Rafael, thanks for sharing your experience with Thanos! It looks like it has many rough edges and needs careful cooking :)

Did you consider using other long-term remote storage solutions for Prometheus? They also support global querying view. Some of them are easier to set up and operate comparing to Thanos. I’d recommend taking a look at VictoriaMetrics, since:

• it has low operational overhead
• it is fast, so a single node may substitute moderately sized cluster built with competing solutions
• it is optimized for high cardinality data
• it seamlessly integrates with prometheus-operator via remote_write config
• it scales horizontally with cluster version
• it doesn’t require connections to Prometheus instances for querying recent data
• it has better on-disk data compression, so less disk space is required for long-term storage