Unsolicited opinion: If #Mastodon wants to be a part of the #Fediverse that encourages small #selfhosted instances to be part of the whole, then it has to be a lean system able to be installed on second-hand/hand-me-down hardware and it should run reas...

Edwin G. :mapleleafroundel:

@chris I spent a couple of days without ElasticSearch… it was horrible for searching my own posts.

Now, I will have a look at OpenSearch (?). It might fit my purpose better than ES.

[email protected]

IRC, or XMPP at worst, should be the gold standard to follow here. Those things can be installed on Pentium IVs that you can get from rural school trash yards, and still run reasonably well.

Jérôme

@EdwinG @chris it's a fork of ElasticSearch, it's unlikely to make a big difference. It was mostly a licensing dispute.

SomethingGeneric

@chris id argue it should also try its best to become single exec, so ir can be OS/distro agnostic too, but that's a major PITA

Chris Alemany🇺🇦🇨🇦🇪🇸

@SomethingGeneric a long term but very worthy goal!

Chris Alemany🇺🇦🇨🇦🇪🇸

@EdwinG yes I saw mention of it in the last release notes. I will be interested in your experience!

Sean Boyer 🇵🇸 FREE PALESTINE

@chris @EdwinG
I switched to opensearch from es and there was no appreciable difference to the memory it will consume. It's the same thing different license basically.

Ed Wiebe

@jerome @chris Java is used for portability, not performance. There are lots of things that would be much better. So many 'database' and 'indexing' things are done really badly with tools that aren't 'database' and 'indexing' tools. However, I am not in charge and happy not to be.

Philipp Krenn

@chris
1. not sure you're seen it already, but if you have docker you can just run "curl -fsSL https://elastic.co/start-local | sh" to get going (https://www.elastic.co/guide/en/elasticsearch/reference/current/run-elasticsearch-locally.html#local-dev-quick-start)
2. while elasticsearch won't be tiny, the most common problem is that people don't configure the heap size and then elasticsearch assumes that it should own the entire machine. will depend on your activity and the initial data import but 1GB should get you going.
I'd also say that some of the java dislike is a bit dated at this point

Philipp Krenn

@jerome @EdwinG @chris I'm clearly biased here but we'd say that it's rather the opposite: https://www.elastic.co/blog/elasticsearch-opensearch-performance-gap

Chris Alemany🇺🇦🇨🇦🇪🇸

@xeraa thanks! I’m running a strait install on Debian. Can you point to some configuration options for the Heap size?

Jérôme

@xeraa @EdwinG @chris haha yeah I did some research yesterday and indeed the performance seems to be on the elastic side.

Always happy to see some elastic engineers over here!

Philipp Krenn

@chris yes: https://www.elastic.co/guide/en/elasticsearch/reference/current/advanced-configuration.html
1. create a file in /etc/elasticsearch/jvm.options.d/ with the custom config (this will survive upgrades; don't change the jvm.options file directly)
2. try something like 700m as a starting point. you might be able to go a bit lower. or with more activity you might need to go higher. there are details on the setting page for Xmx + Xms — this is what you want

Philipp Krenn

@jerome @EdwinG @chris now I'm interested. anything you can share?

and it's great here. always happy to jump into some discussions!

Chris Alemany🇺🇦🇨🇦🇪🇸

@xeraa @jerome @EdwinG this is really wonderful that you're here Philipp. Your advice is really interesting and opposite to what I've seen elsewhere. My old iMac (2008) has 16GB of RAM, I had been told previously to look at allocating 25-50% of the RAM to Java which translates to a mammoth 8GB. I'm going to try cutting it right down to 1GB and see what happens.
What are some metrics or tools I can use to determine if the RAM it is using is suitable?

Philipp Krenn

@chris @jerome @EdwinG so there is on and off heap (mostly caching) memory needed; normally you'd split it into equal parts. our smallest cloud instance has 1GB of memory and 500MB of heap
your use-case might need more (depending on how hard you push it). but just to get started you should be able to go with 1GB total or maybe even a little less
and I'd start with the JVM memory pressure to debug: https://www.elastic.co/guide/en/elasticsearch/reference/current/high-jvm-memory-pressure.html

Philipp Krenn

@chris @jerome @EdwinG the main pain you'll experience is that by default elasticsearch assumes it should use the entire machine / VM / container. so it allocates 50% of the memory for its heap. if you have a 16GB instance, it doesn't automatically need 8GB for the heap. but you need to tell it what you feel is appropriate

Chris Alemany🇺🇦🇨🇦🇪🇸

@xeraa @jerome @EdwinG thank you very much for the advice! I will tweak things and see how it goes!

Chris Alemany🇺🇦🇨🇦🇪🇸

@xeraa Thanks again for your help.
I set it down to
-Xms1g
-Xmx1g
elasticsearch reports: heap size [1gb], compressed ordinary object pointers [true]

I have a fairly active account, but my little instance is relatively quiet as there are only three of us on it.

I had originally set the heap size to Xmx/Xms2G as recommended in the Mastodon docs. But maybe that's too much for my purposes and I should have reduced rather than increased its memory.

The log seems very quiet, but no obvious errors that I can see.

Chris Alemany🇺🇦🇨🇦🇪🇸

@xeraa I'm not sure setting it to a lower value has done much good. This is the report from htop. There are 50+ processes from elasticsearch all claiming 7.1% memory which obviously adds to more than 100% so I'm thinking this is causing some serious issues here. The system is very sluggish.