## TL;DR
"Insufficient data" means the analysis window lacks the metrics to compare: the canary got no traffic, the metrics have gaps, or the baseline period has no data for a new service. Fix it by ensuring the canary actually receives traffic (synthetic if needed), aligning metric windows with data availability, and giving new services a baseline period before canarying. An analysis with no data is honest; forcing it is not.

## The query
```text
canary analysis "insufficient data" errors: how to fix metric gaps
```

## Use this when
- Canary analysis fails with insufficient data
- New services cannot establish baselines
- Metric gaps invalidate comparisons
- After changing metric collection

## Not for when
- Canary promotion/rollback strategy
- Metric query correctness (different issue)
- Load testing

## Steps

### Step 1: Verify the canary received traffic
Check request counts for the canary during the analysis window. Zero traffic means zero data; the fix is traffic (real or synthetic), not analysis tuning. New endpoints with no callers cannot be canaried.
Expected output: canary traffic confirmed present, or the traffic gap found.

### Step 2: Check metric continuity in the window
Look for gaps in the metrics during baseline and canary windows: collection outages, new metric names after a deploy, or retention boundaries. Gaps invalidate the statistical comparison honestly.
Expected output: the metric gaps located and explained.

### Step 3: Align windows with data availability
Shorten or shift the analysis windows to periods with complete data. A 30-minute analysis over a window with 20 minutes of data is the common shape of this error.
Expected output: windows covering only well-measured periods.

### Step 4: Establish baselines for new services
New services need a soak period with traffic before canary analysis means anything. Run the baseline version under load first, collect the metrics, then start canarying changes against it.
Expected output: a real baseline existing before the first canary.

### Step 5: Add synthetic traffic where real traffic is thin
For low-traffic services, synthetic checks generate the steady baseline canary analysis needs. Real user traffic is ideal; consistent synthetic traffic beats sparse real traffic for statistical validity.
Expected output: sufficient data density for reliable analysis.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_6F7LfTFIRSfv9KyqD4YpPg