maggie44
commented
11 months ago Here is a bash script for testing:
#!/bin/bash
# Set the API URL
api_url="https://api-url-to-test"
# Function to check API status
check_api_status() {
# Use curl to make a GET request to the API URL
response_code=$(curl -s -o /dev/null -w "%{http_code}" "$api_url")
# Check if the response code is 200 (OK)
if [ "$response_code" -eq 200 ]; then
echo "$response_code"
else
echo "API is not reachable. Response code: $response_code"
fi
}
# Infinite loop
while true; do
start_time=$(date +%s.%N) # Get the start time
# Call the function to check API status
check_api_status
end_time=$(date +%s.%N) # Get the end time
duration=$(echo "$end_time - $start_time" | bc) # Calculate the duration
echo "Loop took $duration seconds."
# Sleep for 0.5 seconds before the next iteration
sleep 0.5
doneDuring the restart one of the requests drops (i.e. active users receive errors):
Loop took 0 seconds.
200
Loop took 1.0 seconds.
API is not reachable. Response code: 503
Loop took 0 seconds.
200
joliveirinha
Describe the bug
I am running multiple replicas of Cloudflared on a Kubernetes cluster to serve traffic to a Kubernetes service. When I restart the deployment the tunnels initiate a rolling update, starting a new instance before terminating the running one. Each of the Cloudflared instances appears to be exiting gracefully as expected:
I have the probes set up for starting new replicas as follows. I can observe them taking effect as expected:
I have
minReplicas: 3and the following rolling update strategy:When I observe the rolling update I can see all three instances running, then it scaling up and it waiting 10 seconds as expected for the startup probe before exiting the old instance.
From a Kubernetes perspective everything seems to be working as expected. Yet when trying to access through the domain name during these restarts I keep encountering
HTTP Response Code: 502:This is through a Cloudflare WARP installation. I can replicate it through an online domain (also with Cloudflare) as well though, although instead of the 502 as above one of the refreshes in my browser never makes it.
Expected Behaviour
With the graceful shutdowns happening as expected I had expected Cloudflared to inform Cloudflare services to not relay any requests to the shutdown/shutting down Cloudflared instances.
Environment and versions Kubernetes 1.29 Cloudflared: 2024.1.1