Stuff that's happened since November on the Infra side: - Racked some new hardware in our datacenters and got it burned in and provisioned - Migrated Timelines/Following to its own database cluster to greatly improve perf - Moved static asset css/js delivery to a CDN making first-load much faster - ThreadSky

jaz.bsky.social • 42 days ago

Stuff that's happened since November on the Infra side:

- Racked some new hardware in our datacenters and got it burned in and provisioned
- Migrated Timelines/Following to its own database cluster to greatly improve perf
- Moved static asset css/js delivery to a CDN making first-load much faster

Comments

datadrivenmd.social•42 days ago

The improvements have been noticeable esp during yesterday's new user influx. Thank you for your effort, and glad to hear that you stepped away to recharge

jaz.bsky.social•42 days ago

Some stuff still coming down the pipe to tackle soon:

- Service Discovery (lite) for our internal services to provide better reliability during unexpected hardware outages and provide warmup modes for services
- More serious caching architecture to reduce cold-start pain
- Making more use of new HW

jaz.bsky.social•42 days ago

And my stretch goal for this year is probably something like building a custom database for Timelines since it turns out the fixed-length, circular-buffer-per-user style with high write throughput (>1M writes/sec) pattern is not a great fit for LSM-tree based DBs or like, any conventional DBs really

aparker.io•41 days ago

oooh you and @quail.wtf should talk

jaz.bsky.social•42 days ago

I've been pretty busy doing ^ and more, needed to take a bit of a step away from posting over the holidays during the higher stress moments and to get a bit of space away from work. Turns out it's kinda hard to take a break from work if I spend lots of my free time on the site.

richburroughs.dev•42 days ago

I’m glad you’re taking care of yourself

leonardopds.com•42 days ago

Thanks for all you hard work!

rahaeli.bsky.social•42 days ago

The perpetual problem with working on a social media site is that you inevitably find yourself stopping using the site -- it's not just you!

shawnsblog.bsky.social•42 days ago

twidget.wtf•42 days ago

Thank you. I've noticed a general improvement in my user experience even as you've had battle explosive user growth.

ls1228.bsky.social•42 days ago

Build build build!

rainbowrewind.bsky.social•42 days ago

Would it be possible to edit our posts at some point? Sometimes my autocorrect hates me and I don't notice until after I've posted and already gotten likes 🙄

mmccue.bsky.social•42 days ago

love that you guys are hosting on your own old school

jaz.bsky.social•42 days ago

It's made scaling a lot less scary from a cost standpoint.

I'm a firm believer that one of the largest scams in the software industry today is egress fees from cloud providers. Every cent I pay for cloud egress makes me incredibly frustrated considering I _know_ how little they're paying for BW.

jaz.bsky.social•42 days ago

It's got drawbacks too of course, you've gotta bake in a lot of headroom when you do scale, you need a lot more lead time, and it gets harder as your physical footprint gets much larger.

That being said, you'd be surprised how far you can get with a handful of modern machines and some DIA hookups.

mmccue.bsky.social•42 days ago

yeah we hosted Tellme on our own back in the day in three data centers. Flipboard is all on AWS. Seriously tempted to go back to hosting on our own given all the massive hw and sw progress. Kudos for pushing this. Will keep the hosting providers honest!

justingarrison.com•42 days ago

Hey Mike 👋 I host the @fafo.fm podcast about building and maintaining software.

Would you or someone from flipboard be interested in coming and talking about your challenges scaling and maintaining what you’ve built?

mmccue.bsky.social•42 days ago

sure! Cc @marcimccue.bsky.social

Comments

Posting Rules

Reply