I’ve recently been wondering if Lemmy should switch out NGINX for Caddy, while I hadn’t had experience with Caddy it looks like a great & fast alternative, What do you all think?

EDIT: I meant beehaw not Lemmy as a whole

  • lp0101@kbin.social
    5·
    2 years ago

    nginx 100% isn’t the problem, and you’re right on all counts. I’ll also add that I’ve seen reports that Lemmy has some pretty poorly optimized SQL queries.

    They need to add support for a message broker system like RabbitMQ. That way their poor postgres instance stops being the bottleneck.

    • TheTrueLinuxDev@beehaw.org
      2·
      2 years ago

      PostgreSQL is tricky to get right and I can’t fault anyone for wanting different solution like RabbitMQ to workaround it. One of the thing I did back in the day was that when dealing with high-write traffic and the data itself is not mission critical, I would set up a tmpfs on Linux for specified amount of RAM to serves as a cache to create a duplicate of the same data table used for storing on SSD/HDD and then I create a view that combines them both where it would check the cache first before querying the HDD/SSD.

      During an insert/update statement, it would trigger a condition that increment a variable (semaphore) and if reached a certain value, it would run a partitioned check on the cache table and scan for any old data that aren’t in active use based on timestamp and then have those written to HDD/SSD as well as writing to HDD/SSD if the data have been on cache long enough. Doing it this way, i was able to increase the throughput more than a 100 folds and still have data that can be retained on database.

      Obviously, there are going to be some additional risks incurred by doing this like putting your data on a volatile memory although it’s less of a risk on ECC Memory on Servers. If the power goes out, whatever stored on the RAM would be gone, so I assumed in cloud they would have backup power and other solutions in place to ensure it doesn’t happen. They might have a network outage, but it’s rare for servers to do a hard fail.