Lemmy.ml front page has been full of nginx errors, 500, 502, etc. And 404 errors coming from Lemmy.
Every new Lemmy install beings with no votes, comments, postings, users to test against. So the problems related to performance, scaling, error handling, stability under user load can not easily be matched given that we can not download the established content of communities.
Either the developers have an attitude that the logs are of low quality and not useful for identify problems in the code and design, or the importance of getting these logs in front of the technical community and trying to identify the underlying patterns of faults is being given too low of a priority.
It’s also important to make each log of failures identifiable to where in the code this specific timeout, crash, exception, resource limit is encountered. Users reporting generic messages that are non-unique only slow down server operators, programmers, database experts, etc.
There are also a number of problems testing federation given the nature of multiple servers involved and trying not to bring down servers in front of end-users. It’s absolutely critical that failures for servers to federate data be taken seriously and attempts to enhance logging activities and triangulate causes of why peer instances have missing data be track down to protocol design issues, code failures, network failures, etc. Major Lemmy sites doing large amounts of federation are an extremely valuable source of data about errors and performance. Please, for the love of god, share these logs and let us look for the underlying causes in hard to reproduce crashes and failures!
I really hope internal logging and details of the inner workings of the biggest Lemmy instances is shared more openly with more eyes on how to keep scaling the applications as the number of posts, messages, likes and votes continue to grow each and every data. Thank you.
Three recently created communities: [email protected] – [email protected] – [email protected]
You’ve raised many good points to come to a wildly accusatory conclusion.
Get off of this line of thinking if you want to raise support for fixing the valid issues you’ve raised.
Have you made any attempt to see if these issues have been raised on GitHub? Have you made any attempt to create issues on GitHub? Have you made any attempt to submit code enhancements via GitHub?
Have the developers of Lemmy put a message to end-users that data is being lost? have the operators of servers opened issues on GitHub about ‘pending subscribe’ failures on federation? Have the developers of Lemmy put warnings in the Release Notes that scaling is an active concern and needs urgent attention? Have the operators of major Lemmy websites published their nginx server logs and application code logs so that the bugs and design problems fo the code are logged?
Ir is the logging worthless, so poor, that they don’t share logs?
Most of all,e at your own dogfood, where are Lemmy developers using Lemmy to archive discussions about scalability and performance issues causing major crashes?
“I like Rust” but don’t like using Lemmy to discuss the performance concerns of scaling and data integrity in a monolithic Rust application?
It looks like you want to vent and issue demands of others rather than be productive. I hope that serves you well.
And how is this reply productive? **You are just trying to cover up the very real scaling problems of Lemmy code and project mismanagement by insulting me **personally for having autism that you misinterpret. It took me 3 weeks of courage to make this post about how mismanaged this project is, and here you are - DENIAL of the problems in the design and code.
If I want to vent: I’ll go to Reddit and start making postings about how unreliable Lemmy is as an application and how it can’t even share community comments on a daily basis - and the developers don’t give a single shit to read their own Lemmy communities and make postings about needing help with Rust webapp 101 and PostgreSQL.