A/B testing with split_clients in Nginx - Step-by-Step Execution

Process Flow - A/B testing with split_clients

Client Request Arrives

↓

split_clients Directive

↓

Hash Client ID or IP

↓

Assign Client to Variant A or B

↓

Serve Variant Content

↓

Log or Track Variant Served

The client request is processed by split_clients which hashes client info to assign them to variant groups for A/B testing.

Execution Sample

Nginx

split_clients "$remote_addr" $variant {
  50%     "A";
  *       "B";
}

if ($variant = "A") {
  return 200 "Variant A served";
}
return 200 "Variant B served";

This config splits clients by IP address into two groups: 50% get variant A, others get variant B, and returns a message accordingly.

Process Table

Step	Client IP	Hash Result	Assigned Variant	Action	Output
1	192.168.1.10	hash=0.45	A	Serve Variant A	Variant A served
2	192.168.1.11	hash=0.75	B	Serve Variant B	Variant B served
3	10.0.0.5	hash=0.10	A	Serve Variant A	Variant A served
4	10.0.0.6	hash=0.90	B	Serve Variant B	Variant B served
5	172.16.0.1	hash=0.50	A	Serve Variant A	Variant A served
6	172.16.0.2	hash=0.99	B	Serve Variant B	Variant B served
7	End of requests	-	-	-	No more clients

💡 All client requests processed and assigned variants based on hash thresholds.

Status Tracker

Variable	Start	After 1	After 2	After 3	After 4	After 5	After 6	Final
$remote_addr	-	192.168.1.10	192.168.1.11	10.0.0.5	10.0.0.6	172.16.0.1	172.16.0.2	-
$variant	-	A	B	A	B	A	B	-

Key Moments - 3 Insights

Why does a client with hash exactly 0.50 get assigned to variant A?

Because the split_clients directive assigns 50% to variant A using '50%', which includes hash values up to but not including 0.50. However, in nginx split_clients, the percentage is inclusive of the boundary, so hash=0.50 maps to A as shown in execution_table row 5.

What happens if a client's hash is greater than 0.50?

They are assigned to variant B as the '*' catch-all rule covers all remaining hash values above 0.50, as seen in rows 2,4,6.

Is the variant assignment consistent for the same client IP?

Yes, because the hash is based on the client IP ($remote_addr), so the same IP always hashes to the same variant, ensuring consistent experience.

Visual Quiz - 3 Questions

Look at the execution_table, what variant is assigned to client IP 10.0.0.5?

A Variant B

B No variant assigned

C Variant A

D Depends on time

💡 Hint

Check row 3 in execution_table where client IP 10.0.0.5 is assigned variant A.

At which step does the hash value exceed 0.50, causing assignment to variant B?

A Step 1

B Step 2

C Step 3

D Step 5

💡 Hint

Look at execution_table row 2 where hash=0.75 > 0.50, assigned to B.

If the split_clients directive changed to 30% for variant A, how would the assignment change for hash=0.45?

A Assigned to variant B

B Assigned to variant A

C No assignment

D Random assignment

💡 Hint

Since 0.45 > 0.30, it falls outside the 30% threshold, so assigned to variant B.

Concept Snapshot

split_clients "$variable" $target {
  percentage% "variant1";
  * "variant2";
}

- Splits clients by hashing $variable (e.g., IP)
- Assigns variants based on percentage thresholds
- Ensures consistent variant per client
- Used for A/B testing in nginx

Full Transcript

This visual execution shows how nginx's split_clients directive assigns clients to A/B test variants. Each client IP is hashed to a number between 0 and 1. If the hash falls within the first percentage (e.g., 50%), the client is assigned variant A; otherwise, variant B. The execution table traces multiple client IPs, their hash values, assigned variants, and the response served. Variables $remote_addr and $variant track client IP and assigned variant respectively. Key moments clarify how boundary hash values are assigned and consistency of assignment. The quiz tests understanding of variant assignment based on hash thresholds. The snapshot summarizes the syntax and behavior of split_clients for quick reference.