection which my HBA card supports

the disks use 520-byte sectors instead of 512

8 bytes are used for hardware ECC

the rest is given to ZFS to use it's purposes

this allowed for error correction in the cables, not just the data

but the sata drives are just stupid and don't have anything fancy

just consumer drives

btw, FreeBS13 will use the OpenZFS codebase (the same one OpenZFS on Linux uses)

or you could use it right now by installing the zfs.ko from ports

which you should do anyways if you want things like encrypted datasets

do you use external SLOG on the SSDs?

not right now. I used to but currently I just don't have a lot of synchronous writes

apt-get is painfully slow for me under zfs, even with sync=disabled. I hope to solve it with a slog but I'm not sure

keep in mind that a raidZ(n) has the random performance of 1 spindle

if sync=disabled did not fix your problem try adding a special vdev xorman. It sounds like the apt problem may be database expensive

I'm using a mirror

how many mirror vdevs?

2

oh, so 2 spindle write perf

apt-get over spinning ext4 is way faster

I would suggestion adding a solid state special mirror, and then setting the pool's special allocation class to 64K

assuming your other datasets's recordsize is 128K

yeah, ext4 isn't copy on write

I might try with sync=enabled and libeatmydata

what is that?

it's a workaround to disable sync writes in dpkg

why does dpkg use a lot of sync writes?

zfs should handle the synced writes anyway with sync=always

as a test at least

the only things that need to be persitent in a package manager is the list of installed packages

and ya know the packages it installs 😉

everything else can be redownloaded or regenerated

I heard apt-get has performance issues with btrfs too

the thing about zfs, well hard drives in generally really, is that if you send too many random writes to them you can starve out everything else

a drive might be capable up pumping out 200MB/s largeio and 400MB/s sequential, but if you starve it with randomio it's capped to 4MB/s

or smallio

you know what has performance issues on zfs? lmdb and monero

that's a big mess. I have to occasionally copy and delete the data around to defrag it

hopefully I won't have to do that as much with the special class, or anymore ounce i get bigger ssds

it's because using LMDB is CoW on CoW

Can you not chcon and disable cow for the folder/files?

chattr*

I guess ethereum may the same problem bc I tried to run a validator and it missed all the blocks due to i/o lag

zfs is cow

that's how it fails safe

any DB on top of a Cow filesystem is not going to be fun :)

Somebody from www.conversejs.org online ? Got a HSTS-Error, your cert is only named for conference.conservejs.org, but not for www Maybe your should add it. Https//conversejs.org is working well

I've pinged JC

Thanks ...

HI! I am evaluating setting up an XMPP server. Do you suggest to me ejabberd or openfire? Or other?

nicola, i'd suggest ejabberd or prosody

or snikket if it fits your usecase

https://snikket.org/

> nicola, i'd suggest ejabberd or prosody Tks. I seem that ejabberd it simpler to install. Am I wrong?

both are equally simple

> or snikket if it fits your usecase Interesting.

nicola, snikket is in my opinion simplest to set up, it is geared towards friends&family and I recommend it for that

(disclaimer: I am a contributor :))

@jonas’ tsk ;-)

> nicola, snikket is in my opinion simplest to set up, it is geared towards friends&family and I recommend it for that What are the differences with ejabberd?

nicola, Snikket is a complete suite: Android client, iOS client (closed beta), Server including web interface for easy management

the XMPP server component is a preconfigured prosody with hand-selected modules for ease of use and setup.

as the package is always tested together, many of the issues like "which modules do I need?" "does X work with Y?" don’t occur

being standard XMPP, you can federate with other Snikket (and normal XMPP) domains or use it with other (non-Snikket) XMPP clients such as Dino or Gajim on non-phone systems. ✏

> being standard XMPP, you can federate with other Snikket (and normal XMPP domains) or use it with other (non-Snikket) XMPP clients such as Dino or Gajim on non-phone systems. Tks. Very interesting. I see that the installation is very simple by docker

that’s the goal :)

I was also about to recommend Snikket if you are new to server hosting

we also have a chat: xmpp:general@channels.snikket.org?join

if you have any questions or issues with setting it up or something, that’s the place to go :)

Ok. I’ll try to install Snikket and I’ll come back here. Thank you again

have fun! :)

The iOS app is only for Snikket or is it possible to add other XMPP server accounts?

I’m not sure, I don’t have an apple device myself

I suggest you hop into general@channels.snikket.org and ask there, there are people who are in the iOS beta.

however, during beta, standard snikket server users are preferred because that’s the main focus currently.

Here I am, again.

I installed Snikket

Using the Snikket iOS app with servers other than a Snikket server or a Tigase server will lead to sub-optimal results

hi nicola :)

I am using on iOS Siskin IM

Probably this is not the right place, but I see a warning on the SSL connection. Why?

Which server are you connecting to?

https://chat.nicfab.it - self hosted

nicola, hmm, do you have a reverse proxy in front?

The certificate on that site says it is for a different site

> The certificate on that site says it is for a different site Ok. I try to understand

I have a firewall and I thought that the SSL connection was automatically set during the installation

I installed certbot. Can I use it to generate certificates for the Snikket site or it is managed by the docker?

Snikket will generate its own certificates

It just needs to listen directly on port 80/443, or you need to configure a reverse proxy if you have other things on the same machine

> Snikket will generate its own certificates If I set port 80 it doesn’t work

Because I'm guessing you have something else already using port 80

nginx or apache for example

Documentation is here: https://github.com/snikket-im/snikket-server/blob/master/docs/advanced/reverse_proxy.md

Ok. Now it works fine

Great :)

I need to set up a reverse proxy. Ho can I access to nginx ?

Depends on your system and how you installed nginx. On most systems you'll find the config at /etc/nginx

Sure! I thought nginx was installed inside the docker

There is an nginx inside the docker containers but you never need to touch that (and shouldn't... any changes you make will be lost when it updates)

The documentation I linked is for if you already have nginx or another reverse proxy on your system, so you can share port 80 and port 443 between multiple sites

> The documentation I linked is for if you already have nginx or another reverse proxy on your system, so you can share port 80 and port 443 between multiple sites Ok. Tks

that is a very bad connectivity issue

yeah thanks for the heads up, on a bad connection

Alex, the lowercase you though

seems like a bug more than anything else

not me then

I would be curious actually, maybe the lowercase thinks it is in conflict with the uppercase and prevents the final join, then it reconnects after

sadly we’ll never know

I was wondering if in the future it will be implemented the registration from the web page instead of by the admin via shell

nicola: ejabberd and prosody both support registration via web

And both support inband

> nicola: ejabberd and prosody both support registration via web Yes, I know. I was referring to Snikket

Snikket allows registration via invite to a webpage too.. You can just send people a link and they registrate from there..

I don't think snikket will ever do open registration..

Easy to setup spam server

I like the link + web page method

I might eventually that up on my prosody server

I wonder if you have multiple vhosts, if they can choose from any of them

Probably better for the prosody channel

There is IMMENSE lag to yax.im from my server

Talking like over 30 seconds

What's going on over there?

Is your server paging in and out of swap or something?

Be gentile to yax.im its handling the vaxvbot flood > https://yaxim.org/blog/2021/04/09/vaxbot-performance-challenge/

Yes yax.im is under a lot of stress

💚

I'm getting the urge to write a bot that monitors various system things on your server and reports issues, but also responds to queries. As though I have time for more projects

aioxmpp library looks good and would be fun to practice Python

rob: Jonas has that?

Oh, the bot? That works be cool

Or the library? In which case yes it's theirs

This is hilarious https://yaxim.org/blog/2019/04/01/yaxim-enters-the-matrix/

Thanks for writing this

Especially this part: » cvwright on Feb 21, 2019 [–] » » If everyone shared your (lack of) optimism, then we'd still be stuck with: » » * Wonky unwieldy hypertext systems instead of the WWW » » * The Nomad instead of the iPod » » * The Blackberry instead of the iPhone » » * GNU Hurd instead of Linux » » * Geocities instead of Facebook (-- OK maybe that one's a wash :-) » » The point is, sometimes it's worth trying again until you can make it work. Hello, I would fuckin love to live in a world like that

Can you imagine?