first time seeing this btw

I've figured out that I've been measuring latency a little wrong

it takes around 23 seconds to process all ~35M logs

which is, well, still seems suboptimal, but if I filter a specific service by label, that has ~4M logs -- I get ~3s per query of 100 lines, which is actually, I would say, decent

but I'm scared to image what horrors would happen if I try to query the data over longer periods of time

and also I realized that the service that has the least useful logs -- produces 80% of the logs

chat check out the grafana I've deployed: http://127.0.0.1:3000/

deploying VictoriaLogs to compare it with Loki...

took me some time to figure out how to properly do log levels when using VictoriaLogs datasource in Grafana

it's still undocumented, but you can specify it as:

          jsonData:
            logLevelRules:
              - field: level
                operator: equals
                value: INFO
                level: info
                enabled: true
              - field: level
                operator: equals
                value: WARN
                level: warning
                enabled: true
              - field: level
                operator: equals
                value: DEBUG
                level: debug
                enabled: true

in the datasources yaml

levels: critical, error, warning, info, debug, trace, unknown
operators: equals, notEquals, greaterThan, lessThan, regex

it also has ingestion compatibility layers with every possible protocol/format in the world

crazy

consuming logs

set up both Loki and VictoriaLogs, will wait for a day and see how it performs after consuming many kiloliters of logs

I suspect that memory usage might go quite a bit up, surprisingly enough it consumes less memory and CPU than Loki at the moment

didn't notice this initially, but it also has a "beta query builder"

200k logs/second ftw

comment maxxing on pull requests

16 million logs later, the results are insane

VictoriaLogs still consumes less resources than Loki, runs queries 100x faster and yet supports high-cardinality indexes with full-text search

same query over 16M of logs on both Loki and VL:

Grafana Loki: 10-11s
VictoriaLogs: 250ms

official Jaeger integration just got approved today too

we've never been so early

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

dear god…

7 seconds away

7 seconds away...