Pod Phase Versus Readiness

Kubernetes Fundamentals / Guide 8 of 46 / Beginner

• Kubernetes1.36.4
• Cluster4 nodes
• Runtimecontainerd 2.2.6
• CNICalico v3.32.1
• TimeAbout 15 min

conditions.ready on EndpointSlice is the field that matters here. The older Endpoints object expressed the same thing by moving addresses between addresses and notReadyAddresses; EndpointSlice keeps one list and flags each entry, which is why the address never disappears below.

This guide includes

Use this when a Pod says Running and the Service in front of it answers nothing. This matters because phase and readiness are different states, and only readiness decides whether traffic is sent to a Pod.

• running one Pod with a readiness and a liveness probe reading two different files
• seeing a Running Pod refuse every connection, and finding ready=false against its address in the EndpointSlice
• creating the file the readiness probe wants and watching traffic switch on
• deleting the file the liveness probe reads and watching the container get killed and restarted
• telling the two probes apart by what their failure costs - no traffic, versus a restart

Before you start

• A cluster and kubectl. Both probes are exec probes over ordinary files, so nothing needs to be installed or reachable from outside.
• The session creates namespace kx-probe, a Deployment web with a readiness probe on /tmp/ready and a liveness probe on /tmp/live, a Service web, and a client Pod to call it from inside the cluster.

1. 1

   One Pod, two probes, two files

2. 2

   Running, and refusing every connection

3. 3

   One file turns the traffic on

4. 4

   The other file gets the container killed

Official sources

• Configure Liveness, Readiness and Startup Probes
• Pod Lifecycle - container probes
• EndpointSlices