Problem
A node that starts slightly earlier than its peers probes them before they are
listening, fails once, and then waits the full first backoff interval before
retrying. With the default policy this is ~3s, during which the node reports
"service not ready" although its peers become ready within milliseconds.
Observed
Three nodes started ~14ms apart: the earliest node needed ~3.0s to pass the
cluster ready check; the other two needed ~5ms.
Problem
A node that starts slightly earlier than its peers probes them before they are
listening, fails once, and then waits the full first backoff interval before
retrying. With the default policy this is ~3s, during which the node reports
"service not ready" although its peers become ready within milliseconds.
Observed
Three nodes started ~14ms apart: the earliest node needed ~3.0s to pass the
cluster ready check; the other two needed ~5ms.