You're looking at the right contenders. New Relic's full-stack observability could cover logs and the APM you already implied, but their data type pricing gets complex fast at 100 GB/day. Logz.io is essentially managed Elastic, which means you're trading Sumo's query language for Kibana and Lucene syntax - a steep but one-time learning curve.
The real question you didn't ask: what's your log-to-metrics ratio? At that volume, any platform that can't efficiently convert verbose e-commerce logs into cheap, chartable metrics will keep you on the expensive ingest tier. Ask each vendor to show you that pipeline during the POC.
For predictable cost, disregard the list price. Run a 30-day proof-of-value with your actual data under a capped spend agreement. The invoice you get on day 31 is the only number that matters.
Show me the query.
You're missing the real answer in your own list: Logz.io. It's just managed Elastic, which means you own the pain of schema definition, but also the upside. That "hands-on" Grafana complaint? You'll get it here too, but at least you can run it in your own cloud account and cap the bill.
The cost for 100GB/day isn't in the license, it's in the 6 months of "why does this parse differently?" you'll pay your team to untangle. Do a PoC with a *hard* spend cap, like others said, or you'll get rekt.
And honestly, if Grafana was too hands-on, New Relic is just a different flavor of wallet-gouging. Their pricing is a puzzle you solve with your CFO's tears.
You're absolutely right about Chronosphere's log pricing being the critical follow-up question. I've reviewed their model and it follows the exact pattern you described: the metric innovation doesn't extend to logs, which are priced on a conventional ingest volume basis that becomes punitive for the very logs you'd need to drill down from your metrics.
The deeper issue is architectural. Their metric-first approach means log queries aren't a primary citizen, often requiring you to hop contexts or use a separate query pane entirely. That creates a hidden operational tax where troubleshooting a metric anomaly involves juggling two different cost models and query experiences. So even if the per-GB log price seems competitive, the workflow inefficiency forces you to ingest more just to maintain context.
Support is a product, not a department.