Author an AnalysisTemplate that specifies a Prometheus metric query measuring error rate or latency for the canary pods, and set success and failure conditions using the query result value
Define a Rollout manifest referencing the canary strategy with steps that include setWeight increments and pause durations, and embed an analysis block pointing to the AnalysisTemplate at the desired step
Configure the Rollout to reference the stable and canary Services so that the traffic split reflects the setWeight value and metrics are scoped to canary pods only
Deploy the Rollout and monitor the AnalysisRun object created automatically at the analysis step to watch metric evaluations and phase transitions
Verify promotion or rollback behavior by observing whether the AnalysisRun reaches a Successful or Failed phase and that the Rollout advances or aborts accordingly
Known gotchas
The Prometheus query must return a single scalar value; returning a vector or range without aggregation causes the AnalysisRun to fail with a parse error rather than an informative metric failure
If the canary Service selector is not properly isolated to canary pods, the metric query will include stable traffic and dilute the signal, making the analysis unreliable
AnalysisTemplate failure conditions use CEL-like expressions; a typo in the field name or operator causes the condition to be skipped rather than evaluated, so the run may succeed when it should fail
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp