David @mleku - 1y
i discovered that the decode cache was not counting the capacity of buffers but only their usage, and process memory was up around 500mb... i changed the allocation strategy so now every time a new buffer is used it adjusts a simple moving average, and when new buffers are made, they are 20% larger than this to avoid reallocation relay now runs comfy in around 100Mb this is more like what i expect it should be, and the size of the encode cache is configurable, as is the runtime garbage collector trigger (with more memory you want to raise the GC percent towards 100 and the decode cache size is default at 10Mb but you might want to raise that as high as 500Mb for heavy traffic) fantastic seeing the memory utilization where it really should be, it will not fluctuate nearly as much as without the JSON encode cache, and 100Mb memory usage for a small local database is so good i think it could even be compiled to WASM and used in a browser client https://github.com/Hubmakerlabs/replicatr/releases/tag/v1.2.3 need to finish the vacuum thing i was gonna make so i could populate a goodly sized data set to push to the layer 2 data store to demo a fresh deployment syncing up rapidly to a shared layer 2 data store today #devstr #replicatr #nostr #relay #layer2
