tl;dr: Join rooms on hackint.org from Matrix via
#channel:hackint.org. Say hello from Matrix in #hackint:hackint.org (for #hackint)!
Back in 2018 we did announce the sunsetting of the matrix bridge. Back then our primary points of concern were:
- memory intensiveness
- excessive logging to database, even though we're only relaying information
- clients are occasionaly not properly rejoined, but might be able to read messages nevertheless user connections are never culled, even when they become unused
In the last couple of years we've tested the bridge in multiple rounds in regards to these issues. Over the course of that time the bridge Matrix ecosystem and the bridge code has improved a lot. Especially on the Matrix side there have been huge improvements in terms of memory and cpu consumption as well as retention policies. We'd like to take the time to address each of these points and why we think that we are now able to provide a bridging setup again.
On memory intensiveness: Earlier this year the Synapse team put a lot of effort into reducing the memory consumption of Synapse. The requirements to join large channels has since dropped significantly and we haven't seen a lof of memory (or CPU) usage on regular (fairly busy) homeservers. That being said, the matrix bridge is likely the single most busy server that hackint operates. :)
On logging & database growth: Around the time of our sunsetting the Matrix team started the discussion on message retention (MSC1763). Since then synapse gained support for purging messages based on configurable schedules. We will purge messages (and media) rather agressively and not store it for extended periods of time. This means that we will only backfill history to users that were joined within an hour of a conversation happening, and even that is conditional on a channels privacy settings. This should cover most network segmentations while keeping the data on our servers limited. A note on media: Even if media has already been purged from our services it can (and will) be retrieved from the origin server again when requested.
Last but not least the actual IRC bridge problems: The IRC bridge hasn't seen much love for the last couple of years until the Freenode accident happened. The Matrix team was very eager to provide Matrix bridging to Freenode's spiritual successor Libera.Chat and they started discussing their concerns! This has led to fixes for most of our long-standing gripes with the bridge. You can read more about this in This Week in Matrix 2021-06-18 matrix-appservice-irc notes and the related GitHub issues, conversations etc..
Do we think it is perfect? We wouldn't say so. It is a long way away. You will certainly see issues but the overall experience isn't as rocky as it used to be. This is why we feel confident enough to offer the bridge again.
Heads up! We're going TLS1.3 during this month. The first machine has already been upgraded today and everything looks good.
- If your clients supports TLS1.3 (or TLS1.2) you don't need to do anything, you will be able to reconnect without reconfiguration.
- If you're however running on an older system, please note that we disable support TLS1.0 and TLS1.1 in this change.
Older news can be found in the archive.