Hello!

Good question - let me come back to you. can you link me to your workspace please?

Yep: 89f79d0c-6f3c-4937-be66-a82e42a608a3 this is our project id

Thank you!

Thanks!

Oh Eden! Sorry I didn’t recognize you on the fly on mobile.

No worries lol

@rahul any updates on this?

Hey - will come back to you this evening if that’s ok?

What are you doing on the cloud side that you need to egress to? We have bucket storage now on platform - are we missing another piece you need?

Not questioning your needs - just want to cover on all angles make sure we get you what you need

We are a very media heavy application, so that means a lot of video/image processing happens on our servers and we stream data back to clients (full videoplayback).

For this we use Cloudflare R2 at the moment because it's one of the only providers with truly free egress pricing and a pretty competitive per TB pricing ($15/TB/mo). Other competitors like wasabi/telnyx/idrive e2 have 3x/4x of stored size limits on egress/mo.

We'd love to use the bucket storage on Railway, but egress pricing is the main concern for us here as well. From what I can tell you don't have pricing figured out yet for it right?

Ok - I can’t make any promises yet but help me understand: if we could do something on the egress, could you move a bulk of compute from elsewhere to railway?

Yes! currently we're testing using https://runpod.io as a proxy basically. like our railway services call a runpod serverless endpoint that does egress heavy work.

If the bucket pricing makes sense we'd love to just move everything to railway. For some context, this 18.84TB is with only about our 30ish beta test users in the last month of iteration

Bucket pricing will be in line w most places - I expect $20/TB or so. The egress problem though - where does that processing happen? Railway or elsewhere?

Ie. is there a fleet of vms somewhere that could be here? Or is it just storage?

There's two parts to the egress stuff:

1. compute: we process videos/files/do transformations and then upload them back to R2. This causes egress when doing it from within railway
2. client side: we serve the media to clients from within storage (this is basically free for us on r2 atm)

Ok that part I understand. But 1 is happening on railway already right?

It’s process > ship (egress) on railway.

Store > distribute (free egress) on r2.

Correct?

process > ship (egress) this month we moved to runpod as a test

Got it.

How much egress across each leg?

process > ship: 1.5x what we store

store > distribute on r2: >2x what we store (just based off of the data being loaded in for processing)

Will come back to you but I’m not hopeful there’s much I can do here 😔

ahh okok let us know!

even with the dedicated vms (the 10k spend), we'd be paying the same 5c/GB egress?

I’m going to run some modelling and see. Sorry - kids bedtime etc

No worries, sounds good. And thank you so much!

where does egress need to be for Railway to be viable for you on dedicated?

wdym by where?

like price range?

yeah from the 5c

off the shelf we have no discount on egress -- it's not something we maek money on, we just pass it through.

QQ: What is the expected object store size on R2?

Would you be open to having the obj. store in Railway?

yes if we can figure out the egress

R2 is free egress on the distro

wait, Im blind

20 TB yea?

20TB rn, but we're looking at closer to 100-200TB in the next few months

20TB is the size of the bucket from 0 in the last month with only a small set of users

Cool, so hypothetically, if we can handle the obj. store and you can avoid a network hop. We get you a win/win, one less upload and less cost yea?

Yep!

I do wanna confirm with our vendor that we can eat these objs. before we commit to something.

But! Rahul is already batting for you internally, but egress we'd likely not be able to touch since it's sold at cost and a loss in some cases.

(We don't region price our egress)

ahh i see got it

Thank you so much Rahul!

He's fighting for you hard btw

Omg bless ❤️ ❤️ We love railway and it has made our dev experience so much nicer coming from building https://kortex.co on ec2 and k8s ourselves and now https://eden.so

Like it has saved us so much time/energy/money, AND everyone your team is just fantastic.

give us a bit to spin the rubks cube here -- whats your urgency?

is this like a 'solve tonight' or a 'few days is ok'

no urgency on it atm!

we're onboarding about 3-4k users on the 24th of November

ok so before that you need a solution

yeah, ideally

but we can go about it iteratively, like if you guys know you can support our use case but it'll take a bit longer we can continue with our hacks for now until you guys are ready. Our public access is set for sometime in January which is where we have a lot more than 3-4k users coming on lol

i think we can do something here -- give us a day or two though to bottom it out

i want to stress - cautiously optimistic, but aiming to be more confident in 24-48 hours

i don't liek ove promising.

Ari, one more question, the egress charges you'd expect is because you would be streaming media to your clients from Railway? Or, is it exfil to your object store that you'd expect to be the main source of egress fees?

Sorry about the delay, didn't see the ping

yes! the object store is expected to be the main source of egress

ill ping you tomorrow and we can run through some options!

Sounds good

Thank you so much!

@Ari how is 3pm est for you?

I have a podcast 2-5, could we do either before or after

1pm est? ill send something if so

yep

sounds good

you shoudl also haev a slakc invite from me - let's move there

your whole worksapce should

Yes will do!

Just got it

Closing out as we have moved to Slack! <:salute:1137099685417451530>