@Moritz@sfba.social

@announcements@sfba.social

· 6mo ago

Yesterday later afternoon, our site experienced a massive traffic spike, sustaining almost 1Gbps for about 50 minutes.

What happened?
Our diagnostic tools revealed that a single IP address was stuck in a high-speed loop, requesting a single video file over 54,000 times in just 50 minutes—an average of more than 1,000 requests per minute. This was likely a broken client or an aggressive scraper.

Why there was no impact on you:
Crucially, there was no performance impact on other users. Because of our multi-layered Nginx caching, these requests were served directly from memory cache. This prevented the spike from ever reaching our Ruby/Puma application servers, keeping the site fast and responsive for everyone else while we identified the culprit.

This event was a perfect stress test for our new server environment, and it passed with flying colors.
Thanks for being part of SFBA! 🌉

30

2

11

1

@announcements@sfba.social
#SFBA #Mastodon #MastodonAdmin #Fediverse

· 4mo ago

🔧 Pardon the dust, we just bumped the dial to v4.5.11!

The servers are happy, the caching is cached, and everything is looking sharp. If you notice any weird quirks or edge cases, give us a shout.

Go forth and toot! 🌉🐳

SFBA.social

12

2

1

0

@announcements@sfba.social

· 6mo ago

Blocking Aggressive Crawlers

I just updated the blocklist for the Mastodon and BookWyrm instances. The logs showed meta-webindexer getting stuck in ridiculous infinite /rss/rss/ loops — a total waste of server resources pushing BookWyrm to 100% CPU utilization.

Additional Bans:

Meta-Webindexer/ExternalAgent: Recursive scraping gone rogue.

ClaudeBot: Keeping local content out of AI training sets.

Semrush & SERanking: Commercial SEO bots have no business here.

If you’re self-hosting and notice weird CPU spikes or odd path patterns, I highly recommend auditing your User Agents. It’s an easy way to protect your performance and your users' privacy. 🛠️

#SelfHosted #MastodonAdmin #BookWyrm #Nginx #Privacy #SysAdmin

21

3

13

0

· 6mo ago

🚀 We are officially caught up!

@announcements@sfba.social We just upgraded sfba.social to v4.5.8. This puts us right on the latest release—no more running behind! Thanks for your patience during the migration and these subsequent hops. Enjoy the latest features and the snappier performance! 🌉✨

23

2

5

0

· 4mo ago

Hello SFBA!

We've just upgraded to Mastodon version 4.5.10. Everything should be running smoothly, but please let us know if you encounter any issues. Thanks for being part of the community! 🐘✨

#SFBA #Mastodon

SFBA.social

11

2

0

· 7mo ago

Replying to

@announcements@sfba.social 🛠 Infrastructure & Performance Upgrades sfba.social just got a major engine upgrade to keep the neighborhood fast: Hardware: Migrated our main database to a dedicated host with 16 CPU cores and 32GB RAM for more memory headroom. Monitoring: Enabled pg_stat_statements to track and optimize slow queries in real-time. Database Tuning: Created targeted composite indexes, dropping some profile load times from 10-20s to <1ms. Cache Optimization: Tuned RAM to keep timelines and notifications "hot" for instant loading. The result should be a much snappier experience. Thanks for being part of the community! 🌉🚀

26

2

0

@announcements@sfba.social

· 6mo ago

Hello #SFBA!

As many of you know, we’ve recently migrated to a new infrastructure stack. While our logs show a 99.999% success rate, we know that only real-world usage can sometimes uncover edge cases that automated monitors miss.

We are still in the fine-tuning phase and want to ensure everything is buttery smooth before we decommission the old nodes.

How you can help: If you encounter errors, high latency, or posts failing to load, please don't be shy! Drop a reply here or message us directly with:

* The type of error (or a screenshot if possible).
* What you were doing at the time (e.g., favoriting a post, loading a thread).
* Whether you are using the web client or a specific app (like Phanpy or Mona).

We’re currently optimizing Puma worker threads and sidekiq queues to handle spiky workloads, and your real-world feedback is the best data we have.

Thanks for being the best testing community we could ask for! 🐘✨