Mooar debrid toys

It's too soon to tell whether it's "just a passing phase", but the "infinite" streaming app bundles seem to be a popular alternative to traditional torrenting / usenet. It's a good fit for us, since (a) the apps can be complicated to interconnect, (and automating complexity is our jam), and (b) they rely primarily on compute and network resources, which are cheap and easily expandable, rather than storage, which is a stone-age PITA .

So for your instant-streaming enjoyment, I present the following "instant streaming" guides:

• Plex, with "infinite storage"
• Plex, with "infinite storage", managed with Arrs
• Jellyfin, with "infinite storage", managed with Arrs
• Emby, with "infinite storage", managed with Arrs

We've also got an alldebrid bundle ready for a trial, if that's your flavor.

Read more for capacity expansion, and changes to TDarr and WebDAV access to meet resourcing constraints...

100% more compute

All these new trial users have put some pressure on our nodes, and rather than loosing the momentum of new trials, or turning users off with a bad experience, I've added 100% more compute nodes to manage the load.

I'm taking the opportunity to experiment with "right-sizing" our compute nodes, based on what's available to us on Hetzner's server auction. It turns out that we max out the nodes on CPU well before RAM becomes a constraint, so our i9-9900K 128GB nodes may be too "dense". It may be more effective to replace these with more 64GB nodes, and a combination of i9s (QSV support) and i7s (no hardware transcoding), letting Kubernetes schedule the streamer pods onto the QSV-supporting nodes.

BYO compute

A conversation in #elf-friends sparked a left-field idea.. users with large Plex libraries / audiences may not "fit" in our regular product usage.. and may want more CPU, RAM, transcoding space, or network capacity.

It's hard to accommodate these users on our shared infrastructure, but it may be possible to provide services on "dedicated" nodes - for example, we'd be able to offer "Plex Unchained" on a dedicated i9-9900K 64GB node like this one:

We'd have to work out the details like setup fees, pre-payment (I have to pre-pay for the nodes), and management overhead, but we could effectively dedicate an entire node to an individual user, giving them all the CPU / GPU for their streaming / transcoding requirements.

If this is interesting to you, reach out and let's talk :)

Tdarr removed for bad behaviour

There have been a few kernel lockups over the past few days, and the evidence seems to suggest it's related to TDarr instances (unless configured explicitly per the docs, TDarr will eat much more RAM than allocated, quickly get itself OOMKilled, rinse and repeat, causing storage i/o drama and generally making a nuisance of itself). I've taken it "off the menu" for now, and may need to bring it back if / when BYO compute is a thing.

If you were one of the handful of users with TDarr on your subscription, it'll be removed, so you won't be charged for it any longer.

WebDAV egress dropped to 5MB, ingress unlimited

Until now, WebDAV access (which provides the facility to "exfiltrate" your data and config from any attached storage) has been throttled to 10MB/s via rclone CLI arguments. On a hunch that this may be doing more harm than good re long I/O delays, I've removed the CLI arguments, but used Cilium instead to limit the egress of WebDAV to 5MB/s.

This means that taking data out of ElfHosted is limited to 5MB/s, but putting data in will go as fast as the storage can handle (we've seen about 40MB/s for one user today)

Today's scoreboard

Metric	Numberz	Delta
Total subscribers	143	+13
Storageboxes mounted	19	0
ElfStorage in TBs	48	+1
Rclone mounts	43	+10 1
Tenant pods	1518	+178
Bugz squished	1
New toyz	1

Summary

As always, thanks for building with us - feel free to share suggestions, and your own ideas for new apps to add!