Wilfried ROSET

Totally agreed for exposing this information into the cardinality api. This would help tenants better understand their usage

Thank you for sharing your work around BI like workload on loki. I've a somehow similar usecase. I would like to store an enriched [network flow](https://en.wikipedia.org/wiki/NetFlow) data model to provide...

> If you are using ebpf to monitor network requests, I recommend pre-aggregating on the agent side, and then pushing the pre-aggregated metrics to prometheus. I cannot pre-aggregate the data...

> The short version is that we are able to get query throughput out of Loki at rates of 50-100GB/s, we've had good success making Loki queries fast. Although there...

I'm facing a somehow similar situation while assessing Cortex where I'm unable to query data from swift due to missing index files. **Environment**: - Infrastructure: baremetal - Deployment tool: deb...

> There is no such functionality planned right now for ingesters. We suggest to address this by 1) making sure ingesters don't crash, and 2) make them reuse the hostnames...

> If you run Mimir in a single zone, you should restart 1 ingester at a time. If you run Mimir with multi-zone replication, you can restart 1+ ingesters in...

> I think we could reconsider it (e.g. re-shard to healthy ones only if unhealthy are < a given % of total ingesters). This would be a good middle ground...

If users run a single monolithic mimir (meaning a single alertmanager) they will face the bug explained in #2910. Starting from two instances, until the bug is fixed, they should...

This PR will probably need to be rebased after the merge of https://github.com/grafana/mimir/pull/3196