Jaeger remote sampling: strategy file updates not picked up by clients after collector ConfigMap change

Answer to the questions

1. Does a strategy file change require a collector restart to take effect, and if so, is there any supported hot‑reload mechanism?

• The 1.4.x collector loads the strategy file only once, at startup, via the --sampling.strategies-file flag. It has no file‑watcher and does not reload the document when the underlying ConfigMap changes.
• Therefore, a collector pod restart (or a rolling rollout of the deployment) is mandatory for new rates to be served. No hot‑reload flag exists in the 1.4.x line.

2. What is the defined client‑side fallback when the served strategy document version is unsupported, and is it logged?

• Clients poll the collector’s /sampling endpoint roughly once per minute. If the JSON strategy uses a version or field that the client library does not understand, the client usually keeps the last‑known strategy or falls back to its configured default sampler.
• Parse failures are logged at the client’s debug level; they are not surfaced as application errors. The exact log message and whether the client falls back to the default sampler depend on the language implementation (Java, Go, Python, Node, C++).

3. In a mixed‑version fleet, is there a recommended strategy format that all supported clients can parse?

• Use the classic JSON strategy format that consists of a defaultStrategy and a serviceStrategies array, each entry containing strategyType: "probabilistic" and a param: { sampling_probability: } field.
• Avoid newer fields such as per_operation or strategy_version that older clients ignore or reject. This format is understood by all Jaeger client libraries up to 1.4.x.

Practical steps for a 1.4.x deployment

1. Update the ConfigMap that mounts the strategy file with the new JSON content.
2. Trigger a rolling restart of the collector deployment, e.g. kubectl rollout restart deployment/jaeger-collector.
3. Verify the new strategy is served: curl http://collector:14270/api/sampling?service=my-service and check the probabilistic field.
4. Check client logs at debug level for a message such as "Updated sampling strategy for service my-service" to confirm the poll succeeded.

Because the collector does not hot‑reload, any ConfigMap change that you want to take effect must be followed by a collector restart. If you need a fully automated approach, consider adding a sidecar that watches the ConfigMap checksum and triggers a pod restart when it changes.

Do you know the exact collector version you are running and the client libraries in your services? That information would allow a more precise recommendation for fallback handling in your specific environment.