ECS Deployment FAQs

Frequently asked questions about ECS Blue-Green deployments.

This article addresses some frequently asked questions about ECS deployments in Harness.

What's the difference between the Swap and Traffic Shift steps?

There are two distinct flows for ECS Blue-Green deployments. Only one can be used in a pipeline:

Swap Step (Traditional Flow)

• The BG deploy step asks for prod and stage listener rules (two separate rules, typically on different ports)

• Does NOT ask for stage target group input

• Target groups are determined automatically based on which listener rule they're attached to

• Running the pipeline multiple times with the same configuration automatically alternates two versions of services between Blue/Green

• At the end of each deployment, the target group attached to the prod listener rule will have the blue service

Traffic Shift Step (New Flow)

• The BG step asks for one listener rule only (production rule)

• This rule must be attached to precisely two target groups: stage and prod

• Requires manual input for the stage target group ARN to determine which of the two target groups is stage vs prod

• This design choice was intentional because users often have configurations like 60-40 or 20-80 traffic splits, not just 0-100

• With non-zero traffic on both target groups, Harness cannot automatically determine which is stage vs prod

Why doesn't Harness automatically determine the stage target group like it does for ASG?

In ASG deployments, Harness automatically determines the prod target group as the one with 100% weight. However, for ECS traffic shifting:

• Users often maintain traffic configurations like 60-40, 20-80, or other splits

• It's not always 0-100, making automatic determination unreliable

• Requiring explicit stage target group input gives users complete control over which target group receives the new deployment

Why does my active service get deleted during deployment?

Symptom: You see logs like:

Deleting existing non-blue version Service: echo-service__1
Waiting for existing Service echo-service__1 to reach inactive state
Deleted non-blue version Service: echo-service__1

Root Cause: Harness deletes any non-blue services attached to the stage target group at the start of each deployment. If you reuse the same stage target group that contains your current blue (active) service, Harness will delete it.

Solution: You must alternate the stage target group between deployments:

1. First deployment: Use target-group-a as stage target group → After 100% shift, target-group-a has blue service

2. Second deployment: Use target-group-b as stage target group → After 100% shift, target-group-b has blue service, target-group-a has green service

3. Third deployment: Use target-group-a as stage target group → Continues alternating

If I constantly shift 100% traffic, do I still need to alternate target groups?

Yes. Even with a 100% traffic shift, you must alternate the stage target group between deployments if you want to maintain two versions of services alternating between Blue and Green (similar to the traditional swap flow).

What happens if you don't alternate:

• First deployment: Blue service __1 is created in the stage target group with 100% traffic

• Second deployment (same stage target group): Harness sees service __1 attached to stage target group, considers it non-blue (because it's in the stage target group), and deletes it before deploying new service __2

• You lose the ability to maintain two alternating versions

What happens when you do alternate:

• First deployment: Blue service __1 in target-group-a with 100% traffic

• Second deployment (using target-group-b): Service __1 is now tagged as green (in the prod target group), blue service __2 is created in target-group-b with 100% traffic

• Third deployment (using target-group-a): Service __2 is now green, blue service __3 is created in target-group-a with 100% traffic

How can I achieve traffic shifting without manual stage target group selection?

You can automate stage target group selection using a shell script step before the ECS Blue Green Create Service step. This eliminates the need for manual selection on each deployment:

#!/bin/bash

# Configuration <a href="#configuration" id="configuration"></a>
LISTENER_RULE_ARN="<+pipeline.variables.prodListenerRuleArn>"
TARGET_GROUP_A="<+pipeline.variables.targetGroupA>"
TARGET_GROUP_B="<+pipeline.variables.targetGroupB>"

# Get current listener rule configuration <a href="#get-current-listener-rule-configuration" id="get-current-listener-rule-configuration"></a>
RULE_CONFIG=$(aws elbv2 describe-rules --rule-arns "$LISTENER_RULE_ARN" --query 'Rules[0].Actions[0].ForwardConfig.TargetGroups' --output json)

# Extract weights for each target group <a href="#extract-weights-for-each-target-group" id="extract-weights-for-each-target-group"></a>
WEIGHT_A=$(echo "$RULE_CONFIG" | jq -r ".[] | select(.TargetGroupArn==\"$TARGET_GROUP_A\") | .Weight")
WEIGHT_B=$(echo "$RULE_CONFIG" | jq -r ".[] | select(.TargetGroupArn==\"$TARGET_GROUP_B\") | .Weight")

# Select the target group with 0% traffic (or lower weight) <a href="#select-the-target-group-with-0percent-traffic-or-lower-weight" id="select-the-target-group-with-0percent-traffic-or-lower-weight"></a>
if [ "$WEIGHT_A" -lt "$WEIGHT_B" ]; then
  STAGE_TARGET_GROUP="$TARGET_GROUP_A"
  echo "Selected Target Group A: $STAGE_TARGET_GROUP"
else
  STAGE_TARGET_GROUP="$TARGET_GROUP_B"
  echo "Selected Target Group B: $STAGE_TARGET_GROUP"
fi

# Export as output variable <a href="#export-as-output-variable" id="export-as-output-variable"></a>
export STAGE_TARGET_GROUP_ARN="$STAGE_TARGET_GROUP"

Then reference the output in your ECS Blue Green Create Service step:

stageTargetGroupArn: <+execution.steps.SelectTargetGroup.output.outputVariables.STAGE_TARGET_GROUP_ARN>

Can Harness provide a plugin or automatic inference for stage target group selection?

Stage target group selection is a manual input to support various traffic split configurations. If you have a specific use case that requires automatic inference, you can raise an enhancement request with Harness Support to explore options like:

• Publishing a plugin for automatic target group selection

• Adding configuration options for automatic inference based on traffic weights

• Supporting specific traffic patterns (e.g., always select 0% weight target group)

What are the best practices for stage target group selection?

1. Use runtime input: Configure stageTargetGroupArn: <+input> to manually verify before each deployment

2. Document your target groups: Keep a record of your two target group ARNs and which one was used in the last deployment

3. Implement automation: Use a shell script step to select the target group with lower/zero traffic automatically

4. Verify before deployment: Check the AWS Console to confirm which target group currently has 100% traffic (this is your blue service)

5. Select the opposite: Choose the other target group as the stage target group for the new deployment

Does Harness support Network Load Balancers (NLB) with multiple listeners for Blue/Green deployments?

Short answer: No, multi-listener Blue/Green deployments with NLB are not currently supported.

Detailed explanation:

Harness supports Blue/Green deployments using Network Load Balancers (NLB), but only with a single pair of stage/production listeners. If your ECS service uses an NLB with multiple listeners (multiple ports exposed), the Blue/Green deployment strategy with dynamic target group switching will not work.

Current Harness limitation:

• Only one stage listener and one production listener pair is supported

• Multi-listener configurations are not supported, even though ECS natively supports them

What this means:

• If your NLB exposes multiple ports (e.g., port 80, port 443, port 8080), Harness cannot perform automatic traffic shifting across all these listeners

• Only a single listener/target group pair will be managed by Harness during Blue/Green deployments

Alternative approaches:

• Use Application Load Balancer (ALB) instead, which fully supports Blue/Green deployments with Harness

• Use Rolling deployment strategy instead of Blue/Green if NLB with multiple listeners is required

• Configure separate load balancers for each port/service if Blue/Green is required

Related documentation

• ECS Blue-Green Traffic Shifting Step

• ECS Deployments Overview