Honeycomb

Use BubbleUp to select the slow request outliers on a heatmap.

Inspect the attributes surfaced (e.g., 'customer_id' or 'region').

Identify that a specific legacy database shard is the bottleneck.

Pivot to the trace view to see the exact SQL query timing.

Filter data by 'build_id' to compare the new version vs the stable version.

Monitor the SLO error budget for the new build.

Roll back automatically if the error budget burn rate exceeds the threshold.

Query for unique 'user_id' where 'status_code' is 500.

Export the list of affected users for proactive communication.

Identify the 'plan_type' of affected users to prioritize VIP customers.

Open the distributed trace for a specific slow request.

Count the number of spans associated with database calls.

Pinpoint the loop in the code generating excessive sequential queries.

Deploy Refinery in your infrastructure.

Set a rule to keep 100% of errors but only 1% of successful HTTP 200 requests.

Apply dynamic sampling based on traffic volume.

Group by 'trace.parent_id' and 'error' to see failure patterns.

Look for correlation between failures and specific 'host' or 'pod' names.

Identify a misconfigured load balancer in a specific availability zone.

Instrument LLM calls as individual spans with attributes for 'model_name' and 'token_count'.

Analyze the correlation between prompt length and response time.

Monitor for 'finish_reason' outliers indicating model hallucinations or safety blocks.