dear god…

7 seconds away

7 seconds away...

vibecoding in it's purest form

I'm done, anysphere

I use neovim, btw

https://github.com/VictoriaMetrics/VictoriaMetrics/pull/8988

victoriatraces
victoriatraces

I've collected roughly 100M logs over the last few days in both Loki and VictoriaLogs

VictoriaLogs finally started to utilize a little more memory than Loki

Query time amazingly is still ~250ms for full-text search (not labels)

I don't even want to try running same query with Loki, it will just explode...

(it did explode)

actually, it didn't

It looks like under the hood Loki splits up the data into 24h chunks and fetches them one-by-one

I've seen this in the docs, but didn't realize it works this way

However, it still takes astronomical amount of time to search, ~20-25s per each chunk. So the request to do full-text search on a week timeframe would take ~2.5 minutes

and it also uses all of the available CPU time to do a search, I wonder if it has any negative effects on the ability to effectively consume the data

After all it does seem to have sense to run Loki only as a Simple Scalable Deployment, where you can have a dedicated writer target(s) and a bunch of read target(s) that can process your request in parallel. But this will get expensive really fast

VictoriaLogs also seem to consistently use less storage than Loki

$ du -sh victoria-logs-data
936.1M  victoria-logs-data

$ du -sh loki
1.6G    loki

step 1. download cursor
step 2.

holy moly vector.dev is powerful

all my concerns regarding quality are lifted

I've been considering building something similar myself in Grafana Alloy. vector.dev provides this out of the box...

there's even an ability to unit test your configurations: https://vector.dev/docs/reference/configuration/unit-tests/